# Maisa — Full Site Content > Maisa builds hallucination-resistant AI Digital Workers that run regulated enterprise work end to end, prove every step, and leave you owning the process. This file contains the full text content of the public pages of https://maisa.ai, generated from the live site. For a concise index of the same pages, see https://maisa.ai/llms.txt. # Main pages ## Home URL: https://maisa.ai/ Deploy AI with confidence in regulated industries. Maisa's Digital Workers automate your most demanding processes, reliably, traceably, and auditably. ### Where frontier enterprises run their operations Hallucination-resistant AI that runs regulated enterprise work, proves every step, and leaves you owning the process. ### For the work that has to be right - Subpoena Request Intake Automation · Auto-Loan Processing Automation · New Client Onboarding Automation · Equity Research Automation - Banking & Finance - Progress Reporting · Certification Generation · Tender Preparation · Subcontractor Compliance - O2C & P2P Automation · FNOL Claims Transformation - Maintenance Work Orders · Supply Chain Task Resolution · Supplier Invoice Matching · Non-Conformance Handling ### How Maisa gets the work done Define the process once. A Digital Worker then executes it end to end across your existing systems, applying your rules, taking actions, handling exceptions and keeping a complete record of what happened. Onboard a Digital Worker Define the objective, process knowledge, business rules, controls, exceptions and expected outcome in Maisa Studio. Connect the applications, data and systems the process already depends on, including legacy systems. The Knowledge Processing Unit Every step leaves a trace you own Your work trains your models ### When Digital Workers run your operations, benefits compound Own your intelligence Every run leaves a trace of how your company works: information used, rules applied, outcomes reached. In time that record becomes intelligence nobody else has. 100% proprietary process intelligence Change the economics of AI AI gets more expensive the more you use it. Maisa goes the other way: specialised models on infrastructure you control, at a cost per execution that keeps falling. ~90x cheaper per run at scale Reproducible, down to every step Nothing runs unexplained. Every step is on record, debugging is quick, and a failed validation is repaired inside the run before it reaches the next step. 50M+ production runs, each one on record ### Our clients speak for us Scaling AI transformation across Santander Read full story → Deploying automations in 6 weeks for Indigo View the video ### Real Work Ready for ### Santander ### Indigo ### frontier enterprise +21 countries running Maisa for Santander Santander trusts Maisa as its global process automation platform, with thousands of Digital Workers across 15+ business units. 3 months from start to production for Indigo Indigo put four processes into production across finance, procurement, invoicing and tax. 93% lower cost per job After a customer's process worker moved from a frontier model to a company-owned SLM built with Maisa. 100,000+ executions a month, same accuracy. ### Get value from day one. Go live in less than six weeks. Deployment works because our team builds alongside yours guided by the 1-3-6 method that takes one use case from scoped to production in six weeks. Build We build the first version of your workflow in one week, using your real process and systems. Pilot We run it against live cases alongside your team, refining until the results hold up. Go-Live The Digital Worker moves into production, backed by real business data you can measure. ### Recognized by Gartner across reports that shape Enterprise AI Request the reports ## Product URL: https://maisa.ai/product The AI platform where business teams build their own Digital Workers with Maisa Studio. Scalable, governed, and ready for the enterprise. ### Deploy Digital Workers with Maisa Studio The AI platform where business teams build their own digital workers. Scalable, governed, and ready for the enterprise. ### Meet Maisa Studio Create Digital Workers in natural language, connect them to any system, and monitor every run from one place. ### 01 Onboard a Digital Worker ### Use natural language to explain the task, the data to use, and the goal to reach. ### 02 Connect your systems and tools ### 03 Run it before it goes live ### 04 Monitor every run in production ### Engineered for Reliability Powered by the KPU, Digital Workers turn complex processes into reliable automation. Hallucination-Resistant Code-backed execution reduces errors and enables self-healing to reach the goal. Complete Traceability and Audit Ready Code traces show every step, making fixes fast and compliance simple. ### Making AI economics work for enterprise Every run produces traces that let you switch to smaller, cheaper models over time — so cost falls as volume grows, without giving up accuracy. Freedom from frontier labs The process lives in your traces, not in the model that ran it. You can move a worker from one provider to another, or to a model you own, and it keeps doing the same work. Custom trained SLM that drops the cost Cost control at Enterprise scale Up to ~90x cheaper per run at scale ### Built for the most demanding processes New Client Onboarding Automation A global financial institution modernized its client onboarding process using Maisa Digital Workers, automatically classifying, analyzing, and processing onboarding documentation to dramatically reduce activation time while improving accuracy and scalability across regions. Auto-Loan Processing Automation A top-20 global bank deployed Digital Workers to handle document review, fraud detection, and deal structuring across auto loan origination, cutting time-to-fund from 4-5 days to 2 hours while processing over 23,000 loans monthly. Equity Research Automation A leading investment management group automated the extraction and structuring of insights from broker reports, turning unstructured content into actionable intelligence in minutes and enabling analysts to focus on high-impact investment strategy. Trade Finance Process Automation A leading global bank's Corporate and Investment Banking division transformed one of its hardest processes to automate. Certification Generation for Electricity Network Maintenance A global energy and utilities company transformed its electricity network maintenance by automating certification generation, removing manual bottlenecks and speeding up billing cycles. Supply Chain Task Resolution Automation A global manufacturing leader transformed its exception handling workflow by deploying Digital Workers to automate supply chain task resolution, achieving dramatic gains in speed, accuracy, and capacity. RFP Response Generation Automation A leading financial institution transformed RFP management with intelligent workflows that search internal documentation, generate compliant answers, and deliver ready-to-submit drafts in hours instead of days. Subpoena Request Intake Automation A leading financial institution restructured subpoena and legal request management with automated intake and processing, replacing a manual, cross-department process prone to delays and compliance risk. ### Built for work that must stand up to scrutiny. Governance Control who can build, run, and change Digital Workers — and prove every change was approved. Control Access by role, workspace or employee Deployment choice by architecture, region or security Version control with approval pipeline Risk Management Every execution is encrypted, explainable, and reproducible — so a decision can always be traced back and re-run. Encryption of all AI data at motion and at rest Encrypt PII on the fly for GDPR Reliability testing with deterministic re-execution Compliance Built to the standards regulated industries answer to, with an audit trail examiners can inspect. Configurable rules for FINMA, Basel III/IV, DORA, and the EU AI Act Immutable audit database with e-discovery Audit trails with proactive flagging and alerting Cybersecurity Your models and data stay inside your perimeter, with defense-in-depth against attacks aimed at AI. Runs inside your firewall, VPC, or on-premise Prompt-injection guardrails across text, files, and system data 24/7 monitoring with full stop-and-escalate control ## Use Cases URL: https://maisa.ai/use-cases See how Maisa AI's digital workers automate compliance-heavy workflows in banking, insurance, supply chain, and beyond. Real enterprise AI use cases. Explore how Maisa’s digital workers automate complex, high-stakes processes across industries. New Client Onboarding Automation A global financial institution modernized its client onboarding process using Maisa Digital Workers, automatically classifying, analyzing, and processing onboarding documentation to dramatically reduce activation time while improving accuracy and scalability across regions. Auto-Loan Processing Automation A top-20 global bank deployed Digital Workers to handle document review, fraud detection, and deal structuring across auto loan origination, cutting time-to-fund from 4-5 days to 2 hours while processing over 23,000 loans monthly. Equity Research Automation A leading investment management group automated the extraction and structuring of insights from broker reports, turning unstructured content into actionable intelligence in minutes and enabling analysts to focus on high-impact investment strategy. Trade Finance Process Automation A leading global bank's Corporate and Investment Banking division transformed one of its hardest processes to automate. RFP Response Generation Automation A leading financial institution transformed RFP management with intelligent workflows that search internal documentation, generate compliant answers, and deliver ready-to-submit drafts in hours instead of days. Subpoena Request Intake Automation A leading financial institution restructured subpoena and legal request management with automated intake and processing, replacing a manual, cross-department process prone to delays and compliance risk. Fund Performance Reporting Automation A global investment bank redesigned its fund performance reporting with Maisa Digital Workers. Manual data extraction, cleaning, and consolidation across platforms is now an automated workflow that delivers compliant, audit-ready reports in minutes, so teams focus on analysis instead of data prep. Adverse Media Screening Alerts Review Automation A global corporate and investment bank uses Maisa Digital Workers to automate adverse media alert reviews in compliance. A once slow, inconsistent manual process is now automatically structured, documented, and policy-aligned, delivering faster, consistent decisions with full regulatory traceability. Correspondent Account Transaction Reconciliation Automation A major financial services provider transformed reconciliation with Maisa Digital Workers. Instead of teams reviewing thousands of unmatched transactions, mostly false positives, adaptive automated logic now delivers faster exception handling, fewer manual steps, and higher accuracy. Deceased Account Processing Automation A leading retail bank transformed how it manages deceased client accounts with Maisa Digital Workers. A once manual process that varied by region, verifying death certificates, closing accounts, and contacting families, is now automated end to end, handling sensitive procedures quickly, consistently, and with full regulatory assurance. Tax Withholding Reconciliation Automation A European investment firm automated its annual tax withholding reconciliation with Maisa Digital Workers. What once meant manually comparing fund data and counterparty certificates across formats, slow and reliant on specialists, is now fully automated, with faster cycles, complete auditability, and minimal manual intervention. Power of Attorney Automation After struggling with low automation accuracy and heavy manual review in loan applications, this leading European bank used Maisa to revolutionize intelligent document processing and boost customer satisfaction. Certification Generation for Electricity Network Maintenance A global energy and utilities company transformed its electricity network maintenance by automating certification generation, removing manual bottlenecks and speeding up billing cycles. “We are proud to strengthen our partnership with Maisa, whose unique vision to scale the value of AI in real, complex and regulated environments truly inspires us.” Ignacio Bernal Chief AI Officer at Grupo Santander O2C & P2P Automation By replacing rigid RPA and unreliable AI tools with hallucination-resistant Digital Workers, this enterprise transformed billing, payments, and financial operations with scalable, trustworthy automation. FNOL Claims Transformation A leading European insurer deployed a specialised FNOL Digital Worker that takes a claim from first contact to a fully structured, validated case file without human intervention — collapsing a four-to-five-hour triage cycle to minutes and processing 98% of claims straight through. Supply Chain Task Resolution Automation A global manufacturing leader transformed its exception handling workflow by deploying Digital Workers to automate supply chain task resolution, achieving dramatic gains in speed, accuracy, and capacity. ## Blog URL: https://maisa.ai/blog Expert articles on agentic AI, digital workers, cost control, and enterprise automation trends. ### News & Insights All News Insights Digital Workers AI in the Enterprise Research and Tech Maisa x Santander: 6 Lessons from Scaling AI Inside a Large Global Bank Exploring the organizational and operational challenges of deploying AI systems responsibly within regulated financial institutions. May 26, 2026 Maisa Named as a Front Runner in Gartner’s “Emerging Tech: AI Vendor Race – Startups to Watch in Agentic AI” Report Mar 30, 2026 Maisa Wins the Rising Star Award at Deloitte’s Technology Fast 50 Spain Feb 13, 2026 How Maisa Digital Workers Actually Work? Jan 12, 2026 Why Maisa Is Not an n8n-like tool or RPA+ AI Jan 7, 2026 Agentic AI Cost Control: The Enterprise Leader’s Guide to Avoiding “Surprise Opex” at Scale What is business process automation (BPA)? Nov 11, 2025 Agentic AI vs RPA: differences, similarities, and examples Oct 24, 2025 Rising star of agentic AI delivering trustworthy ‘digital workers’ raises $25M from Creandum and Forgepoint Aug 28, 2025 Maisa raises $25M seed investment Maisa Studio on AWS Marketplace: removing barriers to AI adoption Jul 23, 2025 CLATTER: Academic Validation for Our Maisa AI Hallucination Detection Strategy Jun 20, 2025 HALP: Maisa’s breakthrough in delivering reliability for enterprise automation Jun 4, 2025 Why we built Maisa this way: scientific proof we’re on the right track Apr 24, 2025 Advancing our vision for Accountable AI together with Microsoft Apr 7, 2025 Black Box AI. How can we trust what we can’t see? Apr 1, 2025 Making AI accountable: Maisa raises pre-seed round Dec 25, 2024 Introducing Vinci Knowledge Processing Unit (KPU) Nov 26, 2024 Introducing Maisa, a new AI architecture Mar 15, 2024 Maisa Updates Maisa recognized as a front runner among agentic AI startups by Gartner's latest vendor evaluation report. ## Careers URL: https://maisa.ai/careers Explore open roles at Maisa and help design accountable, enterprise-grade AI Digital Workers. Grow your career in a fast-moving, global team. ### Come and define the future of enterprise AI Maisa builds Digital Workers for the world’s most demanding industries. Behind our technology is a team of humans who care intensely about their craft. This is your invitation to become one of them. See Open Roles A note from our founders Benefits ### The people behind the AI revolution. Digital Workers may run the processes, but humans build everything here. Meet Maisa's cofounders David Villalon (CEO) and Manu Romero (CSO). ### Great work needs great conditions. Equity upside Everyone owns a piece of Maisa. When we win, you win with us. Health & wellness Mind and body support through Alan’s wellness programs and gym membership. Remote friendly Work from where you do your best thinking. We hire for output, not office hours. Team offsites in Spain The whole company together, two times a year, where Maisa was born. Gear you would choose Top equipment and the best AI tools on the market. Nothing slows you down. Transportation covered Get to work on us. We cover your monthly transit pass. Growth & Learning Courses, books, workshops, conferences and Preply language classes. If it makes you better at your craft, we back it. ### Build for global impact ### Five core values. Accountability Outcomes are personal here. When something goes wrong, we answer for it, fix it, and say what we learned. Nobody hides behind process. Ambition We aim higher than what is asked and raise our hands for the hardest problems. Safe work is the only work we refuse. Trust We behave consistently enough that nobody needs to check. Trust is the core of our offering, our product, and what we expect from each other. Collaboration We make the people around us better and put the shared result above individual credit. Wins belong to the team. Grit We move through difficulty without handing off problems we should own. Hard is not a reason to stop, it is usually the reason to start. Open Roles ### Build the future of AI ## Contact URL: https://maisa.ai/contact Get in touch with Maisa AI to explore Digital Workers, book a demo, discuss partnerships, or get support from our team. ### Learn why Maisa is used by leading enterprises Get a custom demo. Discover the value of Maisa for your enterprise and explore our custom plans and pricing. Tailored one-on-one platform walkthrough Expert guidance & consulting Comprehensive Q&A No commitment to sign up Ignacio Bernal, Chief AI Officer "We are proud to strengthen our partnership with Maisa, whose unique vision to scale the value of AI in real, complex and regulated environments truly inspires us." Leading companies already trust Maisa First Name Last name Corporate email Phone Company size I am reaching out to How did you hear about us? In short: By submitting this form, you agree to our Privacy Policy Your information is kept confidential. # Industries ## Banking & Financial Services URL: https://maisa.ai/industries/banking-financial-services Reliable AI for banking and finance: Maisa's Digital Workers give banks and financial institutions a competitive edge in their markets. Reliable AI for Trade Finance Auto-Loan Processing Equity Research New Client Onboarding Application Received Documents Extracted Identity Verified Credit Report Retrieved Affordability Assessed Policy Checks Run Underwriter Review Decision Approved ### Reliable AI for Banking and Finance Maisa's Digital Workers give banks and financial institutions a competitive edge in their markets. ### Why Maisa works for financial entities Works with legacy systems Connect to core systems, document repositories, databases and legacy applications. Controlled deployment options Apply bank rules and SOPs ### Digital workers in action for banking and financial services New Client Onboarding Automation A global financial institution modernized its client onboarding process using Maisa Digital Workers, automatically classifying, analyzing, and processing onboarding documentation to dramatically reduce activation time while improving accuracy and scalability across regions. Auto-Loan Processing Automation A top-20 global bank deployed Digital Workers to handle document review, fraud detection, and deal structuring across auto loan origination, cutting time-to-fund from 4-5 days to 2 hours while processing over 23,000 loans monthly. Equity Research Automation A leading investment management group automated the extraction and structuring of insights from broker reports, turning unstructured content into actionable intelligence in minutes and enabling analysts to focus on high-impact investment strategy. Trade Finance Process Automation A leading global bank's Corporate and Investment Banking division transformed one of its hardest processes to automate. RFP Response Generation Automation A leading financial institution transformed RFP management with intelligent workflows that search internal documentation, generate compliant answers, and deliver ready-to-submit drafts in hours instead of days. Subpoena Request Intake Automation A leading financial institution restructured subpoena and legal request management with automated intake and processing, replacing a manual, cross-department process prone to delays and compliance risk. Fund Performance Reporting Automation A global investment bank redesigned its fund performance reporting with Maisa Digital Workers. Manual data extraction, cleaning, and consolidation across platforms is now an automated workflow that delivers compliant, audit-ready reports in minutes, so teams focus on analysis instead of data prep. Adverse Media Screening Alerts Review Automation A global corporate and investment bank uses Maisa Digital Workers to automate adverse media alert reviews in compliance. A once slow, inconsistent manual process is now automatically structured, documented, and policy-aligned, delivering faster, consistent decisions with full regulatory traceability. Correspondent Account Transaction Reconciliation Automation A major financial services provider transformed reconciliation with Maisa Digital Workers. Instead of teams reviewing thousands of unmatched transactions, mostly false positives, adaptive automated logic now delivers faster exception handling, fewer manual steps, and higher accuracy. Deceased Account Processing Automation A leading retail bank transformed how it manages deceased client accounts with Maisa Digital Workers. A once manual process that varied by region, verifying death certificates, closing accounts, and contacting families, is now automated end to end, handling sensitive procedures quickly, consistently, and with full regulatory assurance. Tax Withholding Reconciliation Automation A European investment firm automated its annual tax withholding reconciliation with Maisa Digital Workers. What once meant manually comparing fund data and counterparty certificates across formats, slow and reliant on specialists, is now fully automated, with faster cycles, complete auditability, and minimal manual intervention. Power of Attorney Automation After struggling with low automation accuracy and heavy manual review in loan applications, this leading European bank used Maisa to revolutionize intelligent document processing and boost customer satisfaction. ### Proven impact for every department Maisa’s Digital Workers give banks and financial institutions a competitive edge in their markets. Consumer and retail banking Automate onboarding, KYC and claims workflows while preserving a reliable audit trail across every customer channel. Use Cases Institutional and wealth Coordinate research, trading, compliance and relationship workflows with governed digital workers embedded in the team. Enterprise operations Reduce manual work across finance, HR, procurement, compliance and IT operations with consistent, measurable execution. “We are proud to strengthen our partnership with Maisa, whose unique vision to scale the value of AI in real, complex and regulated environments truly inspires us.” Ignacio Bernal Chief AI Officer at Grupo Santander ## Insurance URL: https://maisa.ai/industries/insurance Digital Workers help insurers move faster and stay compliant across the policy lifecycle. Reliable AI for First Notice of Loss Quote and Bind Claims Adjudication Policy Issuance Claim Submitted Documents Extracted Policy Coverage Verified Damage Assessed Fraud Checks Run Payout Calculated Adjuster Review Claim Approved ### Reliable AI for Insurance ### Why insurance companies choose Maisa Hallucination-resistant by design Maisa’s Knowledge Processing Unit sits on top of any large language model and turns predictions into deterministic, rule-based execution. Every output is consistent and safe for production use in a regulated environment. Fully auditable with complete chain of work Every decision is logged: data used, checks run, sources queried, reasoning applied. Audit trails are auto-generated and regulator-ready, with compliance built in for GDPR, the EU AI Act, US Banking, SEC, and HIPAA. Predictable pricing Maisa is priced like headcount, not like API calls. No surprise token bills, no usage spikes, no costs that scale with prompt length. The ROI on every Digital Worker is calculable before contracts are signed, with the business case built and validated before deployment. ### Digital workers in action for insurance O2C & P2P Automation By replacing rigid RPA and unreliable AI tools with hallucination-resistant Digital Workers, this enterprise transformed billing, payments, and financial operations with scalable, trustworthy automation. FNOL Claims Transformation A leading European insurer deployed a specialised FNOL Digital Worker that takes a claim from first contact to a fully structured, validated case file without human intervention — collapsing a four-to-five-hour triage cycle to minutes and processing 98% of claims straight through. ### Proven impact across the insurance lifecycle Capture claims from every channel, classify documents, verify policy and coverage, and open the case in the CMS. Eliminates manual triage, freeing handlers for judgment work. Use Cases Claim Intake Automation Document Classification Automation Underwriting: Quote and Bind Pull policy data, run third-party reports, evaluate risk, and produce a structured decision with the reasoning attached. Eliminates back and forth across systems, freeing underwriters for the risks that need a conversation. Quote and Bind Automation Risk Evaluation Automation Policy Issuance and Endorsements Adjudication and Resolution Consolidate damage, liability, and settlement calculations into a complete case file validated against policy terms, then verify disbursement readiness and track recovery inside the core platform. Eliminates routing loops and manual handoffs at close. Case Setup and Adjudication Billing and Booking Automation “We are proud to strengthen our partnership with Maisa, whose unique vision to scale the value of AI in real, complex and regulated environments truly inspires us.” Ignacio Bernal Chief AI Officer at Grupo Santander ## Engineering & Infrastructure URL: https://maisa.ai/industries/engineering-infrastructure Helping engineering and infrastructure teams streamline operations, improve safety and compliance, and deliver value from design to maintenance. Reliable AI for Certifications Project Planning Field Operations Asset Management Per SQM Budget Calculation Materials List Amended Plans Approved Purchase Order Order Sent to Supplier Delivery Received on Site Site Manager Review Review Approved ### Reliable AI for Engineering and Infrastructure ### Why Leading Manufacturers choose Maisa Trust & Auditability Transparent and verifiable actions ensure complete traceability across production lines, supplier interactions, equipment data, and plant records, supporting strict regulatory and quality management requirements. Scalability by Business Teams Operations, quality, engineering, and supply chain teams can deploy Digital Workers using natural language, allowing plants to scale automation across sites and processes without dependence on IT or specialized automation teams. Faster Operations Digital Workers remove delays across sourcing, supply chain, production, quality, maintenance, and service workflows. Manufacturers move from planning to execution to delivery in a much shorter time, improving throughput and operational stability. ### Digital workers in action for engineering and infrastructure Certification Generation for Electricity Network Maintenance A global energy and utilities company transformed its electricity network maintenance by automating certification generation, removing manual bottlenecks and speeding up billing cycles. ### Proven impact across every function Project Planning & Design Digital Workers speed up planning by automating tasks such as document analysis, blueprint verification, model updates, compliance checks, and stakeholder reporting, reducing manual review cycles and ensuring faster alignment across engineering teams and contractors. Use Cases Certification Generation Inspection Workflow Automation Compliance Check Automation Construction & Field Operations Maisa supports site execution by automating inspection workflows, progress tracking, materials verification, safety monitoring, and subcontractor documentation, giving teams real-time visibility across the field. Progress Tracking Automation Materials Verification Automation Subcontractor Documentation Automation Asset Management & Back-Office Ops Organizations enhance efficiency by automating asset documentation, maintenance reviews, procurement workflows, and administrative operations, streamlining finance, HR, compliance, and IT functions so teams can focus on strategic programs. Asset Documentation Automation Maintenance Review Automation Procurement Workflow Automation “We are proud to strengthen our partnership with Maisa, whose unique vision to scale the value of AI in real, complex and regulated environments truly inspires us.” Ignacio Bernal Chief AI Officer at Grupo Santander ## Manufacturing URL: https://maisa.ai/industries/manufacturing Eliminate bottlenecks, improve accuracy, maintain compliance, and move faster from planning to production to delivery with Digital Workers. Reliable AI for Supply Chain Production Maintenance Shipment Received Certificate of Analysis Extracted Batch Details Verified Measurements Checked vs Spec Quantities Reconciled Non-Conformance Screened Quality Manager Review Batch Approved ### Reliable AI for Manufacturing ### Why Leading Manufacturers choose Maisa Trust & Auditability Transparent and verifiable actions ensure complete traceability across production lines, supplier interactions, equipment data, and plant records, supporting strict regulatory and quality management requirements. Scalability by Business Teams Operations, quality, engineering, and supply chain teams can deploy Digital Workers using natural language, allowing plants to scale automation across sites and processes without dependence on IT or specialized automation teams. Faster Operations Digital Workers remove delays across sourcing, supply chain, production, quality, maintenance, and service workflows. Manufacturers move from planning to execution to delivery in a much shorter time, improving throughput and operational stability. ### Digital workers in action for manufacturing Supply Chain Task Resolution Automation A global manufacturing leader transformed its exception handling workflow by deploying Digital Workers to automate supply chain task resolution, achieving dramatic gains in speed, accuracy, and capacity. ### Proven impact across the plant Sourcing and Procurement Automate supplier documents, purchase orders, contract checks, and delivery confirmations to improve accuracy and strengthen supplier performance. Use Cases Supplier Document Automation Purchase Order Automation Delivery Confirmation Automation Supply Chain Management Streamline demand inputs, shipment verification, exception handling, and inventory checks for faster decisions and fewer disruptions. Shipment Verification Automation Inventory Check Automation Production and Maintenance Process production logs, equipment data, inspections, and maintenance records to detect issues earlier and improve plant reliability. Production Log Processing Inspection Record Automation Maintenance Record Automation “We are proud to strengthen our partnership with Maisa, whose unique vision to scale the value of AI in real, complex and regulated environments truly inspires us.” Ignacio Bernal Chief AI Officer at Grupo Santander # Use Cases ## Auto-Loan Processing Automation URL: https://maisa.ai/use-cases/auto-loan-processing A top-20 global bank deployed Digital Workers to handle document review, fraud detection, and deal structuring across auto loan origination, cutting time-to-fund from 4-5 days to 2 hours while processing over 23,000 loans monthly. ### $3.0M Direct processing cost savings per year ### +4% Net volume from faster funding SLA ### 2 hr ### 90,604 Documents processsed monthly ### Challenge Time-to-fund is the primary competitive differentiator in auto lending. Dealers simultaneously submit contract packages to multiple lenders, and the fastest funder wins the deal. Manual processes create significant bottlenecks: 4–5 day time-to-fund Manual review cycles tie up dealer inventory and freeze working capital, giving faster competitors an opportunity to capture approved borrowers. 10–15% deal loss Auto loan applicants submit to multiple lenders simultaneously. Slower funders watch approved deals transfer to faster competitors. Multiple review loops Manual processes often stop at the first error and send packages back for corrections, creating additional delays with each round-trip. 30–60 minutes per package Manual checklist review, document extraction, and cross-validation consume significant labor hours with inconsistent results across teams. Customer service strain Dealers flood call centers with funding status inquiries, increasing staffing costs with each delay. $9.2B industry fraud exposure Synthetic ID fraud up 98%. Income misrepresentation the #1 category. Manual review processes miss patterns undetectable at volume. ### Solution The client deployed two specialized Digital Workers across the origination flow, each handling a specific function. Clients now receive responses before applicants leave the dealership. Uploads RFP documents, reads sections, and automatically identifies all questions and response requirements. Creates first drafts by analyzing historical data, previous submissions, and approved documentation with automatic formatting for Word, PowerPoint, or Excel. Establishes a single global document library providing verified, current content to all teams. Assigns questions to teams, tracks real-time progress, maintains version control and transparency. Searches internal knowledge repositories to locate relevant compliant information. Provides progress insights, alerts, and change control management. ### Digital Workers Deployed Document Review & Extraction Identity Verification & Fraud Detection Ingests full contract packages including contracts, payslips, bank statements, W-2s, and identity documents Extracts all fields and cross-validates data across documents in a single pass Flags all issues simultaneously—eliminating partial reviews or back-and-forth cycles Routes clean files directly to funding; only exceptions require human review Performs ID verification, fake payroll detection, and cross-document identity checks Integrates with credit bureaus and third-party data sources Applies dynamic dealer risk scoring based on delinquency and fraud patterns Outputs explainable decisions: Clean, Soft Flag, or Hard Flag with full reasoning ### Results $3.0M direct processing cost savings annually 110,000+ manual hours saved per year as document review, extraction, and cross-validation shifted from people to Digital Workers. +4% net volume Collapsing the 4–5 day manual cycle to a 2-hour SLA recovered approximately 920 loans monthly previously lost to faster competitors. $1–1.5M dealer call center savings annually Faster time-to-fund eliminated 25–35% of “where’s my funding?” calls from dealers chasing status updates. Fraud prevention Cross-document checks and dealer risk scoring catch synthetic IDs, fake payroll, and income fraud before disbursement. 