Maisa x Santander: 6 Lessons from Scaling AI Inside a Large Global Bank

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.