100% auditability Every fraud and risk decision includes a full chain of work: checks run, signals fired, sources queried, and the reasoning behind the final decision. >$10M projected annual ROI Predictable pricing backed by a clear business case, validated before deployment to ensure measurable returns from day one. ## FNOL Claims Transformation URL: https://maisa.ai/use-cases/fnol-transformation A leading European insurer deployed a FNOL Digital Worker that turns a claim into a validated case file without human intervention, cutting triage from hours to minutes and processing 98% of claims straight through. A leading European insurer deployed a specialised FNOL Digital Worker that takes a claim from first contact to a fully structured, validated case file without human intervention — collapsing a four-to-five-hour triage cycle to minutes and processing 98% of claims straight through. ### 53% Faster claims processing ### 98% Straight-through processing ### 58,000+ Hours returned annually ### Challenge Claims management is one of the most manual, costly, and time-intensive operations in insurance. Every stage runs on people moving information across systems, decisions wait until the right person is available, and the longer a claim takes, the more it costs and the more customer trust erodes. Hours-long manual triage Handlers move each claim across systems one by one. At peak times, triage alone can take hours before a file is ready for an adjuster. Back-and-forth on every file Missing documents, mismatched data, and duplicate cases send the file back to the customer. New issues surface and the cycle repeats. Backlogs and rising costs Funders stop at the first error and send packages back. Dealers fix, resubmit, hit new errors. Days lost per round-trip. RPA is not built for this FNOL is full of variation — claim types, document formats, exceptions. RPA breaks the moment the process deviates from a fixed script. Black-box AI has no audit trail Standard AI systems work as a black box. Inputs go in and an output comes out, but the reasoning in between is invisible — unworkable in a regulated industry. Customer experience suffers FNOL sets the tone for the whole claims experience. A slow, fragmented start damages trust before the case is even assessed. ### Solution Our client deployed one specialised Digital Worker across their FNOL flow, with a single job: take a claim from first contact to a fully structured, validated case file without a human touching it. The four-to-five-hour manual cycle collapsed to a few minutes. Receives the claim through any channel, with photographs, police reports, and supporting documents Reads, classifies, and fully understands all information provided Confirms no duplicate case exists for the same customer and policy Validates coverage against live policy data Creates the case directly in the legacy CMS, uploads a detailed PDF report, and attaches all documentation Hands the adjuster a complete, validated, fully traceable file ### Digital Workers Deployed Julian: FNOL Digital Worker ### Results Intake and triage that used to take hours now complete in minutes, around the clock, across every incoming channel. Manual hours move out of intake and into the higher-judgment work where adjusters add the most value. Nearly every claim moves from first contact to a fully validated case file with no human in the loop. Adjusters spend their day on the complex cases that need judgment, not on triage and data entry. Claims reported overnight, on weekends, or holidays are triaged in real time — no queue, no delay until business hours. Every case file includes a full chain of work — checks run, sources queried, signals fired, and the reasoning behind the final decision. ## Power of Attorney Automation URL: https://maisa.ai/use-cases/power-of-attorney-verification After struggling with low automation accuracy and heavy manual review in loan applications, this leading European bank used Maisa to revolutionize intelligent document processing and boost customer satisfaction. ### 40,000 hours Annual time recovery ### Minutes Verification processing time ### €1.2 million Cost savings in single region ### Challenge Power of Attorney verification is one of the most critical and labor-intensive processes in financial compliance. Every document must confirm that the individual signing on behalf of an entity has the legal authority to do so. Each case involves parsing legal terminology, confirming identities, and validating authorizations under multiple regulatory frameworks. Before Maisa, this global institution managed the process entirely by hand. More than three hundred specialists spent thousands of hours reviewing scanned PDFs of inconsistent quality, often without a text layer. They manually read handwriting, stamps, and signatures to validate against internal policies and regional laws. This manual approach slowed client onboarding, delayed transaction approvals, and created compliance bottlenecks. Attempts to automate the process using OCR and traditional machine learning systems failed because these tools could not understand the nuance of legal language or the diversity of document formats. The result was a process that drained time, created operational risk, and limited scalability. ### Solution Using Maisa Studio, the compliance team designed and deployed a Digital Worker capable of managing Power of Attorney verification end to end. The configuration process required no coding. The team simply described the task and rules in natural language, allowing the Digital Worker to learn and replicate their expertise. The Digital Worker reads both digital and scanned documents, even those without embedded text, ensuring complete coverage across formats and quality levels. It interprets legal clauses, identifies parties, and verifies authority in context. The system applies the institution’s internal compliance logic and keeps every decision traceable, enabling real-time auditing and transparency. With each verification, the Digital Worker strengthens its accuracy. It learns from compliance officer feedback in natural language and continuously adapts to new regulatory requirements without requiring new code or redeployment. Straightforward cases are approved automatically, while complex ones are routed directly to specialists. Using Human Augmented LLM Processing, humans and the Digital Worker collaborate seamlessly, and recurring issues are resolved autonomously through its self-improving capability. Through this collaboration, the bank achieved faster verification cycles, reduced workload for compliance professionals, and built a reliable automation framework that evolves continuously. ### Results Automating Power of Attorney verification freed forty thousand hours annually, releasing thousands of days of productive capacity back to compliance teams. Verifications that once required hours now take only minutes, accelerating customer activation and transaction processing. The Digital Worker evolves with each execution, updating its validation methods automatically to reflect the latest regulations and legal templates. Deploying the solution in a single region generated 1.2 million euros in savings, demonstrating one of the highest returns on automation in compliance operations. ## Tax Withholding Reconciliation Automation URL: https://maisa.ai/use-cases/tax-withholding-reconciliation A European investment firm automated its annual tax withholding reconciliation with Maisa Digital Workers, matching fund data and counterparty certificates with faster cycles and complete auditability. A European investment firm automated its annual tax withholding reconciliation with Maisa Digital Workers. What once meant manually comparing fund data and counterparty certificates across formats, slow and reliant on specialists, is now fully automated, with faster cycles, complete auditability, and minimal manual intervention. ### Weeks to days Processing time reduction ### 100% Automation rate ### Challenge Tax withholding reconciliation is a complex process that requires comparing internal fund records with multiple counterparty certificates, each formatted differently. The task is usually carried out once or twice per year and consumes extensive time from highly skilled analysts. At this firm, the reconciliation process was fully manual. Teams had to review hundreds of certificates, cross-checking dates, amounts, and tax categories across various systems. Inconsistent document structures, missing data, and human error frequently caused delays and discrepancies. Without a centralized view, oversight and auditability were limited. Each fund required custom treatment, and progress was difficult to track, which made it hard to guarantee compliance and reporting accuracy. ### Solution Using Maisa Studio, the firm deployed a Maisa Digital Worker to automate tax withholding reconciliation from data ingestion to final validation. Configured with natural language, the Digital Worker integrates with fund systems and counterparties to streamline every step of the process. Reads and parses incoming certificates from counterparties, regardless of format, and organizes them into a structured dataset for reconciliation. Applies configurable criteria such as date, amount, and fund ID to match entries across systems with high accuracy. Flags unmatched records, consolidates reconciled entries, and generates dynamic reports showing resolved and pending items. Produces structured outputs, including detailed audit logs and traceable Excel files that meet both internal and external reporting standards. Supports ongoing reconciliation as new certificates are received, reducing bottlenecks and allowing continuous progress throughout the reporting cycle. ### Results Reconciliation time dropped from weeks to days, enabling teams to meet deadlines with ease. All transactions are traceable, with clear records for matched and unmatched items across funds and counterparties. Automation drastically reduced manual effort, freeing specialized staff to focus on analysis and strategic reporting. New funds and counterparties can be added quickly, allowing the system to adapt to future growth without additional development. ## Deceased Account Processing Automation URL: https://maisa.ai/use-cases/deceased-account-processing A leading retail bank automated deceased account processing with Maisa Digital Workers: certificate verification, account closure and family contact now run end to end with full regulatory assurance. A leading retail bank transformed how it manages deceased client accounts with Maisa Digital Workers. A once manual process that varied by region, verifying death certificates, closing accounts, and contacting families, is now automated end to end, handling sensitive procedures quickly, consistently, and with full regulatory assurance. ### Challenge Processing deceased client accounts is one of the most delicate and administratively complex responsibilities in retail banking. Each case requires careful verification of documentation, closure of multiple linked accounts, and formal communication with family members. At this institution, the process was entirely manual. Teams had to review death certificates submitted in different formats by region, cross-check customer information across systems, and manually identify every associated account. Each step took time and carried the risk of missed records or inconsistent follow-up. Formal communications with families required precision and sensitivity. Maintaining the correct tone, following legal requirements, and documenting each message added further complexity. As a result, the end-to-end process was slow, error-prone, and difficult to audit, increasing both operational risk and emotional strain for staff. ### Solution Through Maisa Studio, the bank developed a Maisa Digital Worker to manage the entire workflow for deceased account processing. Configured in natural language, the Digital Worker follows the institution’s compliance standards, internal procedures, and communication protocols with accuracy and empathy. Extracts and validates information from both scanned and digital death certificates, recognizing variations in format and structure. Searches internal databases to identify all active and linked accounts associated with the deceased client, ensuring that none are overlooked. Carries out account closure and transfer steps in alignment with bank policies, coordinating actions across core systems and compliance checks. Generates formal letters and notifications using approved templates that maintain appropriate tone and language, ready for review or automatic dispatch. Records every action, from document verification to message delivery, creating a transparent audit trail for regulatory and internal reporting. ### Results Account processing time was significantly reduced as automation now handles document validation, account searches, and communications. Systematic identification ensures that all accounts are found and processed correctly, eliminating oversight and delays. Family notifications follow pre-approved templates and tone guidelines, ensuring clarity and professionalism across every region. All actions are logged and auditable, providing a complete and verifiable record of each case handled. ## Correspondent Account Transaction Reconciliation Automation URL: https://maisa.ai/use-cases/correspondent-account-transaction-reconciliation A major financial services provider automated transaction reconciliation with Maisa Digital Workers, clearing thousands of false-positive mismatches with faster exception handling and higher accuracy. A major financial services provider transformed reconciliation with Maisa Digital Workers. Instead of teams reviewing thousands of unmatched transactions, mostly false positives, adaptive automated logic now delivers faster exception handling, fewer manual steps, and higher accuracy. ### 90% Workload reduction ### 99% False positive identification rate Automation rate ### Challenge Transaction reconciliation is a critical process that ensures financial accuracy across multiple systems. In this institution, the existing reconciliation workflow generated an overwhelming number of false positives, forcing analysts to manually review transactions that were not actual discrepancies. True exceptions were often hidden among system mismatches or data inconsistencies, consuming valuable analyst time and attention. The reconciliation logic in place was complex and costly to update using traditional automation tools. Each exception required human judgment to interpret the transaction context, apply procedural logic, and resolve cases accurately. As volumes increased, the lack of flexibility and automation created delays, higher operational costs, and growing frustration among reconciliation teams. ### Solution Using Maisa Studio, the operations team created a Maisa Digital Worker that learns reconciliation logic directly from standard operating procedures, historical cases, and guided analyst input. – The Digital Worker observes and records the reasoning steps analysts follow when resolving exceptions, replicating their approach with precision and consistency. – It connects securely to internal platforms through application programming interfaces, enabling data access and matching within the organization’s controlled environment. – Maisa applies flexible reasoning to detect alternative attributes or fallback values when direct matches fail, ensuring that the system identifies genuine discrepancies more accurately. – When new exception patterns emerge, the Digital Worker updates its logic automatically, improving its accuracy with every cycle. – The solution was deployed using sample datasets and a few guided sessions, requiring no coding or major system rebuilds. ### Results Ninety nine percent of false positives are now automatically identified and excluded, allowing analysts to focus on true exceptions. Reconciliation workload has been reduced by ninety percent, delivering substantial time and cost savings. Complex reconciliation logic is handled dynamically, ensuring the process remains efficient and scalable without constant technical maintenance. Maisa Digital Workers use contextual data across connected systems to resolve mismatches accurately and consistently. ## Adverse Media Screening Alerts Review Automation URL: https://maisa.ai/use-cases/adverse-media-screening-alerts A global corporate and investment bank automated adverse media alert reviews with Maisa Digital Workers, turning a slow manual process into fast, consistent, policy-aligned decisions with full traceability. A global corporate and investment bank uses Maisa Digital Workers to automate adverse media alert reviews in compliance. A once slow, inconsistent manual process is now automatically structured, documented, and policy-aligned, delivering faster, consistent decisions with full regulatory traceability. ### Challenge Evaluating adverse media alerts is a demanding part of compliance and risk management. Each alert requires analysts to read multiple articles, extract relevant details, and determine potential reputational or financial risks based on internal and regulatory standards. At this bank, analysts performed these reviews manually. Each article had to be read, assessed, and summarized, with findings entered into evaluation matrices by hand. The process was fragmented, relying on several internal systems and third-party data platforms. The heavy dependence on human interpretation created inconsistencies and made it difficult to maintain a clear audit trail. Results varied by reviewer, and missing data or subjective assessments increased compliance risk. As alert volumes grew, so did backlogs and response delays, stretching resources and reducing operational agility. ### Solution Using Maisa Studio, the bank’s compliance team built a Maisa Digital Worker that automates the entire adverse media review process. Configured through natural language, the Digital Worker applies the institution’s internal frameworks for Financial Crime and ESG screening, ensuring each review follows consistent rules. The Digital Worker identifies relevant information within articles, extracting key facts and mapping them directly to predefined evaluation matrices. Each case is processed using uniform logic, ensuring that risk classification and rationale follow consistent patterns that align with internal policies. Maisa produces detailed outputs that include the extracted data, risk category, and rationale, creating structured, audit-ready documentation. It connects seamlessly with the bank’s compliance platforms and data sources, consolidating information flow and eliminating manual re-entry. The Digital Worker is easily updated to include new evaluation criteria, languages, or alert categories as regulations evolve or new data sources are added. ### Results Adverse media reviews that previously required hours of analyst effort are now completed in minutes. Each decision follows a unified logic, producing structured, high-quality assessments that align with regulatory and internal standards. All responses are documented and traceable, providing complete visibility for both internal reviews and external audits. ## O2C & P2P Automation URL: https://maisa.ai/use-cases/o2c-p2p By replacing rigid RPA and unreliable AI tools with hallucination-resistant Digital Workers, this enterprise transformed billing, payments, and financial operations with scalable, trustworthy automation. ### 58,000+ Hours saved annually ### 98% Straight-through processing ### Challenge This global company processes thousands of financial transactions daily—from billing customers to paying suppliers and closing the books. Manual, document-heavy processes in Order to Cash (O2C) and Procure to Pay (P2P) cycles created significant bottlenecks. Orders arrived in countless formats (emails, PDFs, scanned documents), forcing staff to manually validate line by line. Supplier invoices varied in quality, often missing embedded text, requiring rechecks of amounts, tax references, and payment terms. Each case took 10+ minutes of cross-referencing, with errors leading to delays and clarification loops. The organization tested RPA, UiPath, Microsoft Power Automate, and GPT-based AI frameworks—none scaled. Scripts broke on format changes, outputs were unreliable for production, and validation layers multiplied rather than reduced workload. The result: backlogs, delayed revenue recognition, late supplier payments, and operational ceilings that risked limiting growth. ### Solution Maisa introduced trustworthy, hallucination-resistant Digital Workers, giving the finance team confidence to scale automation with traceability, auditability, and resilience. Captures orders across channels, extracting customer info, quantities, and pricing from unstructured or multilingual documents. Cross-checks details against company logic, ensuring compliance and full auditability. Generates standardized invoices and posts to ERP systems, accelerating revenue recognition. Manages anomalies via self-healing and HALP (Human-Augmented LLM Processing). Captures invoices from any format, extracting supplier names, dates, amounts, and tax references. Categorizes invoices, applies workflows, and logs all steps for audit. ### Results Together, O2C and P2P automation saved 58,000+ hours annually, achieving 98% straight-through processing and accelerating both revenue recognition and supplier payments. ## Fund Performance Reporting Automation URL: https://maisa.ai/use-cases/fund-performance-reporting A global investment bank automated fund performance reporting with Maisa Digital Workers, replacing manual data extraction and consolidation with compliant, audit-ready reports delivered in minutes. A global investment bank redesigned its fund performance reporting with Maisa Digital Workers. Manual data extraction, cleaning, and consolidation across platforms is now an automated workflow that delivers compliant, audit-ready reports in minutes, so teams focus on analysis instead of data prep. ### Challenge Fund performance reporting is one of the most repetitive and time-intensive processes in wealth and investment management. Analysts must compile data from various systems such as Bloomberg, Morningstar, and Aladdin to produce monthly, quarterly, and ad hoc reports. The process required significant manual effort to clean, reconcile, and interpret figures across different funds and time periods. Each report followed unique templates and logic that varied by product, making automation difficult. This dependency on fund-specific rules created inconsistencies and made scaling across portfolios nearly impossible. Sensitive client data and strict regulatory requirements added further complexity. Compliance standards demanded accuracy and traceability, but manual handling increased the risk of errors and data exposure. As a result, teams spent days completing tasks that delayed insights and limited their ability to respond quickly to changing market conditions. ### Solution Using Maisa Studio, the institution developed a Digital Worker that manages the entire reporting cycle, from data ingestion to final report delivery. Configured in natural language by operations and analytics teams, the Digital Worker integrates securely with internal systems and applies business rules consistently across all funds. – Connects to internal data sources and reporting platforms to extract and transform performance information automatically. – Performs return calculations, generates narrative summaries, and formats outputs for multiple reporting frequencies and fund structures. – Produces results in standardized formats such as Excel, JSON, and XML, ensuring alignment with regulatory and client requirements. – All outputs are versioned and traceable, allowing compliance teams to verify every figure and change through a secure audit trail. – The Digital Worker is adaptable to future reporting expansion, creating a foundation for goal-based investment reporting and other advanced analytics. ### Results Report generation time was reduced from days to minutes, giving teams faster access to accurate performance insights. The automated system adapts to new fund types, frequencies, and reporting formats without reconfiguration. Each report is versioned and reviewable, providing a clear history of calculations and outputs. All data remains within the bank’s infrastructure, maintaining complete control and compliance with internal policies. ## Subpoena Request Intake Automation URL: https://maisa.ai/use-cases/subpoenas-requests-intake A leading financial institution restructured subpoena and legal request management with automated intake and processing, replacing a manual, cross-department process prone to delays and compliance risk. ### Challenge Handling subpoenas and similar legal information requests demands absolute precision and timely coordination. Each request must be validated, routed, and answered according to strict regulatory requirements, yet the process is often slowed by manual classification and fragmented data systems. At this global bank, legal and compliance teams relied on manual reviews to identify request types and gather data. Information was spread across customer, product, and operational systems, forcing staff to move between platforms and validate details repeatedly. Each submission required careful compilation and review, but without a unified workflow, progress tracking was limited. Inconsistencies between teams increased the risk of delays, incomplete data, and regulatory noncompliance. ### Solution Using Maisa Studio, the bank’s legal operations team created a Maisa Digital Worker that manages subpoenas and legal requests from start to finish. Configured in natural language, the Digital Worker follows institutional rules and compliance standards to ensure accuracy, speed, and transparency. Each incoming request is analyzed and categorized automatically, with routing to the correct workflow or department. Maisa connects securely to relevant internal databases to retrieve necessary client, product, and transaction details. Collected data is organized into standardized documents ready for review, aligned with internal templates and regulatory formats. Routine requests are completed automatically, while sensitive cases are forwarded to legal teams for verification. Every request and response is logged with a complete record of actions taken for compliance checks and audit reviews. ### Results Requests that previously required several days are now completed in minutes, giving legal teams more time to focus on complex matters. Automated classification and response logic eliminate misrouting and reduce the potential for missing or inaccurate information. Each response is traceable, fully documented, and aligned with the institution’s internal standards and external legal requirements. ## RFP Response Generation Automation URL: https://maisa.ai/use-cases/rfp-response-generation A leading financial institution transformed RFP management with intelligent workflows that search internal documentation, generate compliant answers, and deliver ready-to-submit drafts in hours instead of days. ### Challenge Managing RFPs at scale represents a significant operational burden for large financial institutions. Each request involves hundreds of unique questions requiring precise answers drawn from approved internal content. The existing process relied entirely on manual effort. Teams navigated extensive internal documentation libraries, reviewed previous submissions, policies, and technical papers to locate relevant information. Document inconsistencies forced specialists to reconstruct answers repeatedly and validate every response for compliance, leading to extended preparation times, inconsistent messaging, and risks of using outdated materials. Cross-departmental collaboration moved slowly with limited visibility into progress. ### Solution Using Maisa Studio, the bank’s commercial operations and procurement teams built a Digital Worker handling the complete RFP process. Configured in natural language, it connects directly with internal libraries, ensuring responses remain accurate, current, and aligned with corporate standards. Uploads RFP documents, reads sections, and automatically identifies all questions and response requirements Creates first drafts by analyzing historical data, previous submissions, and approved documentation with automatic formatting for Word, PowerPoint, or Excel Establishes a single global document library providing verified, current content to all teams Assigns questions to teams, tracks real-time progress, maintains version control and transparency Searches internal knowledge repositories to locate relevant compliant information Provides progress insights, alerts, and change control management ### Results RFPs complete using verified content, eliminating manual assembly and review Response time reduced dramatically, enabling teams to handle multiple RFPs concurrently Systematic search and validation ensure compliance and alignment with approved documentation Teams redirect focus toward strategic initiatives and vendor management rather than repetitive content retrieval ## Supply Chain Task Resolution Automation URL: https://maisa.ai/use-cases/supply-chain-task-resolution-automation A global manufacturing leader transformed its exception handling workflow by deploying Digital Workers to automate supply chain task resolution, achieving dramatic gains in speed, accuracy, and capacity. ### 6x Capacity increase ### 85% Processing time reduction ### 95% Error rate decrease ### Challenge The supply chain monitoring team received more than one thousand operational tasks every day, each containing different levels of information such as images, emails, supplier notes, and contextual data. Analysts had to manually open every task, interpret the issue, cross check information in the CMMS, confirm shipping plans and delivery status, review program changes, and validate buffer stock levels. With five to ten minutes required per task, the team struggled to keep pace. The speed and volume of the workflow made human mistakes almost certain. Misinterpreting a deviation could disrupt production schedules. Overlooking a delay could trigger penalties or expensive expedited shipments. Closing a task too early could hide a real supply issue. Even small oversights created additional work and financial impact across the supply chain. Under constant pressure, analysts had no room for strategic work and were forced into a purely reactive posture, focused only on clearing the queue instead of improving processes or preventing future issues. ### Solution Maisa Digital Workers automate the complete task resolution workflow with clear, consistent, and fully transparent logic. After the process was defined in natural language, the Supply Chain Monitoring Team deployed a Digital Worker that now manages every step of the workflow from beginning to end. The Digital Worker ingests every incoming task, reads the description, reviews images and attachments, understands supplier notes, and incorporates all available context. It then checks operational data inside the CMMS, including shipping plans, delivery records, supplier schedules, program adjustments, and buffer stock. This allows it to determine whether the issue is real, already resolved, or covered by available capacity. Using the organization’s supply chain logic, the Digital Worker decides whether to close the task, progress it for monitoring, or escalate it with clear follow up actions. It then writes a complete explanation directly inside the task comments, detailing what was checked, what data sources were used, and why the final decision was reached. Analysts receive clarity instantly without opening any other system. ### Results By automating the entire task evaluation process, the organization achieved transformative performance improvements. Task resolution time decreased by eighty five percent, enabling the team to move from one thousand tasks per day to more than six thousand with the same number of analysts. The team now saves between sixty seven and one hundred fifty hours every day. The error rate decreased by more than ninety five percent, preventing production delays, supplier escalations, unnecessary expedited shipments, and penalty costs. Sub minute cycle times now provide real time visibility and a stronger and more predictable supply chain. ## Equity Research Automation URL: https://maisa.ai/use-cases/equity-research-investment A leading investment management group automated the extraction and structuring of insights from broker reports, turning unstructured content into actionable intelligence in minutes and enabling analysts to focus on high-impact investment strategy. ### 5x Processing time reduction ### 30min to < 5 Per-report processing ### 100% Automation rate ### Challenge Equity research plays a central role in investment decision-making, but it depends on time-intensive manual analysis of broker reports. Each document contains a mix of charts, tables, and complex commentary that analysts must read, interpret, and summarize before producing insights for portfolio managers and clients. The research team faced a growing backlog of unstructured documents and lacked standardization across sources. Each report took about thirty minutes and nearly twenty manual steps to process, from data extraction to summary preparation. As the volume of broker coverage expanded, analysts struggled to maintain quality, consistency, and timeliness. Traditional automation solutions could not handle the variety of file formats or understand the financial and qualitative nuances of the reports. This limited the firm’s ability to scale coverage, accelerate research cycles, and respond quickly to market shifts. ### Solution Using Maisa Studio, the firm’s Wealth and Investment Management team created a Digital Worker that processes broker research reports end to end. Built without coding and guided by natural language instructions, the worker extracts, interprets, and structures insights with precision and consistency. The Digital Worker reads and parses text, tables, and charts from PDF reports regardless of layout or formatting. It identifies relevant sections and converts them into standardized, machine-readable fields ready for analysis. Summarizes target prices, recommendations, earnings expectations, and qualitative commentary into concise, structured insights, aligned with the firm’s reporting formats and KPIs. Creates ready-to-use outputs such as Excel templates and narrative summaries, enabling analysts to review and share insights instantly across teams and systems. Scales seamlessly across brokers and research teams with minimal setup, allowing rapid expansion of coverage across asset classes and sectors without additional engineering effort. By integrating document intelligence with domain expertise, Maisa Digital Workers turned a manual research workflow into a scalable, automated intelligence system for investment analysis. ### Results Research reports that once required extensive manual review are now processed in less than five minutes. Structured insights accelerate decision-making and allow analysts to focus on strategic analysis rather than administrative tasks. The automated process enables consistent and rapid processing of large volumes of broker research, creating a foundation for global research scalability. Establishes a roadmap for future AI-driven research synthesis, enabling cross-broker comparisons and multi-document insights at scale. ## Certification Generation for Electricity Network Maintenance URL: https://maisa.ai/use-cases/certification-generation A global energy and utilities company transformed its electricity network maintenance by automating certification generation, removing manual bottlenecks and speeding up billing cycles. ### 80% Reduction in processing time ### 14,600 hours Saved annually ### 95% Automation rate ### Challenge The client’s electricity network maintenance workflow relied on fragmented, manual steps that created operational bottlenecks and delayed billing. Field teams submitted documentation in mixed formats — photos, handwritten reports, PDFs, scans, and spreadsheets — making interpretation and validation slow and inconsistent. Administrators had to manually match tasks to thousands of UCC/baremo pricing codes, a tedious process complicated by unstructured descriptions and regional variations. Inventory updates were also handled manually: equipment usage data had to be sent to the ERP team separately, exposing the lack of integration across legacy systems like EMAX. Errors in price lookup, inconsistent document formats, and repeated rework further slowed the work-to-cash cycle. Traditional BPM and RPA tools failed due to the high variability and complexity of the cases. ### Solution Maisa’s Digital Worker fully automated the client’s end-to-end certification and billing process, eliminating manual data handling and connecting field operations with pricing, documentation, and ERP systems. The Worker collects all relevant input — from photos and handwritten notes to Excel and material usage reports — classifies, reads, and extracts structured information, then applies the client’s pricing catalog to each task. It generates complete Excel certification files, summary reports, and client invoices, while automatically updating inventory in EMAX. A built-in accuracy assurance layer reviews all outputs, sending only 5% of cases for human validation. ### Results By automating certification generation, the company removed major bottlenecks between field work, documentation, and billing. Processes that previously took days now run overnight, dramatically accelerating invoice delivery and improving cash flow. Each certification now takes about 10 minutes instead of 1.5 hours — recovering 14,600 hours annually, the equivalent of nine full-time employees. Automated pricing and rule application now ensure consistent and error-free results across all regions, while real-time ERP updates keep inventory always accurate. The solution also scales easily across divisions, providing a foundation for broader enterprise automation. Consistent, auditable, error-free certification Faster billing and shorter payment cycles Always up-to-date inventory records ## Trade Finance Process Automation URL: https://maisa.ai/use-cases/trade-finance-processing A leading global bank's Corporate and Investment Banking division transformed one of its hardest processes to automate. A leading global bank’s Corporate and Investment Banking division transformed one of its hardest processes to automate. ### Challenge The bank’s Corporate and Investment Banking team processes numerous trade finance transactions daily, each requiring verification of unstructured documents in various formats including contracts, invoices, and certificates across multiple countries and regulatory frameworks. Manual review by over 300 specialists consumed more than an hour per case. Documents contained stamps, signatures, handwriting, and scanned images that resisted automation. Regulatory changes required months of rework and retraining. Previous automation tools such as OCR and RPA could not handle the variety and ambiguity present in trade finance documents, creating bottlenecks, delaying client transactions, and preventing operational scaling. ### Solution Using Maisa Studio, the operations team built a Digital Worker in minutes using natural language instructions, requiring no coding expertise. The Digital Worker manages the complete trade finance verification process: Processes both digital and scanned PDFs without embedded text layers Verifies key data points, beneficiaries, terms, and compliance requirements across multiple sources Identifies information gaps and searches related documents for additional context Compiles complete summaries with traceable, auditable decisions Transforms expertise into reusable assets that evolve with each execution ### Results Trade finance verification time decreased from hours to minutes End-to-end automation eliminated manual bottlenecks and accelerated client service Each execution captured expert knowledge, creating self-improving automation deployable across regions Traceable, rule-aligned automation ensured accuracy, compliance, and audit confidence ## New Client Onboarding Automation URL: https://maisa.ai/use-cases/new-client-onboarding A global financial institution automated client onboarding with Maisa Digital Workers: documents are classified, analyzed and processed automatically, cutting activation time and improving accuracy across regions. A global financial institution modernized its client onboarding process using Maisa Digital Workers, automatically classifying, analyzing, and processing onboarding documentation to dramatically reduce activation time while improving accuracy and scalability across regions. ### 6x Capacity increase ### -80% Processing time ### 93% Automation rate ### Challenge Onboarding new clients in wealth and investment management involves managing a wide range of documents, from identity verification and IBAN confirmations to proof of address and income validation. Each file type has different formats, layouts, and requirements, making automation difficult and slowing customer activation. The bank’s teams had to manually open and review each document, sort it into the right category, and extract key fields such as account numbers, addresses, and signatures. PDFs often contained multiple document types combined into a single file, with irrelevant pages that required manual filtering. These repetitive and time-consuming tasks extended onboarding timelines and created inconsistency between regions and teams. With thousands of clients processed each month, even small inefficiencies translated into major operational costs and slower service delivery. ### Solution Using Maisa Studio, the bank’s operations and compliance teams configured a Digital Worker to automate onboarding document processing from end to end. Built using natural language instructions, the Digital Worker requires no technical expertise and aligns perfectly with internal procedures. Reviews multi-page PDFs and identifies relevant pages for processing, ignoring irrelevant or duplicate content. Recognizes and classifies document types such as identity proofs, bank statements, and payslips without human input. Captures essential data fields from each document category — including client details, account identifiers, and verification information — and prepares structured outputs for system integration. Organizes extracted information into standardized formats that feed directly into onboarding systems and compliance checks. This seamless process ensures that every client document is validated, structured, and stored consistently across departments, enabling faster activation and more reliable compliance documentation. ### Results Client activation time was significantly reduced as document review and validation became fully automated. Documents are now sorted and categorized without manual intervention, eliminating errors and saving operational time. Critical client data is captured systematically, improving the quality of downstream processes like KYC verification and account creation. The new onboarding workflow handles rising document volumes effortlessly, enabling the institution to scale its client acquisition capacity without adding staff. # Solutions & Concepts ## Agentic Process Automation URL: https://maisa.ai/landings/agentic-process-automation Advanced automation powered by AI Agents to handle complex, dynamic business processes. AI Glossary #### What Is Agentic Process Automation? Agentic Process Automation (APA) is an advanced automation approach that leverages AI agents, to autonomously handle and execute complex business processes requiring cognitive abilities such as interpretation, reasoning, and decision-making. Unlike traditional automation methods reliant on predefined rules or scripts, APA utilizes intelligent AI agents capable of understanding unstructured data, interpreting human language, and dynamically adapting their approach as circumstances change. APA systems are characterized by being: - Autonomous: Agents independently determine how to achieve objectives without continuous guidance, reducing the need for detailed scripting. - Goal-driven: Users define desired outcomes, and agents automatically identify and execute optimal paths to achieve them. - Context-aware: Agents adjust in real-time to handle unexpected situations and exceptions smoothly, maintaining efficient workflows. APA enables businesses to automate tasks that were once too complex or nuanced for traditional automation, significantly expanding the possibilities of process automation. #### Why Agentic Process Automation? Agentic Process Automation brings flexibility and intelligence to process automation, making it possible to automate tasks that once required human judgment. ##### Handles higher complexity APA goes beyond simple, repeatable tasks. It interprets unstructured information, reasons through ambiguity, and adapts in real time to shifting conditions. ##### Flexible and efficient to scale With a goal-based setup, there’s no need to define every step in advance. This reduces setup time, lowers maintenance overhead, and makes it easier to scale automation alongside evolving processes and needs. ##### Continuously adapts and improves APA agents learn from feedback and evolve over time. They adapt to context, handle exceptions, and refine their performance with each execution. #### How APA Works Using Agentic Process Automation is simple and adaptable. Instead of hardcoding every step, you define what you want to achieve, and the system figures out how to get there. ##### 1. Define the goal Start by describing the outcome you want. Add any relevant instructions, tools, or context the system should use to get the job done. ##### 2. Run the automation The agent plans and executes the necessary steps using the available tools and data. It adapts along the way, handling unexpected inputs or changes without manual intervention. ##### 3. Improve over time The system learns from each execution. With every run, it becomes more efficient, more accurate, and better aligned with how work actually happens. #### Comparison with Traditional Automation Traditional automation tools are great at executing simple, repetitive tasks—especially when everything follows a predictable path. But most real-world processes aren’t that clean. They involve exceptions, unstructured inputs, and decisions that don’t fit into fixed rules. This is where Agentic Process Automation stands apart. Instead of automating individual tasks with hard-coded logic, APA enables full processes to be handled end-to-end, even when they’re complex, unpredictable, or involve judgment. APA takes a different approach. It understands the broader context, adjusts on the fly, and makes informed decisions at each step. Traditional tools can’t do that. This shift opens up new possibilities. It makes it possible to automate workflows that were previously too complex or dynamic for traditional tools to handle. ##### Agentic Process Automation vs Robotic Process Automation Robotic Process Automation (RPA) is designed to mimic human actions within structured systems. It follows fixed rules and predefined steps—great for repetitive tasks like data entry or invoice matching. APA extends automation far beyond that. Instead of mimicking clicks, it understands goals. It handles variability, interprets context, and adjusts its plan in real time. Where RPA follows instructions, APA reasons. It’s a shift from task automation to process intelligence. Dimension RPA (Robotic Process Automation) APA (Agentic Process Automation) Core Function Mimics human actions following predefined rules Achieves outcomes using autonomous AI agents Flexibility Low — follows rigid, predefined rules High — adapts to changing conditions and open-ended scenarios Task Handling Structured, repetitive tasks (e.g. data entry) Complex, unpredictable tasks requiring judgment and reasoning Data Type Structured data only Structured and unstructured data Error Handling Manual intervention needed Self-healing — detects and corrects errors autonomously Decision-Making None — fully predefined logic AI-driven reasoning and dynamic decisions User Input Requires detailed, step-by-step instructions Requires outcome specification — goal-based execution Learning Capability None Learns continuously from context and feedback Speed of Deployment Slower — requires detailed programming and updates Faster — adapts workflows with minimal setup Explainability High — deterministic by design Varies Maintenance Cost High — sensitive to change Lower — adapts automatically, reducing update needs Scalability Limited to repetitive workflows Broad — spans diverse, dynamic processes Example Tools UiPath, Automation Anywhere, Blue Prism Maisa (Digital Workers), Beam AI, Sema4ai ##### What’s possible to automate with Agentic Process Automation APA is ideal for processes that go beyond simple tasks. It thrives in environments where decisions depend on changing inputs, multiple tools are involved, and the logic can’t be hardcoded upfront. These are often workflows that: - Involve judgment or ambiguity - Require coordination across tools and systems - Depend on multi-step decisions, where each step is influenced by the outcome of the last APA handles them end-to-end, only involving people when needed. Examples: Invoice processing: APA can extract data from invoices, cross-check amounts and vendors against ERP records, flag discrepancies, and trigger payment approvals—without needing predefined templates. It adapts to format variations and learns from feedback to improve accuracy over time. Customer support triage: APA can read incoming messages, identify the issue type, determine the best course of action (respond, escalate, or assign), and update relevant systems. It adapts to changing customer language and can resolve repetitive tickets autonomously, reserving human attention for edge cases. ##### APA benefits for business APA helps reduce friction in day-to-day operations by automating tasks that involve multiple steps, tools, and decisions. It lowers execution costs by cutting manual handoffs, reducing errors through built-in checks, and adapting in real time when conditions change. This means fewer delays, less time spent fixing issues, and more consistent outcomes. Because APA evaluates context at every step, it can act with precision instead of following static rules. It keeps processes aligned with the current state of the business without needing constant updates. Over time, APA improves how processes run. It adjusts to new inputs, tools, and requirements—without having to be reprogrammed each time. ## Chain of Work URL: https://maisa.ai/landings/chain-of-work Chain of Work makes AI execution deterministic, traceable, and transparent, eliminating guesswork and enabling reliable, auditable automation at scale. AI Glossary #### What is Chain of Work? Chain of Work is a structured, traceable log that records every step of AI execution. Every decision, process, and action is systematically logged, making AI workflows fully auditable and transparent. It is code-based, ensuring execution follows predefined logic rather than probabilistic estimations. Unlike Chain of Thought, which relies on sequential LLM calls to generate answers, Chain of Work doesn’t depend on probabilistic outputs. Instead, AI orchestrates tools, processes, and data in a deterministic and structured way, ensuring predictable and verifiable execution. This approach operates like a computational system, where execution follows a clear, logical sequence. This ensures: - Traceability Every step is recorded and reviewable. - Transparency AI actions are explainable and accountable. - Reliability The same input always produces the same output. #### Why is Chain of Work needed? Most AI systems today function as black boxes—they generate responses, but we often don’t know how or why they reached a specific conclusion. This is because they rely on probabilistic models, which predict the most statistically likely answer rather than following a structured, logical process. This lack of transparency creates serious challenges for businesses that need AI to be reliable, explainable, and accountable in critical operations. Key problems with traditional AI systems - Hallucinations AI can generate false or misleading information that sounds plausible but has no factual basis. - Lack of Traceability There is no clear record of how decisions are made, making it difficult to verify outputs. - Inconsistent Results The same input can lead to different outputs, reducing reliability in decision-making. In high-stakes environments—such as business operations, compliance, and automation—this unpredictability makes AI hard to trust. Without a way to verify each step of its reasoning, businesses risk relying on AI-driven decisions that cannot be explained or corrected. #### How Chain of Work solves this Instead of relying on probabilities, Chain of Work ensures AI follows a deterministic process, orchestrating tools, data, and logic in a fully traceable way. Every decision and action is logged, creating a structured audit trail that makes AI execution transparent and accountable. This eliminates randomness, ensuring: - Determinism The same input always produces the same output. - Traceability Every step, tool, and data source is recorded. - Audibility Clear reasoning paths allow verification and refinement. - Error correction If something goes wrong, it’s identifiable and fixable, not an unpredictable failure. Chain of Work makes AI function like a computational system—navigating facts, executing processes logically, and delivering consistent, reliable results. It removes uncertainty from AI-driven automation, making it transparent, accountable, and ready for real-world use. #### Key differentiators of Chain of Work Chain of Work ensures deterministic, traceable, and structured AI execution, eliminating guesswork. Here’s how it compares: Feature Chain of Thought RAG Deterministic ❌ No ✅ Yes Fully Traceable ⚠️ Partial Prevents Hallucinations Uses Structured Execution Works like a Computational System Unlike Chain of Thought, which relies on probabilistic LLM calls, Chain of Work follows a structured process. Unlike RAG, which retrieves external data but still relies on statistical matching, Chain of Work ensures logical, repeatable execution. By removing uncertainty and enforcing structure, Chain of Work makes AI as reliable as a computational system. #### Why Chain of Work matters for enterprises Businesses need execution that is consistent, accountable, and explainable. Yet, many AI systems function unpredictably, making it difficult to trust their outputs in critical areas like operations, compliance, and decision-making. Without a clear execution process, organizations face challenges in maintaining oversight, reducing risk, and ensuring compliance. When AI-generated outcomes cannot be traced or verified, businesses struggle to integrate automation with confidence. Chain of Work provides a structured approach, ensuring decisions and actions can always be reviewed and understood. This makes it easier to implement AI in workflows that demand precision and accountability. Whether automating tasks, enforcing regulatory standards, or improving strategic insights, businesses gain a system they can trust—one that enhances efficiency without sacrificing control. #### Beyond predictions AI should do more than generate responses—it should execute work in a structured, deterministic way. Businesses need AI systems they can trust, where every action is explainable, repeatable, and auditable. Chain of Work provides this foundation by ensuring AI follows clear execution steps, eliminating uncertainty and making automation reliable at scale. It transforms AI from a black box into a transparent system that businesses can fully integrate into critical operations. Built into Maisa’s KPU, Chain of Work ensures that AI execution is no longer a guessing game but a structured, accountable process—delivering consistency, accuracy, and control where it matters most. ## Agentic AI URL: https://maisa.ai/landings/agentic-ai The shift from reactive AI to proactive, autonomous agents. AI Glossary #### What is Agentic AI? Agentic AI refers to AI systems that can decide and act on their own to reach a goal. These systems decide how to act, plan the necessary steps, and execute them independently. This represents a shift from reactive to proactive systems. Reactive systems respond to each input with an independent output, like ChatGPT answering individual questions. Proactive systems are goal-oriented. They work toward objectives on their own, adapting their approach as needed. Agentic AI is the capability behind truly autonomous AI agents. But many systems labeled as “AI agents” in the market don’t exhibit full agentic behavior. They follow pre-scripted logic or fixed workflows. In contrast, agentic systems let the AI itself drive both planning and execution, continuously adapting its approach to reach the intended goal. #### Key Capabilities Agentic systems can take various forms, yet certain core capabilities characterize how these systems operate: ##### Autonomy Agentic AI exhibits goal-oriented behavior, completing full end-to-end workflows without constant human direction. Once given an objective, it determines and executes the necessary steps independently. ##### Adaptability When conditions change or obstacles appear, these systems adjust their strategy dynamically. They don’t fail when the unexpected happens; they find alternative paths forward. ##### Handling complexity Agentic AI manages multi-step tasks that require reasoning and decision-making. It navigates through interconnected processes, making choices at each stage to progress toward the goal. #### How it works Agentic AI usually combines several mechanisms to enable autonomous action. ##### 1. Reasoning At the core, these systems use reasoning capabilities, often powered by large language models. They interpret goals, understand context, and make decisions toward achieving the objective. This reasoning layer determines what needs to be done and evaluates the best approach given current circumstances. ##### 2. Tool and Data Access To act on their decisions, agentic systems connect to APIs, software, and databases. This allows them to gather necessary context, retrieve information, and take concrete actions in external systems. ##### 3. Orchestration These systems orchestrate all the components needed to reach a goal: information, tools, and sequential steps. They coordinate multiple processes, manage dependencies between tasks, and ensure each action builds toward the intended outcome. #### Benefits of Agentic AI Agentic AI offers practical advantages that directly impact operations. ##### Efficiency These systems reduce operational bottlenecks and eliminate repetitive manual work. By automating entire workflows rather than individual tasks, they streamline processes that previously required constant human intervention. ##### Scalability Agentic AI handles increasingly complex processes without proportional increases in resources. As operations grow or requirements become more sophisticated, these systems adapt and manage the additional complexity without requiring equivalent expansion in staffing or infrastructure. #### Challenges While powerful, agentic AI systems present distinct challenges that organizations must address. ##### Reliability These systems face risks like hallucinations or compounding errors across multiple steps. When AI makes sequential decisions autonomously, a single error early in the process can cascade through subsequent actions, potentially amplifying the impact. ##### Control Organizations need guardrails to balance autonomy with oversight. The challenge lies in giving systems enough freedom to operate effectively while maintaining appropriate boundaries and intervention points when needed. ##### Traceability Visibility into decisions and actions becomes critical for accountability. Understanding how and why an agentic system made specific choices is essential for debugging, compliance, and maintaining trust in automated processes. ## AI Agents URL: https://maisa.ai/landings/ai-agents AI Agents explained: what they are, how they work, and why they matter for automation. AI Glossary #### What are AI Agents AI agents are software systems that use AI to pursue goals and complete tasks. AI is at the core of how agents work. It lets them understand goals, reason through tasks, and make decisions based on context. They don’t just follow instructions. They adapt to the context around them. They also take action. AI agents can connect to tools and systems to do things like send an email or update a database. The term AI agent is used in many ways today, and the market is full of noise. Some describe any system that includes AI in part of the process as an agent. We define AI agents as systems with AI at their core—systems that are objective-driven, capable of reasoning, and able to decide how to accomplish tasks. - Workflows with agent-like behavior: These follow predefined paths but may include AI models or tools in some steps. The logic is fixed, and outcomes are limited to what’s been planned. - AI Agents: These are systems where language models dynamically direct their own processes and tool usage, maintaining control over how they accomplish the task from start to finish. #### AI Agents vs AI Assistants We’re all familiar with AI assistants. Tools like ChatGPT or Copilot help us write, summarize, or answer questions. They’re helpful, but they rely on us to guide them, one prompt at a time. AI agents are a step further. They’re not just responding. They’re acting. Agents can make decisions, use tools, and complete tasks on their own, without being told what to do at every step. AI Agent AI Assistant Purpose Acts on its own to complete tasks and reach goals Helps users by following instructions or prompts Capabilities Can handle complex tasks, make decisions, adapt, and learn over time Provides answers, suggestions, or simple actions based on input Interaction Proactive and goal-driven Reactive and prompt-based #### How AI Agents Work AI agents work by combining different parts that let them reason, act, and in some cases, learn from experience. The way these parts are used can vary depending on how the agent is designed. ##### 1. AI Model At the center is usually a language model. It’s what lets the agent understand a goal, break it into steps, and make decisions along the way. This is the reasoning engine. It plans, reacts, and adjusts based on the context. ##### 2. Tool Access To take action, agents rely on tools. These can include APIs, databases, or other software. The model decides what to use and when, depending on the task. This is how agents move from planning to actually getting things done. #### Capabilities of AI Agents AI agents can do more than just respond to prompts. Their capabilities make them useful for handling complex tasks, adapting to new situations, and acting on their own. ##### Autonomy Agents work toward goals without needing step-by-step instructions. Once given an objective, they decide how to move forward. ##### Planning They break tasks into smaller steps, sequence them, and adjust as things change. This includes handling edge cases or exceptions. ##### Action Agents can use tools to complete tasks. They interact with systems, send messages, retrieve data, or trigger workflows. ##### Complexity They can manage tasks that involve multiple steps, decisions, or tools—keeping track of what needs to happen and in what order. ##### Learning Some agents improve over time by storing past experiences, learning from feedback, or adapting to user behavior. #### Challenges of AI Agents As powerful as AI agents can be, using them in real-world scenarios comes with challenges. These systems need to be reliable, understandable, and manageable, especially in business contexts where accuracy and trust matter. ##### Traceability AI models often behave like black boxes. It can be hard to see why an agent made a certain decision. Without traceability into the reasoning process, it’s difficult for teams to trust or verify the outcome. ##### Reliability Since agents often rely on language models, there’s a risk of them generating false or misleading information. This can impact accuracy and trust. ##### Scale Running agents at scale is complex. Their lack of traceability and reliance on AI outputs make it harder to manage errors and ensure reliability. Coordinating tools, tasks, and learning systems adds another layer of difficulty. #### Digital Workers: AI Agents for the Enterprise AI agents open up new ways to automate and scale work. But in practice, teams often struggle with control, reliability, and trust. That’s where digital workers come in. Digital workers are a type of AI agent built specifically for business environments. They can manage full workflows from start to finish, with the kind of structure and visibility companies need. What makes them different is their focus on traceability and accountability. Each step is visible. You can see what decisions were made, what tools were used, and why. That means fewer surprises and more trust. They’re designed to handle real business tasks—coordinating across systems, adapting to changes, and keeping a clear record of everything they do. So instead of just experimenting with AI, teams can put it to work. ## AI Computer URL: https://maisa.ai/landings/ai-computer A new computing paradigm where AI acts as the core orchestrator, managing tools, data, and tasks to deliver real outcomes, not just answers. AI Glossary #### What is an AI Computer? An AI Computer is a computational system designed around artificial intelligence as its central orchestrator. At its core lies an AI orchestrator or kernel, responsible for managing and coordinating all activities within the system. Consider this analogy: traditional computers rely on kernels, like Windows or Linux, to manage hardware and software seamlessly. Similarly, the AI Computer uses an AI kernel to autonomously coordinate tools, modules, and real-time data sources, effectively handling tasks and workflows without continuous human oversight. In essence, just as a computer kernel abstracts complex processes to make computing efficient and accessible, an AI Computer abstracts the complexity of AI-driven tasks, orchestrating various tools and processes to deliver outcomes aligned precisely with user objectives. #### Why the AI Computer? Today’s AI systems come with common frustrations: they can produce incorrect or entirely fabricated answers (known as hallucinations), struggle with maintaining adequate context, and often require complicated setups involving integrations, data flows, and prompt engineering. These issues can make AI unreliable and difficult to work with effectively. The AI Computer addresses these problems head-on: Reliability: Rather than generating standalone answers, the AI Computer orchestrates tasks between tools, processes, and calculations, using clear, code-based steps. This means any error can be traced and understood through a transparent log of actions, avoiding the “black box” uncertainty common in other AI systems. Context Management: AI needs relevant context to produce accurate results, yet most AI models have strict limits on how much context they can handle at once. The AI Computer solves this by incorporating a dedicated memory module, intelligently selecting and providing only the necessary information at each step—much like how traditional computers use RAM and cache to efficiently manage data. Simplified Complexity: Just as your computer’s operating system handles memory management, hardware interactions, and other complex tasks behind the scenes, the AI Computer abstracts away AI-related complexities. Users simply define their objectives, specify tools and information, and let the AI Computer manage the execution seamlessly. #### Key capabilities ##### Real-Time Data Integration Continuously pulls current data from APIs, web services, and databases, ensuring your business decisions are always based on the latest information. ##### Limitless Context Management Intelligently organizes and accesses memory, making sure the AI always uses the precise context needed at every step. ##### Transparent Reliability Executes tasks through clear, step-by-step processes, leaving a detailed trail. This transparency prevents errors and removes uncertainty. ##### Autonomous Orchestration Understands your goals, automatically choosing and coordinating the best tools, models, and processes to complete tasks effectively and independently. #### Maisa KPU The KPU is Maisa’s AI Computer, built from research to orchestrate tools, data, and workflows clearly and effectively based on user intent. It serves as the technological foundation for Maisa Studio and Digital Workers, enabling automation of complete business processes seamlessly and transparently. At the heart of Maisa’s mission is making AI accountable—ensuring every decision and step is clear, auditable, and reliable. By combining AI’s creative problem-solving capabilities with the transparency and reliability of traditional computing, the KPU delivers the clarity and control businesses truly need. ## AI Hallucinations URL: https://maisa.ai/landings/ai-hallucinations AI hallucinations occur when models generate plausible but false information. Learn why they happen and how to reduce them for more reliable AI outputs. AI Glossary #### What are AI Hallucinations? An AI hallucination happens when a generative AI model produces information that sounds correct but is factually wrong. The output reads fluently, uses the right tone, and might even cite sources. But the underlying content is fabricated. In short, the model is optimizing for plausibility, not accuracy. The term borrows from human psychology, but the mechanism is completely different. A person who hallucinates perceives something that isn’t there. An AI that hallucinates generates something that was never there, and presents it with full confidence. #### Why Do AI Hallucinations Happen? Large language models don’t understand what they’re saying. They predict the next word in a sequence based on patterns in their training data, which is billions of examples of human text. When a model hits a question it wasn’t explicitly trained on, it doesn’t stop and say “I don’t know.” It fills the gap with whatever sounds right. That’s the core issue. These models are built to sound convincing, not to be accurate. A confidently written paragraph about a court case that never happened looks identical to one about a real case. The model can’t tell the difference, and unless you check, neither can you. ##### How to mitigate hallucinations There are several strategies to mitigate hallucinations in AI systems’ output. ##### What Actually Reduces Hallucinations? There’s a common assumption that bigger models hallucinate less. That’s partially true, but it misses the point. What matters more than size is how the model was trained, what data it learned from, and what guardrails sit around it. Techniques like reinforcement learning from human feedback (RLHF) teach models to favor accurate, grounded responses over fluent-sounding guesses. Better data curation removes contradictions and low-quality sources from training sets. And frameworks like RAG (covered in more detail below) give models access to verified external information when generating a response, rather than forcing them to rely on memory alone. Scale helps. Larger models can hold more nuanced representations of knowledge. But scale without these other improvements just produces a model that’s better at sounding right while still being wrong. ##### How to Reduce AI Hallucinations Hallucinations can’t be fully eliminated, but they can be reduced significantly. Some of this is on the user side, some on the developer side, and the most effective approaches combine both. - Writing Specific Prompts: The more context and detail you give a model, the less room it has to guess. Vague or open-ended prompts are an invitation for fabrication. Tell it what you need, what format you want, and what constraints apply. - Provide reference material. If you give an AI specific documents, datasets, or sources to work with, it will ground its output in that material rather than falling back on pattern-matching from training data. - Constrain the model’s behavior. Models should be configured to flag uncertainty rather than fill gaps. If a system is allowed to say “I don’t have enough information to answer this,” it’s far less likely to fabricate a confident-sounding response. - Build in human escalation. This is the approach Maisa’s Digital Workers take, rather than fabricating answers or forcing process completion, they log the issue, analyze the outcome, and either retry or escalate to a human. - Test continuously. Hallucination patterns are easier to catch if you’re actively looking for them. Regular testing across diverse inputs helps identify failure modes before users encounter them. ##### What to Do If an AI Hallucination Occurs Prevention reduces the frequency, but hallucinations will still happen. When they do, the response matters as much as the prevention. - Fact-check the specifics. Once you suspect a hallucination, verify the details independently. Dates, names, statistics, and citations are the most common fabrication points. Cross-reference against primary sources, not just other AI outputs. - Rephrase and retry. A more specific prompt often produces a more grounded answer. Add constraints, narrow the scope, or ask the model to cite its reasoning. - Correct manually. For critical outputs, replace fabricated data points (dates, names, regulatory references, citations) with verified information before publishing or acting on the content. - Flag the error. Most AI tools include feedback mechanisms. Reporting hallucinations helps improve the model over time and contributes to better outputs for everyone. ##### Real-World Examples of AI Hallucinations Hallucinations have already caused real damage. - Legal: Fabricated case citations. In 2023, a New York attorney submitted a court filing that cited multiple cases generated by ChatGPT, none of which existed. The judge sanctioned both attorneys involved after discovering the fake precedents. The case became a cautionary example of what happens when AI-generated legal research goes unverified. - Academia: Ghost sources. Researchers have documented cases where AI tools generate citations that look plausible, complete with author names, journal titles, and publication dates, but point to papers that were never written. Students and researchers who rely on these without checking risk building arguments on foundations that don’t exist. - Customer service: Policy fabrications. AI chatbots have invented refund policies, discount codes, and service terms that the company never offered. In one widely reported case, an airline’s chatbot quoted a bereavement fare policy that didn’t exist, and the airline was later held to honor it. #### Why This Matters at Scale If a business relies on AI-generated outputs without systematic verification, the errors will compound. Financial services firms risk acting on fabricated compliance references. Legal teams risk citing non-existent precedent. Insurance operations risk processing claims against policy terms that were never written. The challenge is that these models produce outputs that look identical whether the underlying information is real or fabricated. At enterprise scale, where thousands of documents are processed daily, manual spot-checking isn’t enough. The verification has to be built into the system itself. #### Retrieval-Augmented Generation (RAG): Grounding Outputs in Real Data Much of the serious work on reducing hallucinations happens at the architectural level. The most widely adopted approach is Retrieval-Augmented Generation, or RAG. RAG gives a language model access to a specific external database before it generates an answer. The model retrieves relevant documents or data points and uses them as the basis for its response. The key word is “external.” The model is drawing on verified, up-to-date information not its own static memory. Developers typically organize this information in vector databases, which store data based on semantic meanings. This allows the retrieval step to surface genuinely relevant context, even when the user’s query doesn’t match the exact wording of the source material. RAG significantly reduces hallucinations, but it’s not a complete solution on its own. The model can still misinterpret retrieved data, or fail to retrieve the right documents. That’s why the most robust systems layer RAG with additional verification steps. #### The One Context Where Hallucinations Are Useful In creative and exploratory work, hallucinations sometimes produce something unexpectedly valuable. A game developer using AI to generate environment descriptions might get an output that doesn’t match the intended setting, but the “mistake” suggests a more interesting direction. A designer exploring brand concepts might find that an AI’s unexpected combination of elements sparks an idea they wouldn’t have reached on their own. The difference is context. In factual, compliance, or operational settings, hallucinations are errors. In creative work, they’re occasionally useful noise. The problem is never the hallucination itself. It’s treating a guess as a fact. #### Maisa: Hallucination-Resistant AI for the Enterprise in Regulated Industries Building on the idea that enterprise tools must be both effective and transparent, Maisa has developed a hallucination-resistant system that goes a step further than standard LLMs. This is possible because of Maisa’s Knowledge Processing Unit (KPU), an OS-level architecture built on three components. The Reasoning Engine orchestrates how the system approaches a problem, focusing on understanding the path to a solution rather than guessing at the answer. The Execution Engine processes and carries out each instruction with built-in self-recovery, meaning it can detect failures, retry, or escalate without human intervention. And the Virtual Context Window manages how information flows through the system, indexing and structuring data so the right context is available at the right step. Because the KPU is model-agnostic, it isn’t locked into any single LLM. As base models improve, performance also improves with them. But the real differentiator is traceability. Every step in the Chain of Work is logged and verifiable, giving teams full visibility into how the system reached its output. This enables real human-in-the-loop and over-the-loop control, not just a checkbox. For enterprises in regulated industries, that means decisions based on documented facts, full audit trails, and AI automation they can actually trust. #### When 99% Accuracy Isn’t Enough In a consumer chatbot, a hallucination is annoying. In enterprise operations, it costs money. A 1% error rate sounds small until you do the math at scale. A bank processing 50,000 trade finance documents a month at that rate means 500 documents with potentially wrong data. Wrong dates, made-up policy terms, mismatched compliance references. In manufacturing, a “guessed” delivery date can stop a production line. In insurance, a hallucinated policy clause can expose an underwriter to liability they never signed up for. Maisa’s architecture is built around this reality. The system cross-references source documents at every step, flags inconsistencies before they hit the output, and keeps a full audit trail, possible thanks to the KPU. Every output traces back to the data that produced it. #### Frequently Asked Questions ##### How often does AI hallucinate? It depends on the model, the task complexity, and the guardrails in place. Simpler factual queries tend to produce fewer hallucinations than open-ended or niche topics. The more important question isn’t frequency. It’s whether you have a system in place to catch errors when they occur. ##### Do all AI tools hallucinate? The short answer is yes. Most generative AI tools have the potential to hallucinate. It mainly comes down to how the AI tool was trained and the data it has access to. ##### Will AI hallucinations ever go away? Probably not, at least not as long as AI systems work by predicting text. The real question is whether you can catch the errors before they cause damage. That’s the approach Maisa takes. Instead of trying to build a model that never gets it wrong, the system verifies every step so mistakes don’t make it into the final output. ##### How can you tell if AI is hallucinating? The harder the hallucination is to spot, the more dangerous it is. Well-trained models produce highly confident outputs, which means the best defense is independent verification. Check specific details: dates, names, citations, regulatory references, and statistics. If the model cites a source, confirm that source exists. ##### What are some examples of AI hallucinations? The most frequent categories include ghost citations (fabricated academic or legal sources), policy fabrications (non-existent terms, discounts, or clauses), and logic failures (incorrect medical, legal, or mathematical reasoning presented with full confidence). ## Digital Workers URL: https://maisa.ai/landings/digital-workers AI Agents built for business processes that adapt, collaborate, and operate with full accountability and transparency. AI Glossary #### What are Digital Workers? - Digital Workers are AI Agents built to execute tasks within business processes. They adapt to changing conditions, make decisions, and interact with both teams and tools to complete work. - Unlike traditional automation, Digital Workers operate with clear goals and follow the logic of the process. They handle complex tasks that require reasoning, communication and judgement while keeping every step traceable. - They are built for operational environments where work changes often and where traditional automation falls short. #### Why Do Businesses Need Them? - Most business processes mix predictable steps with those that demand interpretation, validation, or decision-making. AI Digital Workers can automate these processes while preserving control, adaptability, and auditability. - Traditional automation tools follow fixed rules, failing when work demands judgment and flexibility. Conventional AI solutions could process complex information but they act as “black boxes”, making it difficult for organizations to trust or audit their outputs. - Digital Workers solve both problems. They adapt to real operating conditions, follow business rules, and leave a transparent trail of their decisions. This allows companies to automate work that was previously too variable, too risky, or too costly to automate. ##### Key Capabilities of Maisa Automated Workers Our Digital Workers are built for enterprise processes and the operational demands that come with them: traceability, workflow awareness, adaptability, and effective collaboration. ##### Traceability and accountability Every action and decision is captured in a detailed execution log, what we call “Chain of Work”. Teams can review how data was interpreted, which rules were applied, and why a decision was reached. This level of visibility makes the automated work fully auditable. ##### Works with teams and stakeholders Digital Workers operate within business workflows, knowing when to seek approval, escalate issues, or notify the right people. They reinforce business rules rather than working around them. ##### Adaptive problem-solving Instead of following static scripts, Digital Workers pursue goals. When conditions change or information is missing, they evaluate alternatives and select the path that gets the task completed correctly. ##### Understand business data Most processes rely on information scattered across systems, emails, documents, and databases. Digital Workers can interpret both structured and unstructured data and use it to drive decisions or trigger further actions. ##### How setting up a Digital Worker looks like Setting up a Digital Worker is simple and doesn’t require technical expertise. Domain experts, those who know the process best, can configure it directly. ##### Describe the work The user defines the goal of the task in natural language, and the Digital Worker proactively asks for the details it needs. At this stage, the user provides access to relevant data sources, tools, and integrations. ##### Testing & Refinement The Digital Worker operates in review mode, allowing teams to evaluate its decisions and refine workflows through feedback in natural language before implementation. ##### Deploy & Learn After validation, the Digital Worker begins handling tasks, improving over time based on operational feedback and real-world experience. #### How Digital Workers Operate? A Digital Worker isn’t a single model or script. It’s a coordinated system organized into three layers that mirror how people work: Perception, Reasoning, and Execution. ##### Perception Layer – The Input This layer collects and interprets incoming information: documents, emails, forms, system data, or manual inputs. The information is categorized and converted into structured representations the Worker can use. ##### Reasoning Layer – The Brain Here, data is analyzed and broken into actionable steps. Large language models and Maisa’s Knowledge Processing Unit (KPU) collaborate to map out the tasks required to reach the desired outcome. The KPU enforces rules, validates outputs, and keeps reasoning grounded in verifiable data, always backed by code. Making it hallucination-resistant. ##### Execution Layer – The Hands Guided by the decisions produced in the reasoning stage, the Digital Worker carries out actions across enterprise systems. This includes API calls, UI interactions, data entry, document generation, and cross-system workflows. A key difference between basic bots and Digital Workers is how they use tools. Digital Workers interact with business applications as a skilled employee would, consistently and without supervision. ##### Why Most AI Agents Fail, and why Maisa Succeeds AI automation is increasingly appealing, especially in regulated industries like financial services, insurance, and healthcare. But most AI agents fail because they lack transparency. Their reasoning is hidden, their decisions are difficult to audit, and errors go unnoticed. LLMs can generate convincing but incorrect outputs. In a business environment, this is unacceptable. You cannot deploy automation you cannot trust. Maisa avoids this by pairing LLMs with the KPU, which enforces structured, verifiable reasoning, backed by the determinism of code. Every decision leaves a trace: - What data was accessed - Which rules were applied - Why the decision was made Managers can inspect any step of the process when needed. Maisa Digital Workers behave like accountable digital employees whose work is verifiable, reviewable, and adjustable. #### Enterprise Security & Compliance Traceability is great only if it’s backed up by enterprise level security. This is why Maisa’s digital workers are built on the basis of most advanced security protocols for regulated environments. ##### Supports GDPR and EU AI Act compliance with transparent data handling and governance. ##### Data sovereignty Processing occurs inside the customer’s secure perimeter. ##### Enterprise-grade access controls and isolation. #### Maisa Digital Workers Features ##### Adaptability Digital Workers recover from unexpected situations, missing data, or workflow changes by selecting alternative paths rather than stopping the process. ##### Self-Learning Our digital workers learn directly from the workflow and do not rely on endless training cycles or complicated setup. Maisa learns by actually working, like us humans do. ##### Omnichannel Execution They work seamlessly with structured or unstructured data, documents, emails, and system interfaces, enabling automation across diverse channels. #### Use Cases ##### Trade Finance Problem: High-volume trade finance workflows depended on manual checks of contracts, invoices, and certificates, creating long turnaround times, compliance risk, and limited scalability. What the Digital Worker does: Reads and validates all trade documents (even low-quality scans), cross-checks terms and beneficiaries, identifies discrepancies, and produces audit-ready summaries, end to end. Business Impact: 6× capacity increase, 80% faster processing, and 93% automation accuracy across regions. ##### Auto-loan processing Problem: Loan applications required multi-document verification across languages and formats. Legacy OCR/ML tools delivered low accuracy, causing delays, manual review, and customer frustration. What the Digital Worker does: Classifies documents, extracts all required fields with context awareness, validates data against business rules, and produces consistent structured outputs for decisioning. Business Impact: 98% extraction accuracy, 3× faster processing, and an 85% reduction in manual review effort. ##### Power of Attorney Problem: POA verification relied entirely on manual interpretation of legal documents with inconsistent formats, slowing onboarding and creating compliance bottlenecks. What the Digital Worker does: Understands legal clauses, verifies authority, cross-checks identities, routes exceptions intelligently, and keeps each decision fully traceable for audit. Business Impact: 40,000 hours recovered annually, minutes instead of hours per verification, and €1.2M saved in the first region. #### AI Agents That Deliver Real Business Value Digital Workers combine intelligent decision-making with the structure required in real operations. They can understand objectives, follow business rules, and adapt to changing environments, making them reliable for processes that are impossible to figure out for traditional automation. They coordinate actions across systems, interpret both structured and unstructured data, and manage exceptions without losing traceability. This allows them to execute end-to-end workflows with consistency, accuracy, and full operational oversight. For businesses, the benefits are clear: faster cycle times, fewer errors, stronger compliance, and reduced operational load. Digital Workers extend the workforce with a dependable, scalable layer of automation that operates with the same clarity and accountability expected from human teams. # Blog ## Maisa x Santander: 6 Lessons from Scaling AI Inside a Large Global Bank URL: https://maisa.ai/blog/maisa-x-santander Exploring the organizational and operational challenges of deploying AI systems responsibly within regulated financial institutions. Insights May 26, 2026 Most discussions about artificial intelligence focus on potential, benchmarks, and model capabilities. However, the practical challenges of moving AI from experimental settings into daily operations at large regulated institutions like banks rarely receive attention. David Villalon, Cofounder & CEO of Maisa, and José Palacio, Chief Data and AI Officer and Head of Global AI Adoption at Santander, have spent the past year implementing AI within high-stakes banking operations. They shared their experiences at Revolution Banking 2026 in Madrid, revealing six critical lessons about enterprise AI deployment. #### Lesson #1: Scaling AI in Banking Is More Difficult Than Most Other Industries While impressive AI demonstrations create expectations that moving from proof of concept to production should be straightforward, banking presents unique obstacles. The financial services industry operates under extensive regulatory requirements, meaning every AI decision must be explainable, traceable, and thoroughly documented. "When you finally face the reality of scaling AI in production and do things that go beyond conversational agents, you realize it is not so pretty." Banks require systems that can demonstrate exactly how decisions are made with complete reproducibility, rather than accepting occasional errors as inevitable. This fundamental requirement reshapes how AI systems must be designed and deployed. #### Lesson #2: We Hold AI to a Higher Standard Than We Ever Held Humans Traditional banking processes account for human error through provisions and reserves. However, AI introduces a different problem: hallucination, where systems generate confident, well-structured answers that are factually incorrect. Banks appropriately demand stricter standards because incorrect AI decisions—whether affecting lending, compliance, or regulatory outcomes—can have irreversible consequences. Systems must be architected specifically to prevent and detect hallucinations rather than merely optimize for speed and capability. #### Lesson #3: ROI Has to Be Built in From Day One, Not Figured Out Later Many AI initiatives begin without clear business justification, leading to disappointing outcomes. Operating AI at scale involves substantial costs: infrastructure, token expenses, support teams, change management, and adoption efforts. At Santander, every use case meets three criteria: revenue increase, cost reduction, or risk mitigation. Projects unable to demonstrate clear connections to these metrics do not proceed. Platform flexibility—such as Maisa's model-agnostic approach—enables teams to define processes with capable models then switch to lighter, faster, more cost-effective alternatives during execution. #### Lesson #4: The Right Team for AI in a Bank Does Not Come From One Place Building an effective AI team requires diverse expertise that most organizations lack in one location: - Business domain experts who understand existing processes and decision-making logic - External perspectives from dynamic environments, challenging established assumptions - Technical specialists in data engineering, AI science, and systems deployment - Governance experts who can quantify and control AI-introduced risks When business users possess sufficient tools to build automation themselves—rather than depending entirely on technical teams—deployment speed and accuracy both improve significantly. #### Lesson #5: Data Quality Is Where New AI Use Cases Succeed or Fail Before They Even Start "Garbage in, garbage out." Before considering which model or technology to deploy, teams must evaluate available data: its quality, accessibility, and integration with target processes. In banking, data frequently resides across legacy systems never designed for interoperability. Preparing fragmented data for AI use requires substantial foundational work. Teams attempting to bypass this step and build use cases immediately typically must rebuild them later. #### Lesson #6: Three Must-Have Requirements for Every New AI Project in Enterprises José Palacio identified three essential organizational foundations: ##### Data Quality and Connectivity Clean, integrated data accessible across systems is fundamental. Most large banks maintain data silos that obstruct comprehensive AI deployment. Without addressing these structural issues, any AI system will face inherent limitations. ##### Governance, Risk, and Compliance Framework Regulators can halt AI deployments lacking proper risk controls and governance structures. Establishing these frameworks initially costs far less than remediation following deployment. ##### Transformation Mindset This represents the most challenging requirement. Effective AI implementation involves questioning whether existing processes are optimal, not simply automating them faster. Leadership must encourage this critical evaluation. These demands are organizational rather than technological, requiring genuine commitment and capable teams. ##### Conclusion Maisa specializes in regulated industries with high-stakes processes requiring complete transparency and accountability. The Santander partnership demonstrates what successful transition from pilot programs to production banking operations requires. Organizations serious about enterprise AI deployment should examine practical case studies rather than relying solely on technological capabilities. ## Maisa Named as a Front Runner in Gartner’s “Emerging Tech: AI Vendor Race – Startups to Watch in Agentic AI” Report URL: https://maisa.ai/blog/startups-to-watch-in-agentic-ai-gartner Maisa recognized as a front runner among agentic AI startups by Gartner's latest vendor evaluation report. News Mar 30, 2026 #### Recognising the Next Wave of Enterprise AI Gartner's research evaluates 129 global startups building in the rapidly evolving agentic AI space, identifying those demonstrating strong market traction, differentiated technology, and the ability to deliver real enterprise value. Within this landscape, Gartner has identified an exclusive group of 4 global companies as "Set Diamonds," which are front runners in the space. These are organizations whose solutions are already proving their value in real-world environments and are beginning to define the next generation of enterprise software. Maisa is proud to be included in this category. Maisa enables business users in regulated industries such as banking to deploy prebuilt digital workers or build custom ones in natural language. Their proprietary Knowledge Processing Unit acts as an enterprise-grade OS on top of any LLM, turning predictions into deterministic code execution with full chain-of-work traceability, auditable and regulator-ready from day one. These digital employees are gaining rapid market momentum because their value is tied to outcomes, allowing business leaders to clearly identify use cases and calculate a direct ROI. #### The End of Single-Task Agents For years, enterprise automation meant isolated tasks, a bot that processes an invoice, a script that moves data from one system to another. Useful, but limited. Gartner's latest report signals that enterprises are now demanding more: complex, end-to-end process execution where AI doesn't just assist, but takes ownership of entire business workflows with accountability and reliability. "AI employees" are also mentioned as autonomous agents with clearly defined roles and domain-specific capabilities, built to deliver measurable, accountable outputs across the full length of a process. One of the biggest barriers to full process automation has always been fragmentation. Enterprises run on dozens of disconnected systems including ERPs, CRMs, legacy platforms, and modern SaaS tools, and the inability to connect them end-to-end has kept automation confined to isolated pockets of the business. Maisa's enterprise-grade OS connects to any modern application or legacy system, with or without APIs, removing the connectivity constraints that have historically kept automation fragmented and limited in scope. In production environments, this means Digital Workers operating across complete, end-to-end workflows rather than isolated steps, with the accountability and reliability that enterprise teams in legal, compliance, security, and IT require. With Maisa Studio, business teams can onboard Digital Workers by simply describing the job that needs to be done. No process mining required. #### Real Business Impact Happens in Production For many enterprises, AI has lived in a pilot phase for longer than anyone anticipated. The technology showed promise, but translating that promise into measurable, auditable business outcomes proved harder than expected. Gartner's report identifies traceable outputs, deterministic execution, and auditability by design as the defining characteristics of production-ready AI systems. These are the capabilities that allow organizations to deploy AI at enterprise scale with confidence and accountability. Maisa is already running in production in some of the most demanding regulatory environments, including banking and financial services. What makes this possible is the ability to meet the full spectrum of enterprise requirements, from legal and compliance to security and IT governance, without compromising on performance or flexibility. In practice, this means Digital Workers processing loan documentation with full audit trails, automating compliance reporting across multiple jurisdictions, and handling sensitive customer operations under strict data governance frameworks. The deployments are live, the processes are real, and the outcomes are being measured against business results. #### A Market Defining Moment As agentic AI adoption accelerates, the vendor landscape has grown increasingly crowded, making it harder for enterprise buyers to identify systems that are truly ready for production at scale. Gartner's report provides that clarity. For organizations evaluating agentic AI vendors, particularly in regulated industries, it outlines the criteria that define production-ready systems: traceable outputs, deterministic execution, auditability by design, and measurable ROI tied to specific business outcomes. These are the baseline requirements that enterprises in banking, financial services, and other regulated sectors are already demanding from their deployments today. Maisa has been operating against these criteria in live production environments, and Gartner's recognition reflects that's where the broader market is now heading. ##### Frequently Asked Questions Why did Gartner name Maisa a Front Runner in agentic AI? Gartner evaluated 129 global agentic AI startups and named only 4 as 'Set Diamonds.' Maisa was selected for its KPU – which turns LLM predictions into deterministic code execution with full chain-of-work traceability – its rapid market momentum in banking, and its ability to deploy prebuilt or custom digital workers using natural language. What does Gartner say makes agentic AI production-ready? Gartner's 'Emerging Tech: AI Vendor Race' report identifies three criteria for production-ready agentic AI: traceable outputs (every decision linked to evidence), deterministic execution (consistent, reproducible results), and auditability by design (not bolted on after the fact). These are the baseline requirements for enterprise deployment in regulated industries. What is the end-to-end process automation challenge Gartner identifies? Gartner notes that enterprises run on dozens of disconnected systems (ERPs, CRMs, legacy platforms, SaaS), and the inability to connect them end-to-end has kept automation confined to isolated pockets. Maisa's enterprise-grade OS connects to any modern application or legacy system – with or without APIs – removing the connectivity constraints that fragment enterprise automation. ## Maisa Wins the Rising Star Award at Deloitte’s Technology Fast 50 Spain URL: https://maisa.ai/blog/maisa-wins-deloitte-rising-star-award Maisa receives Rising Star recognition from Deloitte for sustained growth and innovation in enterprise AI technology. News Feb 13, 2026 #### Recognizing High Growth Technology Leaders The Deloitte Technology Fast 50 Spain Programme identifies and recognizes the fastest growing technology companies within Spain's startup and scaleup ecosystem. The ranking distinguishes organizations that have achieved sustained and significant revenue growth over the past four years, driven by innovation, technology, and scalable business models. Beyond recognition, the Fast 50 Programme serves as a growth platform for participating companies. It facilitates connections with investors, corporations, institutions, and other key ecosystem players, while providing access to specialized expertise, media visibility, and valuable networking opportunities. The programme has become a benchmark within Spain's entrepreneurial and technology landscape. #### A Milestone for Market Trust In today's market, bold claims around AI agents are everywhere. New platforms launch daily, promising autonomy, transformation, and disruption. Real trust in enterprise technology, however, is earned through execution, customer impact, and independent validation. Winning the Rising Star category represents recognition from the market itself. Leadership cannot be declared but must be acknowledged by credible third-party institutions that evaluate performance, growth, and scalability with rigor. Deloitte's recognition signals that Maisa distinguishes itself in a crowded AI landscape, validating both growth and the strength of its business model. #### Building the Future of Enterprise AI Execution As organizations move beyond experimentation into production-grade AI adoption, the need for traceable, scalable, and reliable execution systems becomes critical. Maisa enables enterprises to execute complete processes, maintain full oversight, and scale with confidence. The Deloitte award reinforces the conviction that enterprise AI must be accountable, auditable, and designed for long-term operational value rather than impressive demonstrations alone. ## How Maisa Digital Workers Actually Work? URL: https://maisa.ai/blog/how-maisa-digital-workers-actually-work Maisa Digital Workers generate case-specific code at runtime to handle process exceptions and compliance requirements without pre-built workflow paths. Digital Workers Jan 12, 2026 Maisa Digital Workers take a different approach to traditional automation. Instead of encoding every possible path in advance, they generate and execute code at runtime, adapting to each case as it unfolds. Most enterprise automation covers the core process well enough. The exceptions are the problem. Judgment calls, documents in unexpected formats, fields that moved since last month, data that contradicts itself across sources. Workflow tools handle predictable sequences, but regulated industries rarely offer predictable sequences. Banking, insurance, trade finance: these run on variance. Digital Workers handle that variance without requiring months of workflow engineering upfront. #### Onboarding a Digital Worker Setup works like onboarding a new team member. You describe the process in plain language, explaining what steps are taken manually, what counts as a complete application, what policies apply, and what acceptance criteria look like. The Digital Worker converts this into executable instructions. You attach the knowledge needed: underwriting guidelines, compliance requirements, internal documentation. You connect the systems: your CRM, document storage, credit bureau APIs, core banking. The Digital Worker then gets access to the KPU, the secure environment where it generates and runs code. #### A Case Arrives When a loan application comes in and routes to the worker, it reads the submission and determines what exists, what's missing, and what to do first. This might involve extracting income data from uploaded pay stubs or calling an external API for a credit check. The worker writes specific code for that action—not generic code, but code tailored to this particular submission. The code runs inside the KPU. The output gets checked against the rules defined during setup. If the pay stub shows income of €45,000 but the application form says €52,000, that discrepancy gets flagged and handled according to your policies. The verified result feeds into the next step. This continues until the process completes. Clean applications resolve fast. Messy ones with missing documents, inconsistent data, or unusual circumstances take more steps as the worker addresses each problem. The output is a decision with full supporting evidence, where every number ties back to a document and every check is recorded. #### The Chain of Work The complete record of a case is called the Chain of Work. Each entry shows what the task was, what tools and sources it used, what it returned, what validations happened, and whether they passed. Inputs connect to outputs through explicit steps anyone can follow. This provides something most AI systems cannot: reproducibility. Same input, same configuration, same output. When an auditor asks why an application was declined, you can show them the exact evidence and the exact rules that produced that decision. No reconstruction. No guesswork. While many systems try to log what they do, the Chain of Work makes the reasoning reproducible rather than just falsely recorded. #### Running at Scale Digital Workers are not fixed programs. Each application gets its own chain of work. Three applications processed simultaneously might look completely different, with different documents, different exceptions, and different validation paths, while all following the same policies defined upfront. This is how you scale without drowning in workflow maintenance. New edge cases get handled at runtime. You do not need someone to anticipate them during design and build another branch. Teams often spend months engineering corner cases into workflow graphs before they can go live. Digital Workers skip that phase by adapting dynamically. #### Where Digital Workers Run Digital Workers run inside your infrastructure with your security controls. The KPU operates within boundaries you set. Data does not leave your environment unless you configure external calls. They plug into existing systems through standard connectors. No rip and replace. You keep your core banking platform, your document management, your CRM. Digital Workers sit on top and handle the process layer. #### Who This Is For If your automation challenges are mostly stable integrations with predictable inputs, you probably do not need this. Standard workflow tools handle that fine. If your processes involve documents, exceptions, judgment calls, and compliance requirements like loan origination, claims processing, KYC, or trade finance, Digital Workers make the difference. They handle variance without requiring you to model every possible path in advance. #### Frequently Asked Questions ##### How do Maisa Digital Workers work? Maisa Digital Workers "generate and execute case-specific code at runtime – not following pre-built workflow paths." When a task arrives, the worker reads the submission, determines what's needed, writes specific code, runs it in the KPU, validates outputs against your rules, and adapts step by step until the process completes. ##### How do you set up a Maisa Digital Worker? Setup involves describing the process in plain language, defining policies and acceptance criteria, attaching relevant knowledge such as guidelines and compliance rules, and connecting your systems via standard integrations. No custom development is required. The Digital Worker converts this into executable instructions and begins processing immediately. ##### What is the Chain of Work in Maisa? The Chain of Work is Maisa's complete audit trail for every automated case. It records each task, tools and data sources used, outputs produced, and validations performed—linking inputs to outputs through explicit, reproducible steps. Auditors can see exactly what evidence and rules produced each decision, without reconstruction or guesswork. ##### Can Maisa Digital Workers handle exceptions and edge cases? Yes. Because Digital Workers generate new code at runtime for each case, they handle novel exceptions without pre-built branches or manual intervention. Complex cases with missing documents or inconsistent data are processed through additional adaptive steps, with edge cases resolved dynamically rather than anticipated upfront. ##### Who should use Maisa Digital Workers? Maisa is ideal for processes involving documents, exceptions, judgment calls, and compliance requirements—such as loan origination, claims processing, KYC, and trade finance. If your automation challenges involve mostly stable integrations with predictable inputs, standard workflow tools may be sufficient. ## Why Maisa Is Not an n8n-like tool or RPA+ AI URL: https://maisa.ai/blog/why-maisa-is-not-an-n8n-like-tool-or-rpa-ai Maisa's runtime process formation and execution-backed verification differ fundamentally from n8n's design-time workflow orchestration model. Digital Workers Jan 7, 2026 #### A different model for enterprise automation In almost every enterprise conversation, sales teams hear the same comparison: "So you are like n8n?" This comparison is reasonable. Both platforms automate work across systems, call tools, run code, and connect to operational data sources and SaaS systems. However, when automation becomes business critical, enterprise buyers prioritize different system properties than workflow builder interfaces. They care about verifiable outcomes, resistance to hallucinations, traceability that holds up in audits, governance and security controls that scale, faster iteration without workflow sprawl, and ownership that does not collapse into a small group of specialists. Both approaches use code to accomplish work, but the timing of code formation, process definition, and proof systems differ significantly. These choices determine cost, risk, and speed moving from prototypes into production. #### A focused comparison on what enterprise buyers actually care about ##### The real distinction is the unit of automation n8n excels at orchestration. Users assemble workflow graphs at design time using nodes, conditions, retries, and integrations. Reliability depends on how thoroughly the process was modeled and how well exceptions were handled inside the graph. This works well when processes are stable and inputs are predictable. Maisa operates from a different premise. In Maisa Studio, users define work in natural language at the level of objective, constraints, policies, and acceptance criteria. The KPU then executes the work. The critical distinction is when and how the workflow exists. With Maisa, the process is formed at runtime. The KPU generates executable steps, runs them in a controlled environment, observes the results, validates them against rules and invariants, and then decides the next step. The system does not follow a fixed blueprint anticipated in advance. Instead, it assembles and verifies steps as each transaction unfolds, under defined governance. If processes resemble stable integration diagrams, workflow graphs provide a natural fit. If processes involve documents, exceptions, inconsistent formats, and edge cases, objective-driven execution changes what becomes feasible. ##### How you define the process determines who can own it In workflow graph models, the process is defined as structure. Nodes represent operations, conditions represent decisions, branches represent exceptions, and retries represent resilience. In enterprise settings, this definition has consequences. When processes are encoded as detailed graphs, typically only people who understand the graph as a system can safely modify it. Over time, ownership concentrates because the workflow becomes both business logic and production software simultaneously. In Maisa, the process is defined as intent plus constraints. Users express what success looks like, what policies apply, and what acceptance criteria must be satisfied. This moves the primary process definition closer to business language and makes it easier for compliance and security to review rules in forms they already use. The operational shift is straightforward. Instead of asking who can safely edit the workflow graph, teams ask who owns the policy and acceptance criteria. ##### When the workflow is formed changes how the system handles variance Enterprise processes rarely remain stable. Auto loans, onboarding, KYC, trade finance, claims, and back office operations contain significant variance. Templates change, documents arrive incomplete, fields move, and edge cases appear after deployment rather than before. In design-time workflow graphs, variance is handled by adding more structure. This typically means more branches, more conditional logic, more special cases, and more retries. Over time, graphs become decision trees. Every new exception requires a new path that must be maintained and tested. Change becomes expensive because policy updates can touch multiple branches and regressions become difficult to reason about. Maisa is designed around runtime formation. The KPU generates steps as it executes. It observes what happens and adapts to the specific transaction. When encountering a new corner case, it can generate the code needed to handle that case in the moment rather than failing simply because the path was not modeled in advance. When the same conditions repeat, the system can reuse proven paths rather than regenerating unnecessarily. The practical result is that you do not need months of corner case engineering just to keep the system functional as variance grows. ##### Hallucination resistance is not a prompt problem, it is a system property When enterprise buyers discuss hallucinations, they describe a system producing plausible values unsupported by underlying documents or data, with those values reaching the output. In many workflow-plus-AI setups, typical responses involve improving prompts, enforcing structured output, adding retries, and wrapping model calls with validation. These techniques help, but they do not eliminate the long tail. In production, the long tail is where risk lives. A second complicating issue involves confidence scores used to decide whether to proceed, retry, or escalate. Confidence does not equal correctness. Models can be highly confident about answers unsupported by evidence. Confidence can create the appearance of control without guaranteeing reliability. Maisa addresses hallucination risk by enforcing an execution-backed contract. The system is required to justify outputs through executed steps and validations. If evidence is missing, it should not invent it. If totals do not reconcile, it should not approximate. If a policy forbids inference, that rule is enforced in runtime behavior, not merely suggested in a prompt. This is what buyers want when requesting hallucination resistance: a system that fails closed when proof is missing. ##### Traceability that survives audits requires more than logs Most workflow platforms can show that a node ran, an API call was made, and an output was produced. However, once generative steps are added, logs can end up recording what the model returned rather than the evidence supporting the result. For enterprise buyers, traceability extends beyond debugging to governance and auditability. When problems occur, teams must answer questions like: - What evidence supported this value? - Which rule allowed this decision? - What validations passed? - What checks failed and how was the exception handled? Maisa is designed so the evidence trail is intrinsic. The KPU executes steps and validations in a controlled runtime. Outputs are backed by those executed actions. The trace becomes a receipt of what was done, what was checked, and what evidence was produced. In n8n-like systems, traceability depends on how workflows are built. Teams can add validation and evidence capture, create robust audit logs, and establish strong governance. However, this is not inherent to the model. It is something engineered per workflow, meaning quality varies across workflows, teams, and time. This difference becomes visible during incident reviews and compliance reviews: reconstructing what happened from logs versus having the system produce a clear evidence trail as part of execution. ##### Complexity level for users determines whether adoption scales n8n is described as low code. In practice, as soon as a workflow matters, teams add expressions, error handling, branching logic, and sometimes Code nodes. This is normal and often necessary for production robustness. The consequence is that people who can safely build and change workflows are relatively technical. This is not inherently problematic. It represents a strong fit for IT-owned integration orchestration. It becomes a constraint when the process lives with operations teams and changes frequently. In those environments, each new exception becomes a ticket and each policy update becomes a workflow edit. Maisa is designed to be no code from the user perspective. Operations teams define objectives, constraints, and acceptance criteria in natural language. Compliance can review policies directly. IT and engineering focus on integrations and runtime boundaries rather than continuously rewriting business logic. The outcome is not just ease of use. It is scalable ownership and predictable cost. The people accountable for outcomes can participate directly without needing to become workflow engineers. #### When to use which model If work consists mostly of stable integrations with predictable steps, and correctness requirements are satisfied by deterministic nodes plus straightforward validations, n8n-like workflows can be an excellent fit. n8n 2.0 strengthens this case with improvements aimed at production execution, safer change management, and a more secure-by-default posture. If work involves high variance, documents, exceptions, and high stakes, the runtime execution model matters far more than the workflow builder interface. This describes the world of auto loans, trade finance, onboarding, and back office operations where edge cases are normal and silent error costs are high. That is where Maisa Studio and the KPU are designed to win. Not by adding AI to workflows, but by changing what automation means when correctness, governance, and adaptability are non-negotiable. #### Why this matters to the business These differences map directly to business outcomes. - When outputs are grounded in execution and verified through checks, silent error risk drops. - When systems adapt at runtime rather than forcing pre-modeling of every corner case, time to value improves and expansions do not become months of workflow engineering. - When ownership is expressed as intent and policy rather than graph complexity, iteration speeds up without losing control. - When traceability is an evidence trail, audits and incident reviews become simpler and faster. The n8n comparison appears frequently because n8n represents a strong baseline for orchestration and is improving as a production platform. Maisa is built for cases where orchestration alone is insufficient and trusted execution, verifiable outcomes, work that adapts to messy reality, and scalable governance models are required. #### Frequently Asked Questions ##### What is the difference between Maisa and n8n? n8n builds workflow graphs at design time – users define nodes, conditions, and branches upfront. Maisa forms its process at runtime: users define objectives and policies in natural language, and the KPU generates executable steps dynamically per transaction. Maisa adapts to exceptions without pre-modeled branches; n8n requires engineering every possible path in advance. ##### Why does runtime process formation matter for enterprise automation? Enterprise processes involve constant variance – documents change format, edge cases appear post-deployment, policies update. Design-time tools require new branches for every exception. Maisa's runtime formation handles novel cases dynamically, eliminating the months of corner-case engineering typically required before go-live and reducing ongoing maintenance costs. ##### How does Maisa handle hallucinations differently from AI workflow tools? Maisa enforces execution-backed verification: every output must be justified by executed steps. If evidence is missing, the system fails closed. Most workflow-plus-AI setups rely on prompting, retries, and confidence scores – techniques that improve reliability but do not eliminate the risk tail. Maisa treats hallucination resistance as a system property, not a prompt engineering challenge. ##### Can business teams own Maisa processes without developer support? Yes. Maisa processes are defined as objectives, policies, and acceptance criteria in natural language – not code or workflow graphs. Compliance teams can review rules in the language they already use; operations teams update processes directly. Ownership stays with the people accountable for outcomes, not specialist engineers. ## Agentic AI Cost Control: The Enterprise Leader’s Guide to Avoiding “Surprise Opex” at Scale URL: https://maisa.ai/blog/agentic-ai-cost-control Enterprises can control agentic AI costs by measuring usage at the workflow level and implementing governance before scaling adoption. AI in the Enterprise Jan 7, 2026 Agentic AI is transitioning from pilots into production environments, fundamentally changing the cost structure. What begins as a manageable expense for model usage expands into broader operational costs including orchestration runtime, tool infrastructure, scaling overhead, and observability. Many of these expenses remain hidden during early demonstrations because the drivers are architectural in nature. Without cost control designed from the start, teams typically retrofit safeguards while the business already depends on the system. Cost overruns in real enterprises typically emerge across two concurrent tracks: - Pilot to Production: As systems become reliable for real operations, costs expand beyond tokens into orchestration runtime, tools, observability, reliability engineering, and scaling overhead. - Democratization without guardrails: As access spreads across teams, usage scales faster than governance. Employees use general AI tools for quick wins without cost estimation, budgets, or limits, creating outsized spend from small actions. #### Why agentic AI costs spiral when you don't plan early The most common pattern is that organizations scale capability before scaling control. Early systems optimize for outputs and autonomy, adding more tools, workflow steps, and agents without measuring what happens per run and per use case. ##### Democratization turns cost into a scaling problem, not a procurement problem When AI tools become widely available, spend scales with behavior rather than architecture alone. People treat AI as a silver bullet, iterating through trial and error and pushing larger inputs without understanding consequences. A single action can trigger significant cost increases. "Uploading a one hundred fifty page file into a general purpose assistant and repeatedly prompting over it can trigger large token usage across many attempts, especially when multiple people repeat the same pattern." Without cost estimation, budgets, and guardrails, organizations discover unit economics only after adoption has already spread. ##### Latency and spend rise together As workflows become more complex, prompts grow, tool calls increase, and retries become more common. Execution time and infrastructure and token usage rise together. Without intentionally decoupling performance from cost, both deteriorate simultaneously. Open source and frontier models still hallucinate, especially under tool use and long context workflows. Enterprises respond by adding verification steps, fallback models, larger prompts, and retrieval. These controls improve reliability but also increase tokens, tool calls, and end-to-end runtime. Without measurement and governance, cost rises as a byproduct of making the system safe enough to depend on. ##### Over provisioning becomes the default When demand spikes are unpredictable and system metrics are incomplete, teams provision for the worst case. This is rational when they cannot confidently autoscale using real workload signals. ##### Multi agent designs create hidden overhead Multi-agent systems can be appropriate for some problems but often introduce additional coordination steps, repeated context sharing, duplicated tool calls, and new failure modes that increase retries. This overhead becomes a permanent tax unless measured and governed. #### The real cost model of agentic AI beyond tokens For practical leadership decisions, costs fall into three categories: ##### Model cost This includes input and output tokens, prompt growth from tool usage, reruns caused by uncertainty, and the cost of defaulting to a larger model than necessary. ##### Infrastructure cost This includes compute runtime for orchestration and tool execution, concurrency and queueing effects, storage and state systems such as caches and vector stores, and observability pipelines. ##### Complexity cost This includes engineering time spent diagnosing and tuning behavior, operational risk from unpredictable runs, and the inability to forecast spend per team or business unit. Cost control means making each category measurable, attributable, and tunable. ##### Where small model strategies surprise teams Some frontier labs propose an enterprise stack built around small language models because the per-call cost appears dramatically cheaper. The risk is this framing focuses on unit price rather than end-to-end system economics. In real enterprise workflows, a single small model often proves insufficient. Teams end up using several specialized small models plus routing, orchestration, and verification to reach acceptable quality. This changes the cost profile across every category. This can increase model cost through more retries, longer prompts to compensate for weaker capability, and more frequent fallbacks to larger models. It can increase infrastructure cost through orchestration runtime, additional tool calls, state handling, and observability for a larger number of steps. It can increase complexity cost because multi-model systems have more failure modes, more tuning cycles, and higher operational risk. Fine tuning increases the downside further. Benchmarks show that after tuning on company data, models can regress on general capabilities due to catastrophic forgetting. When that happens, organizations pay twice: first for the tuning effort, then for the extra routing, fallbacks, and guardrails needed to recover reliability. The right takeaway is not that small models are bad. Small model strategies are only cost effective when evaluated as complete systems with measured quality and measured total cost per business outcome. #### Technical choices that strongly affect cost (and how to control them) ##### Capacity, scalability, and granularity are foundational Macro metrics like total requests or monthly spend are insufficient for agentic systems. You need granular measurements that tie cost to outcomes. Important signals include cost per workflow and case type, tokens per step, tool latency distributions, retry and fallback rates, and end-to-end execution time per workflow path. Once you have this visibility, teams can redesign the few expensive steps instead of shrinking the whole system. ##### Real time metrics enable dynamic infrastructure right sizing Enterprises pay for peaks and the safety margin around peaks. With real-time workload telemetry, you can scale orchestration and tool execution capacity with actual signals such as queue depth, concurrency, throughput, and per-tenant demand spikes. This demonstrates that agentic cost control is not only an LLM selection problem but a systems engineering problem. ##### Execution time estimation changes how you architect workflows Different use cases have very different runtime profiles. A simple routing case might complete in seconds. A deep analysis plus tool chain case might take minutes. Estimating execution time and cost per use case enables realistic SLAs, asynchronous routing of heavy jobs, budget and circuit breaker implementation, and different strategies for different workload classes. ##### Single agent versus multi agent should be a data driven decision Multi-agent can help with decomposition, verification for high-stakes tasks, and parallel exploration. It can also increase spend through coordination overhead, repeated context exchange, and duplicated tool usage. A cost-controlled environment treats this as a measurable tradeoff. You compare total tokens, runtime, tool calls, and success rates for single-agent and multi-agent designs on the same workflow. You default to single-agent where sufficient and upgrade selectively when outcome improvement justifies added cost. ##### Small models are not the goal best sufficient model is A common misconception is that using smaller models always reduces cost. The cheapest system is one that succeeds quickly. If a smaller model leads to more retries, weaker tool selection, longer prompts to compensate, or more human escalations, it can cost more end-to-end. The best approach uses the best sufficient model for each step. Smaller and faster models handle classification, extraction, and routing. Stronger models are reserved for complex reasoning steps. ##### Model agnostic capacity protects you from pricing and lock in risk When your architecture is tightly coupled to one model provider, you inherit pricing changes, policy shifts, and capacity constraints. A model-agnostic design decouples agent logic, routing policies, telemetry, and evaluation from the underlying provider. This enables cost-performance routing across providers and reduces replatforming risk when the model landscape changes. ##### Lightweight orchestration reduces infrastructure overhead The orchestration layer is not free. At scale, it can become a meaningful portion of runtime cost. Using a lightweight runtime and efficient implementation reduces memory footprint, improves concurrency, and lowers baseline compute per run. Choices like using Go or lighter development languages instead of heavier orchestration approaches like JavaScript have direct cost impact, especially when combined with strong metrics and autoscaling. #### A simple executive checklist (useful for governance reviews) If you're reviewing an agentic AI initiative, ask these questions: - Can we state the cost per completed business outcome? (not just monthly spend) - Do we know which workflow steps drive 80% of cost? - Do we have real-time metrics that let us scale infra down after peaks? - Is model choice dynamic per task/step, or fixed by default? - How many retries/fallbacks happen per workflow, and why? - Where are we using multi-agent, and what measurable uplift justifies it? - Can we switch model providers without re-platforming? - Is orchestration lightweight enough that it doesn't become a hidden tax? If several of these answers are no, costs usually trend up and confidence trends down. #### What cost control by design looks like in Maisa Studio Maisa Studio is built to keep agentic AI costs measurable, controllable, and predictable in production through instrumentation, elastic execution, model flexibility, and an execution architecture designed for efficiency at scale. ##### Built in cost and performance visibility Maisa Studio provides visibility into cost and performance at the workflow and step level. Teams can track model usage, tool execution time, retries, and end-to-end runtime per workflow path. This supports cost attribution by use case and business unit, plus fast identification of the few steps that drive most spend. ##### Elastic infrastructure aligned to real usage Maisa Studio aligns infrastructure capacity to actual demand signals such as concurrency, queue depth, and tool execution load. This supports stable performance during spikes and efficient utilization across normal usage patterns, with capacity right-sized to the workload profile. ##### Code backed execution that minimizes LLM calls Maisa uses a code-backed approach for workflow execution. Deterministic steps run in code and tools, and LLM calls are reserved for parts that truly require language understanding or flexible reasoning. This reduces unnecessary token usage, lowers latency, and improves predictability by keeping the model in the loop only when it adds measurable value. ##### Model agnostic capability for best sufficient model selection and operational continuity Maisa Studio is model agnostic, enabling teams to select the best sufficient model for each step based on cost, latency, and quality requirements. This supports avoiding vendor lock-in and maintaining operational continuity as model options evolve. Teams can update routing and policies while keeping workflow logic and operational controls consistent. ##### Digital workers designed to reduce multi agent overhead Maisa Studio is built around digital workers that execute end-to-end workflows with a single coordinated runtime. This reduces the coordination overhead often introduced by multi-agent frameworks, including repeated context sharing, duplicated tool calls, and additional failure paths. The result is a simpler execution model that supports cost predictability at scale. ##### Efficient orchestration for low platform overhead Maisa Studio uses efficient orchestration to minimize platform runtime overhead across steps and tool calls. This supports higher throughput, better concurrency efficiency, and lower baseline compute cost per run, especially in tool-heavy workflows. ##### Cloud agnostic deployment Maisa Studio supports cloud-agnostic deployment patterns, enabling enterprises to align with their infrastructure strategy, compliance requirements, and cost management preferences. This supports consistent operations across environments and flexibility as deployment needs evolve. #### Closing thought: Cost control is a capability, not a constraint. The goal is not to spend less at all costs. The goal is to spend predictably and intentionally while scaling adoption, so every workflow has clear unit economics and every increase in usage comes with visibility, guardrails, and the ability to tune. This approach brings the reliability, security, and performance enterprises expect from agentic AI systems, with cost control built in as an operational capability. It makes spend explainable at the workflow level, keeps infrastructure aligned to real demand, and ensures LLM usage stays focused on the few steps where it creates real value. The result is a platform that scales adoption without turning cost, governance, and operational risk into afterthoughts. Maisa Studio is designed around that principle. It provides workflow and step-level visibility, elastic infrastructure aligned to tool usage, and a code-backed execution approach that reduces LLM calls to the absolute necessary. Its model-agnostic design supports best sufficient model selection and avoids vendor lock-in as new models emerge, while digital workers and efficient orchestration keep coordination and platform overhead low across production-scale deployments. #### Frequently Asked Questions ##### Why do agentic AI costs spiral in enterprise deployments? Agentic AI costs expand across three areas: model cost (tokens, retries, oversized models), infrastructure cost (orchestration runtime, observability, scaling), and complexity cost (engineering time, unpredictable behavior). Costs spiral when organizations scale adoption before implementing governance – adding agents and tools without measuring what each run actually costs. ##### How can enterprises control agentic AI costs? Effective cost control requires measuring at the workflow and step level – not just total monthly spend. Key actions: use the best sufficient model per step, run deterministic steps in code to minimize LLM calls, implement real-time infra scaling on actual demand signals, default to single-agent architecture, and maintain model-agnostic design to avoid vendor lock-in. ##### Does using smaller AI models always reduce costs? No. Smaller models can cost more end-to-end if they produce more retries, require longer prompts to compensate for weaker capability, or trigger more escalations to larger models. The goal is the best sufficient model per step – smaller models for classification and extraction, stronger models reserved for complex reasoning – optimizing total cost per outcome. ##### How does Maisa Studio manage agentic AI costs? Maisa Studio provides workflow and step-level cost visibility, uses code-backed execution to minimize LLM calls to only where they add genuine value, supports model-agnostic routing for best-sufficient model selection, and employs elastic infrastructure that scales with real workload signals – keeping spend predictable and attributable by use case. ## What is business process automation (BPA)? URL: https://maisa.ai/blog/business-process-automation Streamlining operations through AI-powered automation of repetitive tasks and workflows. AI in the Enterprise Nov 11, 2025 Business process automation is the process of using AI technology to automate repetitive manual tasks and processes while maintaining efficiency, productivity, and accuracy. The goal of BPA is to streamline daily operations and allow businesses to operate smoothly. Automation of business processes can help companies in almost every field to achieve higher effectiveness and improve customer satisfaction and experience. #### Why is BPA important for the companies? Business process automation is not just the next trending topic. It is the new reality – adaptation to technology that will exist in the world, changing the old manual labor into new digital processes. Automation workflow helps in every step of the BPA by making it more transparent and accountable, reducing errors, improving turnaround times, and cutting costs. #### The 5 Types of Business Process Automation ##### Task Automation This is the most basic type of business process automation tool. It automates repetitive, straight forward tasks that are time consuming and require no decision making in the process. What is task automation? Task automation uses no-code platforms or software like scripts to perform repetitive rule-based tasks. Task automation fields and examples - Email management – You can use it to manage your emails by filtering, tagging and organizing them into specific folders. - Finance – automatic reporting, invoicing and overdue reminders. - Data management – moving information from a document to a database, or comparing information in different databases and flagging inconsistencies. - Customer service – instant replay to common questions or assign task based on certain criteria. Task automation advantages - Reduced errors – tasks are performed following strictly the process removing the manual labor and copy/pasting. - Increased productivity – tasks are processed 24/7, no coffee breaks or distractions. - Cost savings – improved performance and less errors mean lower corporate operational costs. - Improved job satisfaction – employees can finally focus on meaningful tasks and move away from repetitive work. ##### Workflow Automation Unlike task automation where individual tasks are executed, workflow automation deals with a chain of tasks to complete a business process based on predefined set of rules. What is workflow automation? The workflow automation makes sure a set of tasks are completed between human and software with minimal or no human interaction. Each task follows the next logical one in the process and only stops when external approval, consent or additional information is required. Workflow automation fields and examples - Sales – user fills out a "Contact us" form, the workflow instantly checks for some additional information that is needed (ex.: company size, location) and assigns the lead to the correct sales personnel and updates the follow-up tasks in the CRM. - Customer service – a task comes in, it is assigned to the right customer support representative but is not completed in the expected timeframe, a reminder is triggered and sent to a senior support specialist to ensure no complaint is left unsolved. - IT Help desk – a ticket comes in, based on the category (ex. "Hardware problem", "Software issue") it is assigned to the right specialist and returns an email to the sender with the expected time when the issue will be fixed. - Finance – an invoice is scanned, the system extracts the data from it and places the relevant information in the system (ex. vendor, due date, amount), then it matches the information with the internal purchasing system and sends it for approval to the relevant manager before executing or scheduling the payment. Workflow automation advantages - Improved efficiency and transparency – automated workflows provide a clear picture of task progress, what the next steps are, who is responsible, and where the bottlenecks and blockers are. This ensures smoother process and improved efficiency. - Compliance – in highly regulated industries workflow automations create a log of every approval by time and person, creating a clear and auditable trail. - Customer support – once a ticket is resolved, the system takes the final score of the interaction and if it meets certain criteria it sets up the customer to receive a bonus, discount, promotional materials or even be asked to write a review. ##### Robotic Process Automation (RPA) Robotic Process Automation uses robots to mimic human actions in a digital environment. Think of it as putting a digital "human" on a computer to process a high volume of tasks quickly and with great precision. What is Robotic Process Automation (RPA)? Robotic Process Automation (RPA) is a technology that allows robots or bots to do tasks in a digital environment through graphical user interface (GUI), mimicking a real person. The bots can perform complex tasks such as logging into systems and applications, moving and editing data into these systems, etc. Using robotic process automation frees up time for users to focus on more complex, out-of-the-box work. Robotic Process Automation fields and examples - Know Your Customer (KYC) – robots can automatically verify customer identity by checking them across multiple databases or asking them to perform a set of actions. - Order processing – organizing post-purchase process by getting the shipping information, printing out the labels, and sending back to the customer a tracking information email. - Policy management – managing policy renewals, information updates across multiple policy pages and sending reminders to the right people for verification once the process is done. Robotic Process Automation advantages - Cost reduction, productivity and speed – robots can perform these simple operations at a fraction of the cost, 24/7, and faster than humans do. - Legacy system integration – allow RPA to work with your legacy systems without the need to rework them with modern interfaces or APIs. - Compliance – robots follow strictly formatted paths and rules making sure of compliance while creating an auditable process for everyone. ##### Hyperautomation Hyperautomation is a business approach combining multiple automation technologies. Unlike traditional automation when one of the tasks is automated, in hyperautomation the goal is to automate the entire process. What is Hyperautomation? Hyperautomation is the strategy of combining different technologies like robotic process automation (RPA), machine learning and AI, to automate as many business processes as possible. Integrating these tools allows businesses to identify the tasks, design the needed workflow, deploy the bot that will do it, add an AI layer for the decision-making phase and monitor the entire operation. Hyperautomation fields and examples - Loan evaluation processing – end-to-end workflow that collects user data, runs all necessary checks, does KYC (Know Your Customer) validation to measure the risk and later makes a decision to approve or deny the loan in minutes. - Fraud detection – using established AI models and scanning millions of transactions at the same time to identify odd behavior and patterns to flag or block these transactions for further review. Hyperautomation advantages - Data-driven insights – ML and AI allow faster and more accurate decisions to be made while backing up everything with solid data. - Cost reduction – automating a larger portion of the work allows reduced operational cost. - Coverage – automates both simple and complex tasks across the entire organization. - Adaptability – adapts to the ever-changing company needs and external factors. ##### Agentic Automation Agentic automation combines AI and machine learning (ML) with simpler automation tools like RPA or workflow automation to achieve a goal. What is Agentic Automation? Agentic automation is the most complex form of intelligent automation. It uses AI to achieve goals rather than following rules. It understands the environment using large language models (LLMs), adapts to it, reasons to break complex tasks into smaller, more manageable steps, and starts moving towards the goal. It can make independent decisions to overcome unexpected problems and continue the process until it is finished. Agentic Automation fields and examples - Finance – processing various invoice formats and processing data to ensure all entries are valid. - Customer service – intelligent chatbot that not only provides information but also gives solutions to the customer, providing better customer satisfaction. - IT operations – monitor entire systems and instead of only sending alerts, attempts to reboot the system, run diagnostics or re-route traffic to keep system stability. - Retail – monitor competitor prices, dynamically adjust prices, monitor inventory and send alerts for restock or do it automatically based on a predefined plan. Agentic Automation advantages - Autonomy – learns and adapts to achieve the goal without human intervention. - Problem solving – complex problems are broken down into smaller tasks so they can be better understood and processed. - Personalized approach – adapts communication to match the tone or vibe of the customer and achieve better customer interaction. - Proactive actions – detecting and solving issues at one stage often cause problems at others, so AI agents proactively take measures to prevent it. #### Risk of implementing BPA in your organization There are some risks in the business process automation (BPA) process which need to be considered. Depending on the industry and the nature of the company they vary, and all need to be carefully estimated before blindly diving into BPA without planning everything ahead. ##### Technical Risks - Integration – connecting multiple systems together is often a challenge that brings unexpected issues. Sometimes the integrations are too complex or difficult to maintain, making the whole system fragile and unstable. - Data security – processing sensitive data in the automation process, whether health, financial or personal in nature, is another great risk if system security is not strong, bringing both legal and financial liability. - Data quality – feeding your BPA system bad or poorly formatted data can result in bad outcomes and move errors further down the digital automated process. - Updates. ##### Financial Risks - Cost – BPA integration can become costly, requiring time, people, software and post-launch maintenance. - Unclear ROI – it can be really difficult to measure "better quality data" and justify the need for automating the process or parts of it. ##### Operational Risks - Lack of flexibility – automated systems are created to follow certain logical steps, rules and workflows, which makes it very difficult for them to solve corner cases and atypical events. If the process is complex and has a lot of variations, this can turn into a bigger problem requiring constant human intervention. - New problems – solving a problem at one stage can often simply move it to another place. Imagine having a problem processing a high amount of data at a very early stage. Once this is solved, the problem moves to the next department or system that cannot handle the new volume anymore. - Downtime – creating a fully automated process is great but if the automation fails due to server outage, update problem or misconfiguration, the entire process may stop. ##### People Risks - Internal resistance – employees may resist the change, thinking they may lose their jobs, or avoid adopting the process and stick to the manual one. Employees may try to find ways to bypass the process or even prevent its integration in the first place. - Knowledge gap – a fully automated process used for a long time and expanded across the organization can lead to employees not knowing why and how certain things are done, causing knowledge to get lost over time. If the automation fails, manual labor cannot keep the workflow going even at a slower speed. #### BPA Advantages ##### Short term - Cost reduction – reduce operational cost by removing manual labor from repetitive, time-consuming tasks. - Speed and efficiency – automated systems work 24/7, which reduces processing time. - Accuracy – removing humans from boring tasks removes the errors where slight distraction can lead to bad data entry or bad information transfer. - Visibility – automating the process makes it entirely transparent and visible to management, showing where the problems are and what can be improved. ##### Long term - Further cost reduction – implementing the process initially will save some money, but optimizing it over time will gradually add more savings to the overall cost. - Scalability – allows the company to deal with higher workload without the need to hire more people. - Decision making – gathering large volumes of data and analyzing it can highlight missed opportunities or performance potential invisible in a manual process, helping management make more informed decisions. #### BPA fundamentals Business process automation (BPA) contains several basic building blocks to create a fully automated workflow. - Process mapping – knowing your processes is the first step in automating them; listing currently existing and used processes is the cornerstone of BPA. - Process orchestration – understanding the processes and making a decision about what process automation tools to use. - Robotic process automation (RPA) – the digital worker that gets things done working exactly like a real human in a digital environment. - Artificial intelligence (AI) and machine learning (ML) – the brain behind the hands, making decisions and adapting to the situation. - Monitoring – allowing humans to understand what happens behind the scenes, giving visibility and process data in real time. #### What is the difference between Business Process Automation (BPA) and Robotic Process Automation (RPA) This question often confuses people. Business process automation is much larger in scale and contains robotic process automation within it. While BPA is focused on the entire process, RPA is focused on a specific task. Think of it as BPA being the coach in a team, arranging players, deciding strategy and assigning each individual player a specific goal, while RPA is the player itself doing what the coach says. ##### Frequently Asked Questions Business process automation (BPA) is the use of technology to automate repetitive manual tasks and workflows, improving efficiency, accuracy, and cost-effectiveness. BPA encompasses five types: task automation, workflow automation, RPA (robotic process automation), hyperautomation, and agentic automation – each suited to different levels of process complexity. What is the difference between BPA and RPA? BPA (Business Process Automation) is the broader strategy for automating entire processes, while RPA (Robotic Process Automation) is one specific tool within BPA. Think of BPA as the coach designing the automation plan, and RPA as the player executing scripted tasks – copying data, filling forms, and interacting with digital systems. What are the main risks of business process automation? BPA risks fall into four categories: technical (integration complexity, data security, data quality), financial (high implementation costs, unclear ROI), operational (lack of flexibility for edge cases, new bottlenecks, downtime), and people risks (employee resistance, knowledge gaps if teams lose familiarity with manual processes). Agentic automation is the most advanced form of BPA. It uses AI and LLMs to achieve goals rather than follow predefined rules – understanding the environment, breaking complex tasks into manageable steps, making independent decisions, and handling unexpected problems until the goal is reached. Unlike RPA, it adapts dynamically without a fixed script. ## Agentic AI vs RPA: differences, similarities, and examples URL: https://maisa.ai/blog/agentic-ai-vs-rpa Agentic AI and RPA represent different automation philosophies, with AI handling complex adaptive workflows while RPA executes predefined tasks efficiently. AI in the Enterprise Oct 24, 2025 #### What is Agentic AI? Agentic AI refers to an autonomous AI system that operates independently to achieve complex, long-term goals without constant human involvement. It can understand its environment, reason, plan, act, and learn from its actions to become more efficient. ##### How it works - Goal intake: The AI collects information needed to understand the objective. - Reasoning: The AI analyzes data, plans steps, and chooses actions using large language models. - Tool & data access: Connects with external APIs, software, and databases. - Orchestration: Breaks down complex tasks into smaller ones, planning their sequence. - Action: Executes the task order across systems and tools. ##### Strengths - Manages end-to-end workflows - Adjusts constantly when processes shift or data changes - Scales to support complex, multi-step operations - Removes humans from repetitive manual work ##### Limitations - Errors in one step can propagate through the process - Needs clear guardrails and human oversight - Highly dependent on quality data - Requires transparency for tracking decisions #### What is RPA (Robotic Process Automation)? Robotic Process Automation uses software bots to follow predefined script-based paths and complete tasks within digital systems. RPA systems allow enterprises to reduce errors and speed up operations while maintaining quality. - Process mapping: Every click, action, rule, and sequence is programmed in advance by IT teams. - Execution: Bots replicate these steps across applications with speed, accuracy, and consistency. - Scope: Ideal for large-scale, well-organized, repetitive processes like data entry and form filling. - Maintenance: Any process or interface change requires script updates. - Reliable for high-volume, repetitive, structured tasks - Reduces human error and increases speed in routine processes - Widely adopted in enterprise operations like finance, HR, and back-office functions - Breaks when processes or systems change - Unable to process unstructured or ambiguous data - Dependent on IT teams for setup and maintenance - Limited scalability for complex or changing workflows #### What are the differences between Agentic AI vs RPA? The two approaches reflect different philosophies of automation. Agentic AI focuses on adaptive goal achievement, while RPA focuses on the strict execution of predefined steps. - Primary role: RPA executes predefined tasks; Agentic AI achieves goals autonomously. - Nature: RPA is scripted and rule-based; Agentic AI is reasoning-driven and adaptive. - Input needed: RPA requires step-by-step instructions; Agentic AI requires a clear goal or outcome. - Flexibility: RPA is brittle to change; Agentic AI adjusts dynamically. - Data: RPA handles structured data only; Agentic AI handles structured and unstructured data. - Maintenance: RPA is IT-heavy; Agentic AI is self-adjusting with guardrails. - Business value: RPA delivers efficiency in stable processes; Agentic AI delivers efficiency and scalability in complex processes. #### RPA and Agentic AI examples ##### Finance - Agentic AI: Interprets invoices in varied formats, checks for differences, updates the ERP, and contacts vendors if information is missing. - RPA: Uses predefined templates to copy invoice data into an ERP system following exact pre-set steps. ##### Customer Support - RPA: Routes tickets according to predefined rules such as keywords or categories. - Agentic AI: Reads entire customer messages, identifies intent, responds to straightforward requests, and escalates complex cases. ##### Operations - RPA: Transfers information between systems on a predefined schedule based on scripted rules. - Agentic AI: Monitors system data constantly, identifies discrepancies, updates records, and notifies stakeholders of unusual patterns. #### What are the similarities between Agentic AI vs RPA? - Both aim to improve business effectiveness by reducing or removing human labor - Both aim to achieve 100% accuracy in operations - Both integrate into existing enterprise environments - Both aim to reduce operational costs - Both are tools for business growth and streamlined operations #### What is Hyperautomation? Hyperautomation is when robotic process automation, AI, machine learning, and analytics fuse into one. RPA serves as the "muscle," working alongside agentic AI as the "brain" to streamline business processes and provide better customer interactions. Hyperautomation is about strategy and orchestration, meaning organizations choose what to automate with RPA and what to leave to agentic AI, chaining processes together for improved problem-solving workflows. #### What is Overautomation? Overautomation occurs when businesses automate tasks and processes that do not actually need AI-driven workflows. Overautomation often creates inefficiency and leads to poor customer experience. Overautomation can hurt customer experience by making customers feel detached and stuck dealing with machines. This may lead to poor customer satisfaction and loyalty, while systems become unstable, difficult to maintain, and hard to improve. ##### Factors to consider when choosing whether to use Agentic AI or RPA - Complexity of processes: RPA suits defined, highly repetitive processes. Use agentic AI for dynamic, decision-based processes. - Type of data: RPA works better with structured, stable data, while AI excels at natural language processing, image recognition, and unstructured data. - Cost: RPA is cheaper to integrate, but AI allows for greater expansion and flexibility despite higher expense. ##### Will AI agents replace RPA? In short, the answer is no. AI agents are still a long way from replacing RPA entirely. For some time, they will operate together, and RPA will continue at scale for non-decision-making processes because of its stability and auditability. As AI agents progress, they will gradually replace RPA in more complex processes requiring decisions. #### Facing the automation challenges Success with Agentic AI depends on addressing challenges such as reliability, oversight, and transparency. Organizations need to build systems designed to reason clearly, act consistently, and explain their decisions, enabling businesses to scale automation with confidence. RPA delivers fast, stable, rule-based processes, while Agentic AI opens the door to automating more complex and adaptive workflows. This marks a shift from following predefined scripts to achieving outcomes with autonomy. ##### Frequently Asked Questions What is the difference between agentic AI and RPA? RPA follows predefined, scripted steps for structured, repetitive tasks and breaks when processes or inputs change. Agentic AI uses reasoning and planning to achieve goals autonomously, handling unstructured data and dynamic exceptions. RPA excels at stable, high-volume processes; agentic AI excels at complex, judgment-intensive workflows requiring adaptability. Will agentic AI replace RPA? Not entirely in the near term. RPA will continue at scale for stable, non-decision-making processes due to its reliability and auditability. As agentic AI matures, it will gradually replace RPA in more complex decision-requiring workflows. Both will coexist in hyperautomation architectures. Hyperautomation is the strategy of combining RPA, AI, machine learning, and analytics into a unified automation approach. RPA serves as the ‘muscle’ executing defined steps, while agentic AI provides the ‘brain’ for reasoning and decisions. What are the strengths and limitations of RPA? Strengths: reliable for high-volume repetitive structured tasks, fast and accurate, widely adopted in finance, HR, and back-office. Limitations: breaks when processes or interfaces change, cannot handle unstructured or ambiguous data, requires IT-heavy maintenance, and is not scalable for complex or frequently changing workflows. ## Rising star of agentic AI delivering trustworthy ‘digital workers’ raises $25M from Creandum and Forgepoint URL: https://maisa.ai/blog/maisa-raises-25m-from-creandum-and-forgepoint Maisa secures $25M in funding to expand its AI agent platform enabling non-technical users to build trustworthy digital workers. News Aug 28, 2025 Maisa, a dual US/Europe-headquartered company, has announced a $25 million seed investment led by Creandum, with participation from Forgepoint Capital International through its European joint venture with Banco Santander. NFX and Village Global, who invested $5 million in the company's pre-seed round last December, also participated in this funding round. The company was founded in 2024 by CEO David Villalón, formerly Chief AI Officer at Clibrain, and CSO Manuel Romero, a top HuggingFace contributor with over 700 open-source models. The funding will support expansion across AI research, engineering, sales, and customer success in Europe and North America. ##### Maisa Studio Platform Launch Alongside the fundraise, Maisa unveiled Maisa Studio, an agentic process automation platform designed for "citizen developers"—business experts without IT backgrounds. The platform allows non-technical staff to deploy powerful, fully auditable AI digital workers using natural language commands. The platform is currently being piloted at global financial institutions, car manufacturers, and energy companies to automate complex, compliance-sensitive workflows. According to the company, digital workers learn on the job by doing, through a method Maisa calls "HALP" (human-augmented LLM processing). ##### Real-World Deployment Success Maisa has demonstrated operational success with major clients: - A global investment bank replaced manual media screening with Maisa digital workers that extract facts from news articles and assess reputational risk - A large financial services firm automated transaction reconciliation, filtering out 99% of false positives and achieving 10x productivity improvement per person with no engineering work required ##### Addressing Enterprise AI Challenges Recent industry data underscores the problems Maisa addresses. MIT's NANDA initiative found that 95% of generative AI pilots are failing and delivering little measurable impact on P&L. Additionally, IDC research shows 88% of AI pilots fail to reach production, and BCG found only 22% of companies have advanced AI beyond proof-of-concept stage. Maisa eliminates barriers by providing a model-agnostic system built on its proprietary Knowledge Processing Unit (KPU) reasoning engine. The platform records every action in what the company calls Chain-of-Work—a transparent, auditable record reducing hallucinations and ensuring traceable decision-making. ##### Leadership Perspectives "This investment is validation that AI in business must be built on trust. Our platform gives teams the power to use AI workers which not only perform with incredible accuracy and intelligence, but also explain and prove their logic." — David Villalón, co-founder and CEO "Maisa is solving one of the toughest challenges in AI: making it reliable and safe to use in mission-critical business operations." — Peter Specht, General Partner, Creandum "The company's early success demonstrates there's real market pull for AI done the right way and Studio will make it easy for anyone to use digital workers quickly." — Alberto Yepez, Forgepoint Capital International ##### Platform Integration and Recognition Studio integrates with over 450 third-party systems and connects to any documented API. The platform was recently named by Gartner alongside Google and Amazon as a leading AI agent vendor—marking the first time a Spanish startup appeared in one of Gartner's influential Hype Cycle reports. ## Maisa raises $25M seed investment URL: https://maisa.ai/blog/maisa-raises-25m-seed-investment Maisa secures $25M seed funding to scale accountable AI technology for enterprise automation. News Aug 28, 2025 #### The AI adoption reality Despite widespread enthusiasm for artificial intelligence, the adoption landscape remains challenging. Statistics reveal that 87% of enterprise AI projects never progress beyond the proof-of-concept phase, with only 4% ultimately delivering meaningful business value. Organizations frequently invest substantial resources into pilots that perform impressively in controlled demonstrations but fail when deployed in production environments where precision and accountability are critical. The core issue is that enterprises require AI systems they can verify and that their business teams can independently build and expand. They need complete transparency into decision-making processes and the capability to implement agentic automation without requiring deep technical expertise. Current AI solutions provide neither capability, leaving organizations stuck in experimental phases while routine work remains manual. #### Digital Workers in action Maisa's approach enables business experts to instruct Digital Workers through conversational language, similar to training a new employee. Users describe processes, outline decisions, and specify goals. The Digital Worker learns from this guidance combined with real-world feedback, creating automation that matches actual work practices. #### Fueling our mission The new capital will support expansion of engineering, sales, and customer support teams across Europe and North America while advancing research and development initiatives. The company aims to build infrastructure enabling enterprises to deploy Digital Workers at scale, with all technical improvements and team growth serving the mission of accountable AI for mission-critical operations. #### Our partners in accountable AI Maisa is solving one of the toughest challenges in AI: making it reliable and safe to use in mission-critical business operations. Peter Specht, General Partner, Creandum Maisa's platform is purpose-built for compliance-conscious industries like finance, where decisions must be traceable and outcomes consistent. Alberto Yepez, Forgepoint Capital International ## Maisa Studio on AWS Marketplace: removing barriers to AI adoption URL: https://maisa.ai/blog/maisa-studio-on-aws Maisa Studio launches on AWS Marketplace to simplify enterprise adoption of transparent, accountable AI agents. Jul 23, 2025 #### The AI Agents landscape in enterprise AI agents represent a shift from responsive assistants to autonomous systems that pursue goals independently. They decide how to complete tasks, use tools, and adapt to new information. However, enterprises encounter significant obstacles: AI agents produce hallucinations, make unexplainable decisions, and operate without transparency. When decision-making processes remain opaque, organizations cannot confidently deploy these systems for critical functions, leaving most businesses confined to basic chatbots. Digital Workers offer an alternative: AI agents purpose-built for business processes where every action remains traceable and every decision explainable. Through Maisa Studio on AWS Marketplace, enterprises gain access to AI with complete visibility into reasoning, steps taken, and tools used—designed to align with actual business operations. #### Removing the barriers to AI adoption Maisa joined AWS Marketplace as an early entrant in the new AI Agents category, making Digital Workers accessible through channels enterprises already trust. Numerous AI initiatives stall during procurement and security evaluations, preventing teams from accessing beneficial technology. AWS Marketplace eliminates these friction points: security and compliance reviews are already finalized, and procurement occurs through existing AWS agreements, avoiding new vendor processes. #### Making AI accountable This initiative advances the core objective of establishing AI accountability for business-critical functions. Through AWS Marketplace, organizations can obtain Digital Workers combining autonomous capabilities with the transparency business requires, enabling movement from experimental AI projects toward genuine implementation where systems execute work under complete oversight. ##### Frequently Asked Questions Is Maisa Studio available on AWS Marketplace? Yes. Maisa Studio is available on AWS Marketplace under the new AI Agents category. Enterprises can purchase it through existing AWS procurement agreements, bypassing lengthy security and compliance reviews. Why is Maisa available through AWS Marketplace? Maisa joined to remove procurement and compliance barriers slowing AI adoption. Enterprises already using AWS can access Digital Workers through existing agreements with completed security reviews, allowing teams to focus on automation priorities rather than administrative tasks. What makes Maisa Digital Workers different from standard AI agents? Unlike typical AI agents operating without transparency, Maisa Digital Workers reveal every action: you can see the reasoning steps, tools used, and data sources consulted. This visibility makes them suitable for business-critical processes requiring auditability. ## CLATTER: Academic Validation for Our Maisa AI Hallucination Detection Strategy URL: https://maisa.ai/blog/clatter-hallucination-detection Academic research validates Maisa AI's systematic approach to detecting and preventing AI hallucinations in enterprise applications. Research and Tech Jun 20, 2025 #### The Hallucination Problem: More Critical Than Ever Analysts estimate that chatbots hallucinate as much as 27% of the time, with factual errors present in 46% of generated texts. This makes hallucination detection essential for deploying AI in production environments, particularly in domains like healthcare, legal services, and financial advice, where incorrect information carries serious consequences. Traditional methods relying on simple fact-checking or confidence scoring often miss subtle fabrications or fail with complex, multi-step reasoning. #### CLATTER's Innovative Divide-and-Conquer Approach The CLATTER methodology introduces a systematic four-step process that mirrors Maisa AI's independent development: ##### 1. Decomposition: Breaking Down Complexity The method decomposes generated text into factual claims and attributes these to source evidence, transforming complex outputs into manageable, verifiable units. This granular approach enables precise identification of problematic content rather than broad rejection. ##### 2. Knowledge Base Verification Each extracted claim undergoes rigorous verification against a trusted knowledge base. The system determines whether there is entailment, contradiction, or insufficient evidence for each claim. ##### 3. Parallel Processing Architecture CLATTER processes multiple claims simultaneously, significantly improving efficiency while maintaining accuracy—particularly important for real-time applications. ##### 4. Intelligent Reconstruction The final step feeds the original input, extracted claims, and their verification status back to the model for response refinement, creating a self-healing mechanism that improves output quality without requiring complete regeneration. #### How CLATTER Validates Maisa AI Architecture ##### Knowledge Base Integration Maisa AI's integrated knowledge base consultation system within the KPU aligns with CLATTER's approach. The orchestrator can query the knowledge base to verify claims in real-time, providing systematic verification. ##### Step-by-Step Verification Pipeline The parallel claim checking that CLATTER employs mirrors Maisa AI's internal verification processes, allowing identification and correction of potential hallucinations before reaching end outcomes. ##### Multi-Hop Reasoning Support CLATTER specifically addresses long-form, multi-hop question answering scenarios—exactly the type of complex reasoning tasks Maisa AI handles. The research validates implementing comprehensive verification at each reasoning step rather than only at the final output. ##### Self-Healing Capabilities CLATTER's reconstruction phase aligns with Maisa AI's self-healing approach to reliability. Rather than simply flagging problems, the system actively works to improve outputs based on verification feedback. #### Beyond Academic Theory: Real-World Implementation CLATTER's value lies not just in its theoretical framework but in practical applicability. The research shows that guiding models through a comprehensive reasoning process allows for much finer-grained and accurate entailment decisions, leading to increased performance. Similar improvements appear in Maisa AI's internal testing, where systematic verification has dramatically reduced false information propagation while maintaining response quality and speed. #### The Broader Implications for AI Reliability CLATTER represents a shift from reactive to proactive hallucination management. Rather than hoping models won't hallucinate or catching errors after the fact, organizations can now systematically prevent misinformation from reaching users. As models become better at generating plausible-sounding but incorrect information, simple confidence-based detection methods become insufficient. Systematic entailment reasoning provides a more robust foundation for reliable AI deployment. #### Looking Forward: The Future of Trustworthy AI The CLATTER research validates that building trustworthy AI requires systematic, architectural solutions rather than ad-hoc fixes. The convergence between academic research and practical implementation in Maisa AI suggests alignment with the direction of reliable AI systems. #### Conclusion: Academic Theory Meets Production Reality CLATTER validates that the systematic approach to hallucination detection implemented in Maisa AI represents the future of reliable AI systems. This alignment between academic research and practical implementation provides confidence that organizations can deploy AI for their most critical tasks. While AI hallucinations remain a challenge, systematic approaches like CLATTER and implementations like Maisa AI are building the infrastructure needed for trustworthy AI deployment at scale. ## HALP: Maisa’s breakthrough in delivering reliability for enterprise automation URL: https://maisa.ai/blog/halp Maisa introduces HALP, a method enabling AI systems to learn from real business work without traditional retraining cycles. Research and Tech Jun 4, 2025 AI has made headlines for its potential to transform work, but inside most organizations, turning that potential into reliable automation remains a challenge. Business teams aren't looking for impressive demos or clever assistants. They need AI systems they can trust to follow business logic, respect context, and stay consistent as things evolve. Yet the methods used to build these systems today often work against that goal. What if reliability didn't depend on perfect data or complex training pipelines? What if AI could learn by doing, through real tasks and real feedback, inside the business itself? #### The limits of training methods for enterprises Human-in-the-loop (HITL) methods are used to make AI systems more accurate and aligned with human expectations. They rely on human feedback such as labeled examples, corrections, and supervision to teach models how to behave. This approach has been key to training today's most advanced language models. Systems like GPT and Claude were refined through large-scale HITL processes, helping them perform well across a wide range of generic tasks. But when it comes to enterprise use, this method starts to show its limits. Business processes are specific, tools are unique, and rules change often. Applying HITL in this context means building custom datasets, coordinating technical teams, and retraining models just to keep systems functional. It is slow, expensive, and difficult to scale. For teams that need automation to adapt with the business, this approach becomes a bottleneck. Business logic should not have to wait for model retraining. #### Human-Augmented LLM Processing (HALP). A new way to teach AI What if AI could learn through real work, just like a new team member? HALP changes how we build reliable systems. Instead of relying on retraining cycles or complex setup, it enables AI to learn by doing. HALP stands for Human-Augmented LLM Processing, and it powers Digital Workers that learn directly from the way work happens. Teams explain the task, walk through the logic, and share the tools they use. The system picks up that knowledge through natural interaction, without prompt engineering or rigid rules. Unlike traditional methods, HALP doesn't require labeled datasets or offline feedback loops. The learning happens in context, during real tasks. The system stays aligned with how the business actually works, even as things evolve. #### The reliability enterprises have been missing HALP unlocks what enterprise automation has long lacked: reliability in real work. ##### Fast setup with less effort Digital Workers don't need large datasets or precise prompts. They start from natural interaction and real context. Teams can build and adjust them without relying on IT or external consultants. ##### Lower cost to launch and maintain Less time is spent configuring, correcting, or integrating. Business users can stay involved, reducing handoffs and rework. ##### Scales across teams and processes Digital Workers adapt to different workflows. Logic can be reused, updated, and shared as the business evolves. ##### Trust built into every step Each decision is traceable to a rule or piece of business logic. There is no black-box behavior. Results can be reviewed, audited, and improved. Hallucinations are avoided by grounding decisions in real context. HALP makes it possible to scale automation with confidence; without losing speed, control, or clarity. #### The future of AI looks more like a worker Most AI tools are still treated as assistants. They respond to prompts and handle isolated tasks. Digital Workers go further. They understand tools, context, and goals. They take on work, adapt as things change, and operate as part of the team. This shift changes how we think about AI. It is no longer something to supervise. It is something that collaborates. Reliability is built in, not added later. That is only possible with approaches like HALP, where learning happens from within the business. For teams ready to move beyond experiments, this shift is already underway. #### Frequently Asked Questions ##### What is HALP in Maisa? HALP (Human-Augmented LLM Processing) is Maisa's method for building reliable enterprise AI. Business teams configure Digital Workers by explaining tasks, logic, and tools in plain language – no datasets, no retraining, no prompt engineering. The system learns in context during real work and continuously adapts as processes evolve. ##### How does HALP differ from traditional AI training? Traditional AI training requires custom labeled datasets, technical teams, and retraining cycles to keep systems aligned with business logic. HALP skips this: Digital Workers learn in context during real tasks, staying aligned with live business processes without offline feedback loops or model updates. ##### Can non-technical teams configure Maisa with HALP? Yes. Business teams configure Digital Workers using natural language – describing the task, decision logic, and tools involved. No prompt engineering, labeled datasets, or IT support needed. The system learns from real work and adapts continuously as processes change. ##### How does HALP ensure traceability in automated decisions? Every HALP-powered decision is grounded in a specific rule or piece of business logic defined at setup. There is no black-box behavior: results are reviewable, auditable, and improvable – making HALP suitable for compliance-heavy environments where every automated action must be fully justified. ## Why we built Maisa this way: scientific proof we’re on the right track URL: https://maisa.ai/blog/science-behind-maisa-architecture Maisa's architecture combines code execution with reasoning loops to overcome fundamental limitations of language models. Research and Tech Apr 24, 2025 #### Bridging reasoning and execution ReAct: Synergizing Reasoning and Acting in Language Models remains one of the most important foundations in the evolution of agentic AI. It introduced a core loop: reason, act, observe and repeat. This concept helped reframe LLMs as active decision-makers rather than passive responders, triggering a shift toward treating AI systems as agents capable of planning, adapting, and executing tasks in dynamic environments. LLMs are prone to fluent but inaccurate output due to architectural limitations. Multiple studies confirm that pairing LLMs with code execution significantly improves reliability through verifiable logic, runtime validation, and structured task decomposition. While visible reasoning chains can appear coherent, they often mask logical gaps. Reliability increases when reasoning is grounded in execution, where each step is tested rather than merely described. Research highlights that exposing reasoning chains through Chain-of-Thought prompting does not ensure factual accuracy. The presence of detailed explanations can create false confidence, even when underlying logic is flawed. The model may appear to reason deeply, but steps often serve as post-hoc rationalizations rather than evidence-based logic. Coherence does not equal truth—executable validation remains essential. ##### How this shapes Maisa At the core is a reasoning engine structured around iterative decision-making loops, where each action is informed by observation and continuously adjusted until a defined goal is met. Instead of following fixed instructions, the system adapts continuously as conditions change. Maisa integrates a live code interpreter within the reasoning process, enabling the system to test assumptions, validate outcomes, and apply logical operations as part of its workflow. Code is fundamental, not supplementary. Actions are embedded in executable logic and connected to internal tools, allowing the system to interact with APIs, perform data transformations, and drive outcomes based on real results. #### Beyond the token limit: smarter context management Research shows that LLMs often miss or ignore critical context, especially when it appears mid-sequence or lacks direct phrasing overlap with queries. More context doesn't guarantee better results. Retrieval-Augmented Generation (RAG) offers a middle-ground solution that works but leaves significant room for improvement. Common approaches show clear limitations: fine-tuning often leads to overfitting where models memorize information and lose general reasoning ability, while RAG leaves room for improvement in relevance, structure, and control. ##### Maisa's solution The Virtual Context Window (VCW) is a dynamic system designed to index, navigate, and retrieve user-specific data with precision. Acting as an intelligent memory layer, the VCW delivers structured, real-time knowledge to the reasoning engine before analysis begins. It filters, summarizes, and prioritizes only the most relevant information, reducing token waste, eliminating noise, and maintaining coherence across longer tasks. This approach allows interaction with live, domain-specific data efficiently without retraining or static retrieval strategies. #### Built to adapt: iterative refinement and feedback Research demonstrates that giving LLMs the ability to evaluate their own outputs and refine them during execution leads to significantly better results. The model continuously assesses and adjusts its path forward. The reasoning engine integrates real-time feedback checkpoints. When an outcome diverges from the intended goal, the system generates an internal review, proposes an alternative, and proceeds with a new iteration. This design supports more resilient behavior across multi-step, high-stakes processes, particularly in enterprise workflows where conditions shift mid-task. #### Groundwork for what's ahead The direction Maisa has taken is not accidental. It results from mapping architectural choices to known limitations of LLMs and to emerging techniques that push those limitations further. Scientific foundations remain central as development continues toward enterprise-grade digital workers capable of reasoning, executing, and adapting in real-world environments. ## Advancing our vision for Accountable AI together with Microsoft URL: https://maisa.ai/blog/microsoft-partnership Maisa becomes a Microsoft Strategic Partner and joins the Founders Hub to advance trustworthy, explainable AI automation. News Apr 7, 2025 Maisa has been selected by Microsoft to join the Microsoft for Startups Founders Hub and earn recognition as a Microsoft Strategic Partner. ##### Pushing forward our vision for Digital Workers This partnership reinforces the company's commitment to developing Accountable AI and Digital Workers that automate complex business processes. By leveraging Microsoft's extensive Azure infrastructure and specialized resources, Maisa gains enhanced capabilities to strengthen the reliability, traceability, and performance of its AI technology. ##### Strengthening Accountable AI The collaboration with Microsoft supports the goal of building trustworthy, transparent AI solutions. Maisa continues advancing AI systems that companies can depend on to automate processes through accountable Digital Workers. ##### Frequently Asked Questions What is Maisa's partnership with Microsoft? Maisa is a Microsoft for Startups Founders Hub member and Microsoft Strategic Partner. The collaboration provides access to Azure infrastructure and specialist resources to enhance the reliability, traceability, and performance of Maisa's Digital Workers and advance its Accountable AI mission. Is Maisa Studio available on Microsoft Azure? Yes. As a Microsoft Strategic Partner, Maisa leverages Azure infrastructure to power its Digital Workers, enabling enterprises to deploy AI automation within cloud environments that meet their security and compliance requirements. What does Accountable AI mean for Maisa? Accountable AI means every Digital Worker decision is traceable, evidence-based, and explainable. Unlike black-box AI, Maisa's outputs are grounded in executed logic – not probabilistic guesses – so enterprises can trust, audit, and justify every automated outcome. ## Black Box AI. How can we trust what we can’t see? URL: https://maisa.ai/blog/black-box-ai Opaque AI systems pose accountability and trust challenges that businesses can address through transparent, computation-based approaches rather than probabilistic models. Insights Apr 1, 2025 #### Why is AI opaque? AI systems differ fundamentally from traditional software, which relies on clearly defined rules. Instead, AI learns directly from vast datasets. These models don't have explicit instructions or human-understandable logic guiding their decisions. At their core, AI models use billions of interconnected parameters to convert inputs into outputs through complex mathematical calculations. This method is inherently probabilistic, meaning decisions are based on statistical patterns, not logical reasoning. With billions of these parameters adjusting simultaneously, tracking exactly how or why a specific output was produced becomes practically impossible. Unlike human decision-making, AI doesn't follow structured reasoning steps. It identifies correlations and patterns in data, predicting outcomes without explicit explanations. This absence of clear reasoning pathways means that decisions from AI systems often appear arbitrary, opaque, and difficult to interpret or justify. #### The risks of black-box AI in business Businesses rely increasingly on AI to automate important tasks, yet the opacity of these systems presents clear practical challenges. ##### False confidence and AI hallucinations A major risk of opaque AI is "hallucinations," where AI produces seemingly accurate but entirely incorrect information. These false positives arise when the AI fills knowledge gaps or handles unclear inputs. For example, customer support chatbots might confidently provide false policy details, leading directly to confusion and complaints. ##### Accountability gaps Opaque AI creates accountability issues. Traditional software clearly logs every decision step, making errors easy to track and correct. Black-box AI systems don't offer this clarity. When decision-making relies on hidden AI processes, identifying the exact point of failure becomes difficult, slowing corrections and process improvements. ##### Legal and compliance risks Businesses must increasingly explain automated decisions clearly due to regulations like GDPR. If an AI-driven system, such as a credit scoring tool, makes decisions without understandable reasoning, businesses risk facing regulatory actions, customer complaints, or legal disputes. ##### Uncertainty working with internal data and knowledge Businesses typically want AI to incorporate their specialized data and internal expertise clearly. However, black-box AI models obscure how proprietary business information is actually used. Without clear visibility, enterprises can't confirm that internal knowledge is applied correctly, risking inaccurate outcomes or impractical recommendations. #### Explainable AI (XAI) methods Several methods within Explainable AI (XAI) attempt to clarify AI decision-making processes, helping users understand how AI arrives at specific predictions. Popular approaches include LIME, SHAP, Integrated Gradients, and Chain of Thought reasoning. LIME simplifies individual predictions by creating an easily understandable model around a specific decision. It alters input data slightly, observes the output changes, and builds an interpretable approximation of the AI's decision logic. SHAP highlights the contribution of each feature to an AI's decision, clarifying which factors influenced a specific outcome. Integrated Gradients similarly assesses how decisions change when input data gradually moves from a neutral state to its actual form, identifying the input's most influential parts. Chain of Thought reasoning, used in advanced models like OpenAI's GPT-4, explicitly outlines the AI's reasoning step-by-step, providing clearer insight into the decision-making process. These methods shed valuable light on how AI models function, making their decisions somewhat clearer to users. Still, because AI systems inherently rely on probabilistic reasoning, these explanations remain approximate rather than fully transparent, limiting their ability to offer comprehensive, step-by-step clarity. #### A Computation-First Approach Maisa's mission is to shift AI from probabilistic models towards clear, structured computational systems that businesses can confidently trust. The KPU, its "AI Computer", operates through explicit, step-by-step execution rather than uncertain predictions. Central to this approach is the "Chain of Work," a structured log documenting each decision-making step. This detailed record captures every action, tool, and data source involved, ensuring that decisions are fully traceable and transparent. By clearly logging each step, Maisa helps businesses reduce uncertainty, enabling them to trust and reliably integrate AI into critical workflows. ## Making AI accountable: Maisa raises pre-seed round URL: https://maisa.ai/blog/maisa-raises-pre-seed-round Maisa secures $5 million pre-seed funding to advance accountable AI systems for enterprise automation. News Dec 25, 2024 #### The trust problem in AI Generative AI lacks accountability. Techniques like Chain-of-Thought reasoning, RAGs, and multiagent systems aim to address more sophisticated challenges but still rely on probabilistic predictions, not deterministic computations. AI is unlocking opportunities across countless domains, but the complex world of business demands greater accountability. Mission-critical tasks require not only answers but traceable, evidence-based processes to reach them. Without these, we risk hallucinations—fabricated outputs that render AI results unreliable. This lack of trustworthiness is why fewer than 6% of corporations use AI for anything beyond basic tasks like question-and-answer bots. #### What we are building We believe the solution isn't in refining existing approaches but in creating a new kind of computing system—one that combines AI's creative problem-solving with the determinism of traditional computational systems. With Maisa, you can create bulletproof AI Agents. These are a new generation of Digital Workers that follow natural language instructions to achieve specific outcomes and goals, making intelligent and reliable automation a reality. #### With the best behind us We are fortunate to be backed by visionary investors committed to advancing our mission. Our pre-seed round brought in $5 million, led by NFX and joined by Village Global, the venture fund backed by Mark Zuckerberg, Eric Schmidt, and Jeff Bezos. This funding was further supported by Sequoia's Scout Fund, DeepMind PM Lukas Haas, and other angel investors. This funding enables us to continue developing our product and expanding our research initiatives. We're also genuinely grateful for the recognition our work has received, including a recent feature in Forbes that highlights this important milestone for us. Maisa is going to be a major player in Agentic Process Automation (APA) helping businesses across the world transform their core, business-critical functions through AI. Anna Piñol, NFX David and the Maisa team are building a transformative technology to turn AI agents into actual workers capable of reasoning through complex workflows. Max Kilberg, Village Global #### Come join us If you are interested in joining our mission of making AI accountable, visit our careers page ## Introducing Vinci Knowledge Processing Unit (KPU) URL: https://maisa.ai/blog/vinci-kpu Maisa AI announces Vinci KPU v2, an advanced knowledge processing system that matches leading LLMs on challenging benchmarks while addressing inference compute and scalability limitations. Nov 26, 2024 #### Introduction On March 14, 2024, Maisa AI announced its AI system featuring an innovative architecture called the Knowledge Processing Unit (KPU), designed to address inherent issues in large language models such as hallucinations, outdated information, and context window constraints. The system demonstrated state-of-the-art performance on several benchmarks including MATH, GSM8k, DROP, and BBH. #### Vinci KPU Since the initial March launch, Maisa has focused on addressing inference-time compute limitations and scalability requirements. The company has now released Vinci KPU v2, which the announcement states matches and even surpasses leading LLMs, such as the new Claude Sonnet 3.5 and OpenAI's o1, on challenging benchmarks like GPQA Diamond, MATH, HumanEval, and ProcBench. ##### What's New in Vinci KPU v2 The KPU architecture consists of three main components: a Reasoning Engine for orchestrating problem-solving, an Execution Engine for processing instructions, and a Virtual Context Window for managing information flow. Version 2 improvements include: - Reasoning Engine: Enhanced KPU kernel positioning the LLM as the intelligent core, enabling more sophisticated reasoning and orchestration - Execution Engine: Integration of test-time compute techniques with improved robustness, security, and scalability for tool integration - Virtual Context Window: Refined metadata creation and LLM-friendly indexing for optimized information flow and unlimited context capabilities ##### KPU Architecture Benefits The system's key advantages include: - Model Agnostic Architecture: Better base models yield better performance - Full Multi-Step Traceability: Configurable observability with debug mode and visual representation enabling human-in-the-loop control - Hallucination Mitigation: The system focuses on understanding solution paths rather than generating answers, though execution errors and suboptimal approaches may still occur - Lower Latency: Faster problem resolution than competing systems - Cost Efficiency: Up to 40 times cheaper than RAG and reasoning engines - Continuous Learning: Virtual Context Window enables feedback integration into subsequent iterations - Exception Handling: Autonomous navigation of edge cases and error conditions ##### Reasoning Examples The KPU excels with agentic prompts that require orchestrating multi-step actions across systems without requiring complex prompt engineering. The system follows instructions closely and iterates based on feedback through its traceability features. #### Benchmark Analysis Vinci KPU v2 has been tested on several demanding benchmarks representing current state-of-the-art challenges in AI reasoning and execution. ##### GPQA Diamond This benchmark consists of 198 graduate-level questions from chemistry, physics, and biology. Vinci KPU achieved 70.03% accuracy, ranking second only to o1-preview and demonstrating significant performance separation from other leading models. ##### MATH The MATH dataset includes problems from mathematical competitions with detailed step-by-step solutions. The benchmark is notably challenging—computer science PhD students score around 40% while a three-time IMO gold medalist achieved 90%. Vinci KPU demonstrates superior performance compared to standard LLMs on this dataset. ##### HumanEval This OpenAI benchmark contains 164 handwritten programming problems assessing comprehension, reasoning, algorithms, and mathematical skills. Vinci KPU surpasses all current state-of-the-art models with 94.13% accuracy, positioning it as the leading solution in code generation and functional correctness. ##### ProcBench ProcBench evaluates instruction followability through 23 types of tasks requiring precise step-by-step execution. Vinci KPU performs comparably to o1-preview, achieving 49.75% Sequential Match accuracy and 61.16% Prefix Accuracy. #### Next Steps Maisa plans to: - Open public access to Studio and API, releasing KPU access as serverless agentic functions - Roll out enhanced capabilities including advanced tool usage, multimodal input/output support, and customized observability with debug mode - Deploy other intelligence providers' models in production, including O1 as a reasoning engine - Continue improving the kernel by enhancing intelligence and increasing speed through dynamic kernel proxying The company emphasizes that these results represent an early stage, with the model-agnostic architecture designed for continuous improvement through self-learning capabilities, targeting deterministic, reliable, and traceable outcomes. ## Introducing Maisa, a new AI architecture URL: https://maisa.ai/blog/hello-world Maisa introduces KPU, an architectural framework that addresses fundamental limitations of large language models through decoupled reasoning and execution components. Mar 15, 2024 Recent advances in Artificial Intelligence, particularly in Large Language Models (LLMs) and Vision-Language Models (VLMs), have been driven by sophisticated pre-training methodologies applied to Transformer-based architectures. These models achieve strong performance across benchmarks and provide practical utility through conversational interfaces and API services. Despite their capabilities, language models face several inherent challenges that persist regardless of parameter count or context length: - Hallucinations. The veracity of the response cannot be 100% guaranteed because models generate the most probable token rather than necessarily factual information. - Context limit. The Transformer architecture's attention mechanism has quadratic complexity, increasing processing time and memory as sequences grow. Models also struggle with the "Lost in the middle" problem, failing to retrieve key information from long contexts. - Up-to-date information. Pre-training limitations restrict models to knowledge up to a specific date, affecting their ability to provide current information. - Limited digital interaction. LLMs cannot natively interact with external services like APIs, files, or software systems. #### Architectural Overview Maisa's KPU (Knowledge Processing Unit) comprises three main components: - Reasoning Engine. The system's brain that orchestrates step-by-step planning, relying on LLMs or VLMs and available tools. - Execution Engine. Receives and executes commands while sending results back for re-planning. - Virtual Context Window. Manages data flow between engines, ensuring reasoning stays within context limits while data remains external, and integrates external sources. This design, inspired by operating systems, decouples reasoning from execution, allowing LLMs to focus exclusively on reasoning while reducing hallucination risks and improving factual reliability. #### Benchmarks ##### Setting a new paradigm at Math Reasoning The KPU was tested against established reasoning benchmarks, outperforming models including GPT-4, Mistral Large, Claude Opus, and Google Gemini. On the GSM8k benchmark—consisting of 8,500 grade school math problems—the KPU surpassed all competing models using a zero-shot approach without prompt engineering or iterative attempts. The MATH benchmark, containing 12,500 competition-level problems with step-by-step solutions, similarly demonstrated the KPU's superior performance in generating LaTeX-formatted responses without additional configurations. ##### Boosting reading and discrete comprehension On the DROP benchmark, which requires reasoning over paragraphs including operations like addition and counting, the KPU established a new state-of-the-art result in a zero-shot configuration. ##### Beyond advanced reasoning challenges The Big Bench Hard (BBH) benchmark tests tasks designed beyond current model capabilities. The KPU performed particularly well on tasks requiring logical reasoning, such as Boolean expressions and arithmetic. ##### Summarizing the system's evaluation The KPU has established new standards across multiple reasoning domains: - Elementary math reasoning (GSM8k) - Advanced mathematics (MATH) - Reading comprehension (DROP) - Complex reasoning tasks (BBH) All results and methodology are reproducible through Maisa's published documentation. #### KPU Use Availability Currently in beta phase, the KPU is not yet available for commercial use. Interested parties can join a waiting list for early access, or contact Maisa at contact@maisa.ai or through their contact form. Updates are posted on X (@maisaai_). # Legal ## Terms and Conditions URL: https://maisa.ai/legal/terms-of-service Terms and conditions governing access to and use of Maisa's services. Updated: December 2025 These Terms of Service (“Agreement”) are the agreement governing your access to and use of the Services as defined below. This Agreement is between Maisa, Inc, a Delaware corporation, with offices at 1111B S Governors Ave STE 3624 Dover, DE 19904 (“Maisa”), and the entity you represent by entering into this Agreement (“Customer”). Any capitalized terms not defined throughout the Agreement will have the meaning given to them in Section 17 (Definitions). This Agreement is effective upon the earlier of (i) your acceptance of this Agreement, or (ii) the date you first accessed the Services, as applicable (“Effective Date”), and will remain in effect until terminated in accordance with this Agreement. ### 1. Binding Effect Access to and use of the Website is only permitted for individuals eighteen (18) years of age or older. Access to and use of the Website do not require the creation of a user account. However, in the future, Maisa Inc. may incorporate restricted sections or functionalities that do require user registration. #### Intellectual and industrial property By using the Services hosted in the Platform and/or entering into this Agreement, you represent and warrant that (i) you have read and understand this Agreement, (ii) you understand that the Services provided under this Agreement are for businesses, professionals and developers, not consumers, (iii) you are not a consumer as defined under applicable laws, (iv) you have full legal authority to bind Customer to this Agreement, and (v) you agree to this Agreement on behalf of Customer. If you or Customer do not agree with this Agreement, please refrain from accepting this Agreement and from using the Services. ### 2. Services - Provision of Services. During the Term, Customer will have access to Maisa’s web-based artificial intelligence-powered studio (“Studio”) for the purpose of creating, configuring, and deploying multi-modal AI agentic cloud functions or Digital Workers (“Agents”) on the Platform (collectively, the “Services”) in accordance with this Agreement. - Use of Services. Customer agrees only to use the Services in accordance with this Agreement. Customer’s use of the Services may include deploying the Services to develop Customer Applications and making available Customer Applications to End Users, provided, however, that Customer may not sublicense the Agents or the Services as a standalone or integrated product. Customer will ensure that End User’s use of the Services complies with this Agreement. - Sign up/Account. Customer or End User must sign up on the Platform to create an account (“Account”) to use the Services. The Customer may do so by synchronizing its Google or Microsoft account or by completing the data fields requested by Maisa (name, surname, email) which will be processed in accordance with the Privacy Policy. Customer is solely responsible for all activities that occur under its Account, including using, managing and protecting the Account, including its security, both by Customer and End Users. Customer will not (i) disclose or otherwise share Account access credentials with unauthorized third parties, (ii) share individual login credentials between multiple users on an Account, or (iii) resell or lease access to its account. Customer will (a) promptly notify Maisa if it becomes aware of any unauthorized access to or use of Customer’s account or the Services and (b) use commercially reasonable efforts to prevent and terminate such unauthorized access to our use. - Consent. Customer is solely responsible for obtaining any consent or providing notices required (i) for Customer or End Users to use the Services, (ii) for Customer’s provision of Input or Customer Data to Maisa, and (iii) for Maisa to provide the Services, including using Input and processing Customer Data. - Invite a member. In the “Organisation” tab of the Platform, Customer may have the ability to select the “Invite a Member” feature in order to access the Studio. To invite them to the ‘Team’ they must include the email address of the user who will access as a member and indicate whether they are a “Member” or an “Admin”. ### 3. Use Restrictions; Acceptable Use Policy - Use Restrictions. Customer will not resell, sublicense, transfer or distribute the Services, except as expressly allowed and strictly necessary to make available Customer Applications. Customer will comply with all applicable trade laws, including sanctions and export control laws. - Acceptable Use Policy. Customers must access and use the Services, and must ensure that End Users also access and use the Services, in accordance with the Acceptable Use Policy. Maisa may suspend Customer’s and End Users’ access to and use of the Services if Maisa becomes aware or has reasonable grounds to believe that Customer or End Users are using the Services against or in breach of the Acceptable Use Policy or this Agreement. ### 4. Payment Terms - Fees. Maisa will invoice Customer for the Fees in accordance with the appropriate pricing tier selected by Customer, and Customer will pay Maisa all invoices within 15 days from the invoice date (“Payment Due Date”). Payments are nonrefundable. - Taxes. Customer is responsible for any applicable taxes and will pay invoiced amounts without any deduction, withholding or other reduction to account for taxes, unless Customer provides Maisa with a valid tax exemption certificate for such taxes. - Overdue Payments. If Customer’s payment is overdue: - Maisa may charge interest on overdue amounts at the highest rate permitted by applicable law until the payment is paid in full; - Customer is responsible for all commercially reasonable expenses (including legal fees) incurred by Maisa in collecting unpaid or overdue amounts, except where these unpaid or overdue amounts are due to billing inaccuracies attributable to Maisa; and - Maisa may suspend Customer’s and End Users’ access to and use of the Services or, if Customer’s payment is overdue for more than 30 days after the Payment Due Date, terminate this Agreement. - Disputes. Customer may dispute invoiced amounts if Customer believes in good faith that Fees were inaccurately invoiced (“Invoice Dispute”). Maisa will review in good faith all Invoice Disputes. If an Invoice Dispute is submitted before the Payment Due Date, then (i) Customer may only pay the amounts not subject to the Invoice Dispute; and (b) the rest of unpaid Fees will be paid after Maisa’s resolution of Customer’s Invoice Dispute. If Customer pays Fees that were incorrectly invoiced, Maisa will not issue a corrected invoice, but instead will issue a credit equal to the agreed amount to be used in future invoices until the credit is expired. - Free Access Services/Studio Beta. Maisa may make the Studio available to the Customer as a free and beta version. This version of the Services is not permitted for production use and are provided ‘as is’ on a temporary basis. Maisa is not responsible for the use that the Customer makes of the beta Studio and any resulting Services. - Section 12.1 (Maisa’s Indemnity Obligations) will not apply; - Customer will not use the Services for the processing of any Personal Data. ### 5. Intellectual Property - Except for the limited licenses provided herein, this Agreement does not grant either party any rights, implied or otherwise, to the other party’s intellectual property. Maisa retains all Intellectual Property Rights in the Services and Maisa Technology, and Customer retains all Intellectual Property Rights in the Customer Data, Customer Applications and Input. - Customer Content. Customer and End Users may provide inputs to the Services, such as text, instructions, prompts and other content (“Input”) in connection with its their use of the Services. Customer owns all rights, title and interest, if any, in and to the Input. Customer acknowledges that due to the nature of artificial intelligence services such as the Services, Agents may not be unique and may be substantially similar to Agents received by other customers or users. Content received from the Services by other customers or users are not Output as defined in this Agreement, regardless of the similarity, and this Agreement does not apply to such other content. - Use of Input - By Customer. Customer is solely responsible for: - Input, hereby representing and warranting that Customer has all rights, licenses and permissions required to provide Input to Maisa or Maisa Affiliates via the Services; - use of Agents and evaluation of accuracy and appropriateness of such Agents, taking into consideration that artificial intelligence services, including the Services, are rapidly evolving and therefore Agents may be inaccurate, incomplete or otherwise inappropriate. Consequently, Customer acknowledges that use of Agents requires Customer’s control and supervision, including human review as appropriate. - By Maisa. Maisa will only use Input to provide the Services to Customer and will not use it for any other purposes, including that Maisa will not use Input to train machine learning models, unless expressly permitted by Customer. ### 6. Customer Data; Privacy - Customer Data. If Customer Data includes Personal Data, (i) Customer will process Customer Personal Data in accordance with applicable data protection laws (ii) Customer represents and warrants that it has the right to provide Maisa with such Personal Data; and (iii) the Data Processing Addendum will apply between Customer and Maisa. - Use of Customer Data. Maisa will only use Customer Data to provide the Services in accordance with the Privacy Policy and will not use it for any other purposes. Maisa will implement technical and organizational security measures to prevent accidental or unlawful loss, access, or disclosure of Customer Data. ### 7. Third Party Services When using the Services, Customer may have access to or engage with Third Party Services. If Customer uses or accesses Third Party Services (including by making it available via a Customer Application), Customer understands and accepts that access to or use of Third Party Services may include Output from those Third Party Services are subject to additional terms different to this Agreement. Customer is solely responsible for reviewing any Third Party Services terms. Third Party Services are subject to their own terms, and Maisa is not responsible for them. ### 8. Confidentiality - Confidential Information. “Confidential Information” means information that one party discloses (“Discloser”) to the other party (“Recipient”) under this Agreement and that is marked as confidential or would reasonably be considered confidential information according to its nature and under the circumstances it was disclosed. It does not include information that is (i) independently developed by the Recipient, (ii) is rightfully given to the Recipient by a third party without confidentiality obligations, or (iii) becomes public through no fault of the Recipient. For clarity, Input is Confidential Information. - Protection of Confidential Information. The Recipient will only use the Discloser’s Confidential Information to exercise its rights and fulfill its obligations under this Agreement and will use reasonable care to protect the Discloser’s Confidential Information against unauthorized use or disclosure. Notwithstanding the foregoing, the Recipient may disclose the Discloser’s Confidential Information: (i) to the Recipient’s Affiliates, employees, agents, subcontractors, or professional advisors who have a need to know it and who are bound by confidentiality obligations at least as protective as those in this Confidentiality section; (ii) with the Discloser’s written consent; or (iii) if required by law or court decision, provided that Recipient will try to redirect the request to the Discloser and use reasonable efforts to notify the Discloser in advance, unless prohibited. - Duration of Confidentiality. The confidentiality obligations in this section will last during the Term and for five years following termination or expiration of this Agreement, except that for Confidential Information relating to any product roadmap(s), source code, technical infrastructure security, trade secrets, or compliance documentation, such obligations will survive termination or expiration of this Agreement perpetually or for the maximum duration under applicable law. ### 9. Publicity Subject to Section 5.4.3 (Brand Assets), the parties will not make any public statement or communication regarding this Agreement without the prior written consent of the other party. If the parties wish to make a public announcement regarding this Agreement, they will work together in good faith to decide the content and timing of such announcement. ### 10. Term and Termination - Term. This Agreement will be in effect starting from the Effective Date and will remain in effect until terminated in accordance with the termination provisions in this Agreement (“Term”). - Automatic Renewal. Customer uses the Services under a subscription plan and the subscription will automatically renew for successive periods of the same duration unless Customer provides Maisa with notice of non-renewal at least 30 days before the expiration of the subscription plan . - Termination for Convenience. Subject to any commitments on the duration of the Services, Customer may (i) stop using the Services at any time and (ii) terminate this Agreement for convenience at any time with 30 days prior written notice to Maisa. - Termination for Breach. In addition to any other specific termination or suspension rights in this Agreement, either party may terminate this Agreement immediately upon written notice to the other party if: - the other party is in material breach of this Agreement and fails to cure the breach within thirty (30) days after receipt of written notice of the breach; or - the other party ceases its business operations or becomes subject to insolvency proceedings. - In addition, Maisa may terminate this Agreement or suspend Customer’s or End Users’ access to the Services (a) if required by law or (b) to prevent a security risk or other risk or harm to Maisa or its Affiliates. - Effects of Termination. Upon termination of this Agreement. ### 11. Indemnity - Maisa’s Indemnity Obligations. Subject to Sections 12.3 (Exclusions) and 12.4 (Procedure), Maisa will defend and indemnify Customer for any settlement amounts approved by Maisa or damages finally awarded against Customer by a court of competent jurisdiction in a legal proceeding filed by an unaffiliated third party, to the extent arising from a third party claim alleging that Customer’s use of the Services in accordance with this Agreement infringes any third party Intellectual Property Rights. If Maisa considers that all or part of the Services are likely to become subject to any infringement claim, Maisa may, in its sole discretion, (i) procure, at Maisa’s expense, the right for Customer to continue using the Services in accordance with this Agreement, (ii) replace or modify the allegedly infringing Service so it is no longer infringing with a similar service or functionality, or (iii) if (i) and (ii) are not commercially practicable, suspend the Services or terminate this Agreement upon written notice and refund any prepaid amounts for unused Services. - Customer’s Indemnity Obligations. Subject to Sections 12.3 (Exclusions) and 12.4 (Procedure), Customer will defend and indemnify Maisa for any settlement amounts approved by Customer and damages finally awarded against Maisa or Maisa Affiliate by a court of competent jurisdiction in a legal proceeding filed by an unaffiliated third party, to the extent arising from a third party claim related to (i) Customer’s or End User’s use of the Services against or in breach of the Acceptable Use Policy; (ii) Customer’s use of the Agents, (iii) Customer Applications, (iv) Customer Data, or () Input. - Exclusions. Obligations under this Section 12 (Indemnity) will not apply to the extent that (i) the underlying allegation arises from (a) a breach of this Agreement by the non-indemnifying party, (b) Customer’s combination of the Services with products, services, software or other assets not provided by Maisa, or (c) Third Party Services - Procedure. The party seeking indemnity will (i) promptly notify the other party upon becoming aware of any claim giving rise to indemnification, and (ii) allow the indemnifying party sole control of defense and settlement of such claim. However, the party seeking indemnity is entitled to participate in its own defense at its sole expense. The indemnifying party may not make any settlement on any claim without prior written consent of the other party, which will not be unreasonably withheld. - Sole Remedies. The remedies in this Section 12 (Indemnity) are the sole and exclusive remedies under the Agreement for any third-party claims alleging that the Services infringe third-party Intellectual Property Rights. ### 12. Liability - xclusions. Neither party will have any liability arising out of or relating to this Agreement for: (a) loss of revenues, profits, savings, or goodwill; or (b) indirect losses (whether or not foreseeable or contemplated by the parties). - Limitation of Liability. Except for its indemnification obligations hereunder, each party’s total aggregate liability for direct damages arising out of or relating to this Agreement is limited to the Fees paid by Customer to Maisa under this Agreement for the Services that gave rise to the liability during the twelve (12) month period before the event giving rise to such liability occurred. ### 13. Changes - Changes to Services. Maisa may make changes to the Services from time to time, including publishing new features, changing how certain features look or work or discontinuing specific features. Maisa will make commercially reasonable efforts to inform Customer of material changes to the Services that have a material impact on Customer’s use of the Services. - Discontinuation. Maisa may, at its sole discretion, discontinue the Services in part or in full. Maisa will (i) notify Customer before discontinuing any Service, and (ii) refund Customer any prepaid and unused Fees, in each case unless Maisa replaces such discontinued Service with a materially similar service or functionality. Nothing in this section limits Maisa’s ability to make changes required to comply with applicable law, address a security risk or avoid substantial economic or technical burden. - Changes to Prices. Maisa may make changes to Prices at any time by updating the public URL describing the pricing for the Services. Maisa will make commercially reasonable efforts to notify Customer of changes to price. Unless otherwise agreed, any Price change will become effective 30 days after Maisa informs Customer of the price change. Price changes due to legal reasons will become effective immediately.. Any additional purchases of Services made after a Price change takes effect will be subject to the then-current Price. - Changes to the Agreement. Maisa may update this Agreement from time to time by providing Customer with prior notice, including by posting the updated version of the Agreement online. If the changes to the Agreement materially impact Customer’s rights or obligations at Maisa’s sole discretion, Maisa will provide 30 days prior notice before such material changes become, except that changes made to comply with applicable law will be effective immediately. Any other changes will become effective on the date Maisa posts the updated version of the Agreement. If Customer does not agree with the updated Agreement, Customer may stop using the Services or terminate this Agreement for convenience in accordance with Section 10.3 (Termination for Convenience). Customer’s continued use of the Services after such material updates will constitute Customer’s acceptance of such changes. ### 14. Governing Law; Jurisdiction This Agreement will be interpreted and construed in accordance with the laws of Delaware. As between the Parties, any question concerning the existence, validity, or termination of this Agreement, and any other dispute arising out of or relating to this Agreement, that cannot be resolved by agreement between the Parties shall be finally settled, with express waiver to any other jurisdiction that may correspond to the Parties, exclusively by the courts located in Delaware. ### 15. Miscellaneous - Entire Agreement. This Agreement (i) incorporates by reference any guidelines, policies and terms that apply to the Services; (ii) sets out all terms in relation to its subject matter; and (iii) terminates and replaces all prior and contemporaneous representations, discussions, negotiations and agreements between the parties, whether written or oral, relating to such subject matter of this Agreement. - Notices. Notices regarding compliance by either party with this Agreement, including notices related to termination or breach, must be sent via email, in writing, and addressed to the other party’s Legal Department. Maisa’s address for legal notices is contact@maisa.ai. - Assignment. Customer may not assign this Agreement in whole or in part, including in the event of a change of control, without Maisa’s prior written consent which it may withhold in its sole judgment. - Force Majeure. Except for payment of Fees obligations, neither party will be liable for failure or delay in its performance of this Agreement to the extent caused by force majeure events or circumstances that are beyond its reasonable control. - No Waiver. Neither party will be treated as having waived any rights by not exercising (or delaying the exercise of) any rights under this Agreement. - Severability. If any term (or part of a term) of this Agreement is deemed invalid, illegal, or unenforceable, the rest of this Agreement will remain in effect. - Relationship between the Parties. Maisa and Customer are independent contractors. This Agreement does not create a partnership, franchise, joint venture, agency or other similar relationship between Maisa and Customer. - No Third-Party Beneficiaries. There are no intended third-party beneficiaries to this Agreement. - Equitable Relief. Nothing in this Agreement will limit Maisa’s ability to seek injunctive or equitable relief, including to stop unauthorized access to or use of the Services or infringement of its Intellectual Property Rights. ### 16. Definitions “Affiliates” means any entity that directly or indirectly Controls, is Controlled by, or is under common Control with a party. “Acceptable Use Policy” means the policy set forth in Annex 1 (Acceptable Use Policy). “Brand Assets” means the trade names, trademarks, service marks, logos, domain names, and other distinctive brand assets or features of each party. “Maisa Technology” means all tools, software (excluding open-source software), code, models, algorithms, materials, documentation and other technology developed by Maisa to (i) provide the Services or (ii) in connection with the Services. “Control” means control of greater than 50% of the voting rights or equity interests of a party. “Customer Application” means a program, software, product, service or application developed by Customer by using the Services. “Customer Data” means data provided to Maisa by Customer or End Users through their use of the Services. “End User” means the individuals who Customer allows to use the Services or Customer Applications, which may include Customer’s and Customer’s Affiliate’s employees and other third parties authorized by Customer. “Fees” means the applicable amount to be invoiced by Maisa to Customer for Customer’s or End Users’ use of the Service in relation to Customer’s account. “Intellectual Property Rights” means all patent rights, copyrights, author rights, trademark rights, rights in trade secrets, design rights, database rights, domain name rights, moral rights, and any other intellectual property rights (registered or unregistered). “Personal Data” has the meaning given to it in the Privacy Policy. “Platform” means kpu.maisa.ai “Price” means the then-current price stated on the Platform describing the pricing for the applicable Services. “Privacy Policy” is located at maisa.ai/privacy-policy. “Third Party Services” means any products, services or content offered by third parties that are utilized in the provision of the Services; including, but not limited to large language models. Last update: November 2024 ### Annex 1 – Acceptable Use Policy Access to and use of the Services by Customer and End Users is subject to this Acceptable Use Policy. Customer will not, and will ensure that End Users do not use the Services: - in any way that infringes or otherwise promotes or causes others to infringe any applicable laws or Maisa terms or policies, including on privacy and data protection, child sexual exploitation, child abuse, terrorism, violence, defamation, phishing or other intellectual property rights violations; - for adult entertainment or content, including pornography or other services or content that promote sexual activities; - to distribute malware or any other viruses designed to disrupt, damage or gain unauthorized access to a computer system; - to disable, interfere or circumvent any aspect of the Services; - to generate, distribute, publish or facilitate (i) unsolicited mass email, promotion or advertisements or (ii) hateful, harassing or violent content; - to engage in any conduct that may interfere with the Service’s security; - to copy, modify, create a derivative work of, reverse engineer, reverse assemble, reverse compile, decompile, translate, disassemble, or otherwise attempt to extract or discover any of the source code or underlying components of the Services or Maisa Technology (except to the extent such restriction is expressly prohibited by applicable law); - to develop machine learning and/or artificial intelligence models or related technology that may compete in any way with Maisa’s or Maisa’s Affiliates’ products or services; - to extract data from the Services other than as expressly permitted; - for activities with potential for high risk harm, including fraud, scams, pyramid schemes, plagiarism, fake news or disinformation; - To fully replace advice generally provided by qualified or registered advisors or professionals; including financial advisors, architects, lawyers and doctors, psychologists and any other healthcare professionals. If Customer or End Users use the Services in connection with financial, legal and healthcare activities, Customer must apply reasonable care and judgment and must include, and ensure that End Users include, appropriate disclaimers informing that artificial intelligence is being used, in compliance with any applicable transparency obligations; - To use it for criminal justice, law enforcement, censorship or surveillance purposes; - in any way that infringes or otherwise promotes or causes others to infringe Maisa’s or third party’s rights; - in any way that may damage or otherwise promote or cause others to damage Maisa’s or third party’s reputation or name, including to impersonate or attempt to impersonate Maisa or third parties. ## Privacy Policy URL: https://maisa.ai/legal/privacy-policy How Maisa collects, uses, and protects personal data. ### Privacy Updated: Jul 2026 #### 1. Introduction Maisa Inc. (hereinafter, “Maisa”, “we”, “us”, or “our”) is the owner of the platform (hereinafter, the “Platform”) available at https://maisa.ai/ (hereinafter the “Website“) and is responsible for the processing of the personal data of the users thereof (hereinafter the “Users“). As used herein, “you” and “your” refer to our Users. Through this Privacy Policy, and in compliance with Articles 12 and 13 of Regulation (EU) 2016/679 (hereinafter “GDPR“) and Article 11 of the Organic Law 3/2018, of December 5, on the Protection of Personal Data and guarantee of digital rights (“LOPDPGDD“), Maisa informs Users who use the Website and/or Platform about the processing of their personal data that may be collected through the Website and/or the Platform and processed by Maisa. In order to allow the User to access and use the Platform and Maisa services, Maisa may process the necessary data for the execution of the Terms and Conditions, and more generally to manage our contractual and/or commercial relationship with the Users, and to inform them promptly about any aspect related to the services provided or that can be performed by Maisa in the future. #### 2. Data controller - Data Controller: Maisa AI Inc. - EIN Number: 99-2702335 - Registered office: Delaware, with offices at 1111B S Governors Ave STE 3624 Dover, DE 19904 - Contact: privacy@maisa.ai #### 3. Data processed, purposes, legal basis and retention period - Personal data collected on the Maisa Website: #### Data Processed #### Purposes #### Legal Basis #### Retention Period Browsing data. The IP address of User’s computer and the type of browser User is using. We use this information to analyze general trends and improve our services, including the Website. Except for disclosures described herein, this information is not shared with third parties without User’s consent. Management of the functionality of the Platform and/or Website. Analysis of browsing behavior and statistics: The information collected through cookies and other similar tracking technologies allows an analysis of the browsing of Users. Consent of the User (Not necessary cookies). Legitimate Interest (Necessary Cookies). The retention periods depend on each specific cookie. For more information on the information retention periods for each type of cookie, see the Cookie Policy: https://maisa.ai/cookie-policy/ Waitlist. Name, last name, email, and company. Send information and news about the launch of the Platform. Consent The personal data will be processed by Maisa for the necessary period of time to inform about the Platform launching. Such data will be blocked stored in order to comply with Maisa’s legal obligations and, after that time, will be definitively deleted. - Personal data collected on the Maisa Platform: Browsing and Maisa Platform usage data. The IP address of User’s computer and the type of browser User is using. We use this information to analyze general trends and improve our services, including the Platform. Except for disclosures described herein, this information is not shared with third parties without User’s consent. Analysis of browsing behavior and statistics: The information collected through cookies and other similar tracking technologies allows an analysis of the browsing made by Users. The retention periods depend on each specific cookie. For more information on the information retention periods for each type of cookie, see the Cookie Policy: https://maisa.ai/cookie-policy/. Billing data. Name, tax id, email, address in order to bill and send the invoice. Invoicing Services Execution of the Terms and Conditions The personal data will be processed by Maisa for the necessary period of time to answer the question and/or solve the issue indicated by the User. Such data will be stored in order to comply with Maisa’s legal obligations and, after that time, will be deleted. Registration data. User’s name, surname, email, Google or Microsoft SSO. Sign up on the Platform. Manage the registration of Users on the Platform. Please be aware that if you choose to access certain portions of the Website and/or Platform using the Google or Microsoft login feature, you may be providing us with access to information, attributes, and data in your Google or Microsoft public profile. We may store and use such information in accordance with this Privacy Policy. Such information may be stored by us and may be used in connection with providing the Website and/or Platform or to contact you. Execution of the Terms and Conditions. The personal data provided will be stored for as long as the User maintains its account. In the event that the User wishes to cancel his/her account Maisa will proceed to delete the data. Such data will be stored in order to comply with Maisa’s legal obligations and, after that time, will be deleted Profile data. Name, surname, email, browsing data. Manage and personalize the User experience on our Platform. The personal data provided will be kept for as long as the User maintains its account. In the event that the User wishes to cancel his/her account Maisa will proceed to delete the data. Commercial and services communications. User’s contact email addresses. Maisa may send communications related with Maisa’s services, including the Platform that it may be of interest to Users. Legitimate interest. The personal data provided will be stored for the duration of your use of our services, including the Platform. Once your access to the Platform has terminated, your data will be stored in order to comply with Maisa’s legal obligations and, after that time, they will be deleted. Invite a member. Name, surname and email address. Management of the invitation process to become a member of Maisa, send the communication related to the invitation and provide access to the Platform Personal data will be retained for the time necessary to manage the invitation process. *Maisa may also process the User’s personal data during the period of time in which civil, commercial, administrative, criminal or tax liabilities may arise, in compliance with the applicable regulations in force. Once these periods have expired, Maisa will delete this personal data. #### 4. Data communication to third parties and international transfers For the purposes described above, Maisa may give access to Users’ personal data to technological service providers with which it has entered into a data processor contract, including in particular to: - Database service provider. - Customer Relationship Managment provider (CRM). - Artificial Intelligence services providers. - Analytics service providers. - Hosting service providers. - Payment services providers - Billing and charging service providers. - Email service providers. - Newsletter subscription management service provider. - Survey management service providers - Consent Management platform providers. The aforementioned providers may be located in jurisdictions that generally do not provide adequate safeguards in relation to the processing of personal data. For all entities that are not part of the European Economic Area (EEA), Maisa has entered into contracts with such entities that include such safeguards, including the European Commission’s standard clauses. Also, the personal data may be provided to: - Companies interested in buying or acquiring or investing in Maisa, a part of its business, and consequently, give access to any national or international auditor to carry out their due diligence or your personal data may be provided to third parties in the event that we discontinue our business or file a petition or have filed against us a petition in bankruptcy, reorganization or similar proceeding . - Authorities in order to investigate suspected fraud, harassment or other violations of any law, rule or regulation, or of the Platform or Website, or if we believe in good faith that we are required to do so in order to comply with an applicable statute, regulation, rule or law, a subpoena, a search warrant, a court or regulatory order, lawful requests by public authorities, including to meet national security or law enforcement requirements, or other valid legal process. - Professional advisors, such as lawyers, auditors, bankers, and insurers, where necessary in the course of the professional services that they render to us. - Our “Corporate Affiliates”, meaning any person or entity which directly or indirectly controls, is controlled by or is under common control with Maisa, whether by ownership or otherwise; and “control” means possessing, directly or indirectly, the power to direct or cause the direction of the management, policies or operations of an entity, whether through ownership of fifty percent (50%) or more of the voting securities, by contract or otherwise. In addition to the above, when Users use the Platform or Website, third parties (including without limitation third-party advertisers, analytics service providers, and commercial partners) may directly collect information about our Users’ online activities over time and across different websites. The third parties to which we may provide or who may independently directly collect information may include potential or actual advertisers, providers of advertising products or services (including vendors, analytics services providers, and website tracking services), affiliates and other actual or potential commercial partners, and other similar parties. Please note in particular that the Website uses Google Analytics, including its data reporting features. Information collected by Google Analytics includes but is not limited to web metrics. For information on how Google Analytics collects and processes data, please see the site “How Google uses data when you use our partners’ sites or apps”, currently located at www.google.com/policies/privacy/partners/. For information on opting out of Google Analytics, we encourage you to visit Google’s website, including its list of currently available opt-out options presently located at https://tools.google.com/dlpage/gaoptout. Furthermore, the Website and/or Platform may use third-party service platforms (including to help analyze how Users use the Website and/or Platform). These third-party service platforms may place cookies on your computer or mobile device. If you would like to disable “third party” cookies, you may be able to turn them off by going to the third party’s website. Here are links to the main third-party platforms we use: https://www.google.com/policies/privacy/ https://posthog.com/privacy https://help.x.com/en/rules-and-policies/x-cookies Finally, we would like you to be aware that we use a service provider for payment, analytics, and other business services. This service provider may collect personal data including via cookies and similar technologies, including when you make a payment through the Website and/or Platform or otherwise interact with our service provider that handles billing and payment card processing. The personal data such service provider collects may include transactional data and identifying information about devices that connect to its services. Our service provider uses this information to operate and improve the services it provides to us, including for fraud detection, loss prevention, authentication, and analytics related to the performance of its services. You can learn more about that service provider and read its privacy policy at https://stripe.com/privacy. #### 5. Security and confidentiality Maisa undertakes to adopt the necessary technical and organizational measures, in accordance with the provisions of the applicable regulations, designed to ensure the security of personal data and to prevent the accidental or unlawful destruction, loss or alteration of personal data transmitted, stored or otherwise processed, or unauthorized communication or access to such data. The personal data will be processed confidentially by Maisa, which undertakes to inform and enforce, by legal or contractual obligation, such confidentiality to its employees, partners and any person who should have access to the User’s personal data. #### 6. Users rights The User may exercise the rights of access, rectification, deletion, opposition and, where appropriate, limitation of processing and portability of data by sending an email to privacy@maisa.ai Likewise, the User may file a complaint with a supervisory authority and, in particular, with the Competent Data Protection Agency (in Spain, the AEPD: http://www.aepd.es) if the User considers that the rights set forth above in this Privacy Policy have been violated or the User considers that the processing of personal data concerning him/her violates the applicable regulations. The term “Do Not Track” refers to a HTTP header offered by certain web browsers to request that websites refrain from tracking the visitor. We take no action in response to automated Do Not Track requests. However, if you wish to stop such tracking, please contact us with your request, using our contact details provided above. #### 7. Maisa as data processor If an organization with which you are associated (an “Organization”) signs up to use our services, we may receive information about you in connection with our provision of such services to your Organization. To the extent we process that information solely in order to provide such services to your Organization, we will act as a data processor on behalf of your Organization in respect of that information, which means: we will handle that information solely at the direction of your Organization; your Organization’s privacy policy (and not this Privacy Policy) will apply to the processing of that information; and your Organization (and not us) is responsible for obtaining all necessary consents and providing you with all requisite information as required by applicable law. And in such cases, Maisa will process such personal data as a data processor in accordance with the Data Processing Agreement. To the extent we process your information for any other lawful business purpose of ours, this Privacy Policy will apply to the processing of such information. #### 8. Children We are committed to protecting the privacy of children. The Website and Platform are not designed for or directed to children under the age of 13. We do not collect information from any person we actually know is under the age of 13. #### 9. Links to other websites Our Website and/or Platform may provide you with access to other websites and services. Please be aware that we are not responsible for the privacy practices of any websites or services other than the Website and/or Platform. A link to a third-party website does not constitute or imply endorsement by us. Additionally, we cannot guarantee the quality or accuracy of information presented on those websites. We encourage you to read the privacy policies or statements of each and every such website and service. This Privacy Policy applies solely to information collected by us or on our behalf through the Website and/or Platform. #### 10. Changes in the Privacy Policy Maisa reserves the right to modify this Privacy Policy at any time. We will not make changes that result in significant additional uses or disclosures of your personal data without allowing you to “opt in” to such changes. We may also make non-significant changes to this Privacy Policy that generally will not significantly affect our use of your personal data, for which your opt-in is not required. We encourage you to check this page periodically for any changes. If any non-significant changes to this Privacy Policy are unacceptable to you, you must immediately contact us and, until the issue is resolved, stop using the Platform and/or Website. Effective Date: December 12, 2024 ## Cookie Policy URL: https://maisa.ai/legal/cookie-policy How maisa.ai uses cookies and how to manage them. ### Cookies Updated: Jul 2026 MAISA INC., (hereinafter “MAISA“) is the owner of the Website https://maisa.ai/ (hereinafter the “Website“) and it’s the owner of the platform https://studio.maisa.ai/ (hereinafter the “Maisa Studio”). Both of them use cookies that collect information related to the connection, browsers, and devices used by Internet users who access or use the Website and/or the Maisa Studio (hereinafter the “User/s“). MAISA uses this information to manage and improve the proper functioning of the Website and/or the Maisa Studio. This Policy describes what information these cookies collect, how they are used and for what purpose. It also indicates how the User can restrict or block the automatic downloading of cookies, however, this could reduce or even hinder certain elements of the functionality of the Website and/or the Maisa Studio. Likewise, the User can choose the category of cookies that he/she wishes to activate in the cookies banner that appears the first time he/she accesses the Website and/or the Maisa Studio. ### 1. Definition of Cookies Cookies are small text files that are placed on the User’s computer, smartphone or other device when accessing the Internet. This is done to improve the User’s experience and for other purposes, such as recognizing Users when accessing the Website and/or the Maisa Studio, ensuring the security of your account and delivering targeted advertising. For more general information about cookies, please see the following article. ### 2. How we use Cookies In summary, MAISA uses the cookies listed in Annex I for the Website and the cookies listed in Annex II of this Policy for the Maisa Studio to track how the Website and/or the Maisa Studio is used in order to optimize its operation. ### 3. What Cookies we use The Website and/or the Maisa Studio use both its own and third-party cookies: - First-party cookies: cookies sent to your device by MAISA through the web domain. - Third-party cookies: these are sent to your device by domains that are not managed by MAISA but by another entity that processes the data collected through cookies. According to the purpose of the cookies, the cookies used by MAISA can be divided into the following categories: - Technical cookies (necessary): cookies necessary for navigation and for the proper functioning of the Website and/or the Maisa Studio. Their use allows basic functions, such as access and secure navigation. The legal basis that allows the collection of data through these cookies is the legitimate interest of MAISA in the management of the Website and/or the Maisa Studio. No information collected through these cookies is shared with third parties. See the cookie table below for more details of these cookies. - Analytical cookies: allow monitoring and analyzing the behavior of Users. The information collected through this type of cookies is used to measure the activity of the Website and/or the Maisa Studio and for the elaboration of browsing profiles of the Users, in order to improve the Website and/or the Maisa Studio and their services. The legal basis for collecting this data through these cookies is the consent of the User. See the table of cookies below for more details of these cookies. - Preference cookies: these are cookies that are installed in order to customize the User’s experience (that is, the way in which certain windows will appear or not in the User’s navigation) according to the consents and preferences that, where appropriate, the User expresses while browsing the Website, such as control over the acceptance of this Cookies Policy by the User (this will mean that the cookie warning banner will no longer appear, with its corresponding acceptance or not, for the User). As in the case of analytical cookies, the legitimate basis that allows us to collect such data from the User through these preference cookies is the consent of the User. - Advertising cookies: allow the management of advertising space based on different criteria. Those cookies that track the browsing habits of Users to obtain a profile that allows to offer the most interesting information or adjusted to their tastes and preferences identified on the basis of queries and browsing habits of the User, even on web pages not managed by MAISA. The legitimate basis for collecting this data through these cookies is the consent of the User. See the table of cookies below for more details of these cookies. You can find the detailed list of the cookies we use, the type, purpose and duration of each one of them in the table of cookies below for the Website (Annex I section a) and for the Maisa Studio (Annex I section b). ### 4. Disable Cookies The User may restrict or limit the placement of cookies by adjusting the settings of your browser. However, if you select this setting, you may not be able to access certain parts of the Website and/or the Maisa Studio or the poor display of its contents, which will result in less efficient browsing and you may not be able to take advantage of some of the services available. Therefore, it is recommended that you do not disable them. If you wish to disable cookies, go to the Browser Preferences menu or Browser settings and look for the Privacy section. As an example, you can follow the instructions below: - Microsoft Edge - Google Chrome - Firefox - Safari ### 5. More information We hope we have clarified how and why we use cookies. If anything is still unclear, please do not hesitate to contact us at privacy@maisa.ai Last version: 12/02/2026 ### Types of cookies we use #### Necessary Necessary cookies are required to enable the basic features of this site, such as providing secure log-in or adjusting your consent preferences. These cookies do not store any personally identifiable data. Cookie Duration Description __cf_bm 1 hour This cookie, set by Cloudflare, is used to support Cloudflare Bot Management. cookieyes-* 1 year CookieYes sets this cookie for consent solution management. #### Functional Functional cookies help perform certain functionalities like sharing the content of the website on social media platforms, collecting feedback, and other third-party features. li_gc 6 months Linkedin set this cookie for storing visitor's consent regarding using cookies for non-essential purposes. lidc 1 day LinkedIn sets the lidc cookie to facilitate data center selection. #### Analytics Analytical cookies are used to understand how visitors interact with the website. These cookies help provide information on metrics such as the number of visitors, bounce rate, traffic source, etc. _ga_* 1 year 1 month 4 days Google Analytics sets this cookie to store and count page views. _ga Google Analytics sets this cookie to calculate visitor, session and campaign data and track site usage for the site's analytics report. The cookie stores information anonymously and assigns a randomly generated number to recognise unique visitors. #### Advertisement Advertisement cookies are used to provide visitors with customised advertisements based on the pages you visited previously and to analyse the effectiveness of the ad campaigns. bcookie LinkedIn sets this cookie from LinkedIn share buttons and ad tags to recognize browser IDs. _gcl_au 3 months Google Tag Manager sets this cookie to experiment advertisement efficiency of websites using their services. test_cookie 15 minutes doubleclick.net sets this cookie to determine if the user's browser supports cookies. ## Legal Notice URL: https://maisa.ai/legal/legal-notice Ownership, conditions of access, and terms of use of the maisa.ai website. Updated: Jul 2026 “This legal notice (the “Legal Notice”) governs access to and navigation of the website www.maisa.ai (the “Website”). The Website is owned by Maisa Inc., (“Maisa” or “we”), whose identifying and contact information is as follows: Address: 8 The Green STE R, Dover, Kent County, Delaware 19901. Contact email address: contact@Maisa.ai REG & C.o.Incorporation: State of Delaware, Division of Incorporations SR20233303552FN7632442 This Legal Notice is binding for anyone accessing the Website (the “user” or “you”). Please note that by browsing the Website, you acknowledge that you have read and agree to be bound by the following documents: this Legal Notice, our Privacy Policy, and our Cookie Policy. If you do not agree with any of these texts, you should not access or use the Website. The original version of this Legal Notice has been drafted in Spanish. However, Maisa Inc. may, as a courtesy, provide users with versions of this Legal Notice in other languages (for example, in English). In case of contradiction between versions, the Spanish version will prevail. #### Conditions of access and use of the website Access to and use of the Website is only permitted for individuals eighteen (18) years of age or older. Access to and use of the Website do not require the creation of a user account. However, in the future, Maisa Inc. may incorporate restricted sections or functionalities that do require user registration. #### Intellectual and industrial property Maisa Inc. holds the intellectual and industrial property rights over the Website and all its related elements. This includes, for example: All rights to the source code, object code, interface, databases, and other elements of the Website. All content on the Website (images, texts, videos, etc.). All rights to the trademarks, trade names, and other distinctive signs of Maisa Inc. Users are not authorized to reproduce, distribute, publicly communicate, or transform the Website or its contents. By way of example, this means that users may not extract or reuse, in whole or in part, the information available on the Website, regardless of whether the extraction is done through automated techniques (screen-scraping, bots, spiders, etc.) or manually. #### Permitted uses of the website As a user of the Website, you declare and warrant that you will make appropriate use of it. The following list includes, for example, some of the commitments you undertake: You will not use the Website to transmit or install viruses or other harmful elements. You will not attempt to access restricted sections of the Website or its systems and networks. You will not try to breach the security or authentication measures of the Website. You will not replicate or reverse engineer or decompile the Website (except in cases where the law expressly authorizes it). You will not engage in abusive use of the Website or use it in a way that could cause saturation of the Website. You will not use the Website to extract information that allows you to offer a product or service (analog or digital) competing with the Website or other products and services offered by Maisa Inc. You will not use the Website for illegal purposes. The above list is merely indicative, so you must comply with all other obligations applicable to you according to this Legal Notice and current regulations. If you make unauthorized use of the Website or if, for any other reason, your use of the Website causes harm to Maisa Inc. or third parties, you must hold us harmless. In the event that we receive claims from third parties, this means you must cooperate with us so that we can defend ourselves, and you must also indemnify us for the damages caused by these claims. #### Temporary and/or permanent access restriction to the website Your access to the Website may be temporarily or permanently blocked, without prior notice and without any liability to Maisa Inc., if Maisa Inc. knows or has reasonable suspicions that you are violating this Legal Notice or applicable regulations. We may also temporarily or permanently (as the case may be) restrict access to the Website when we are carrying out maintenance tasks or when, for any reason, we cease to provide our services related to the Website. You will have no right to make any claim against Maisa Inc. related to the temporary or permanent blocking of your access to the Website.” #### Third-party website links The Website may contain links to third-party websites. #### Processing of personal data At Maisa Inc., we take the processing of your personal information very seriously. Please refer to our Privacy Policy and Cookie Policy for more details on how we handle your personal data. #### Representations, warranties, and liability of Maisa Inc. The Website is provided “as is” and “as available,” without any warranties of any kind. Please note that we shall not be liable for damages you may suffer due to: The unsuitability, accessibility, availability, usefulness, or performance of the Website. Errors or lack of accuracy or quality of data and information contained on the Website. Possible compatibility issues between the Website and the device from which you are accessing it. The illegitimate intrusion of a third party or the introduction of viruses into your computer systems. When applicable law does not allow the limitation of liability within the bounds stated in this clause, Maisa Inc.’s liability will be limited to the extent permitted by law. #### Suggestions, inquiries, and complaints If you wish to send us a notification (including any suggestions, inquiries, or complaints), you can write to us at the email address contact@maisa.ai. #### Changes of the website and/or this Legal Notice We reserve the right to make improvements, substitutions, or modifications to any functionality or content of the Website at any time and without prior notice. Additionally, we may modify this Legal Notice at any time. If, after the new version of the Legal Notice is published, you continue to access or use the Website, it will be understood that you have accepted the new Legal Notice and will be bound by its terms. #### “Waitlist” conditions Through the Website, you may be able to join a “waitlist.” Joining a “waitlist” means that we will keep you informed about our updates but does not imply a commitment on the part of Maisa Inc. to effectively launch new products and services or to give you preferential access to such products and services compared to other individuals on the “waitlist” subsequently. #### Applicable Law and Jurisdiction This Legal Notice and the access and use of the Website are governed by Spanish law. If you are a consumer or user according to applicable regulations, when the law requires it, we will submit any conflict related to this Legal Notice and the access and use of the Website to the exclusive jurisdiction of the courts of your place of residence. In all other cases, whenever the law allows, you and we, expressly waiving our own jurisdiction, agree to submit to the courts and tribunals of the city of Madrid. If you are a consumer or user according to applicable regulations, you may also use the online dispute resolution platform of the European Commission, which can be accessed through this link.