IN THIS SESSION

AI maturity is accelerating across Latin American banking, but the advantage is shifting from adopting AI to building systems with compounding returns. The next phase will be defined less by model access, and more by how effectively banks combine proprietary data, reusable platforms, and effective controls to reinvent workflows at scale.

In this session, Rohan Ramanath (General Manager, Artificial Intelligence) joined us to discuss Nubank’s #1 ranking in the 2026 Evident AI Index for Banks - LATAM Edition, how its digital-native foundations translate into AI advantage, and what its experience says about what’s next for AI in the region.

Watch the roundtable

Meet the Speakers

Key Discussion Topics

First Principles vs. Platforms: Rohan drew an important distinction between solving problems and scaling solutions. Banks still need deep expertise to deconstruct individual workflows from first principles, but there’s “alpha” in understanding how a problem should actually be solved. Once those insights are firmly established, reusable abstractions can prevent teams from starting over whenever a new model, product, or technology arrives.

Reusability Creates a Compounding Advantage: NuFormer illustrates what this looks like in practice. Rather than building separate models for individual credit or risk decisions, Nubank has developed a common foundation model for financial behavior that can support underwriting, deposits, growth, recommendations, and personalization. Research improvements flow back into the same backbone, allowing multiple businesses and markets to benefit without rebuilding or retraining from scratch. Rohan linked that reusability directly to faster deployment – with Nubank progressing from roughly one launch every six months, to quarterly, to monthly, and now approximately every other week.

Agentic “Traces” Are the Next Data Flywheel to Unlock: Rohan described agentic “traces” as the new oil. This term refers to the record of user requests, agent reasoning, tool calls, and responses generated as people interact with AI systems. Studied at scale, these traces can reveal how employees work, what customers need, where systems fail, whether controls are effective, and where AI is generating value. The ambition is to turn those interactions into a learning loop where every user contributes new context to a shared “central brain,” making the system more capable for the next user.

Embedded AI Controls Are an Accelerant, Not a Constraint: Rather than treating governance as a constraint on innovation, Rohan argued that it can become a “superpower” because the ultimate winner in AI may be the institution that can make the strongest trust argument. Nubank’s approach rests on reusable pre-launch controls, independent second-line testing, continuous fairness monitoring, and explainability. Crucially, better models provide better signals – but do not independently determine risk appetite, policy, or business decisions. Keeping model development, risk oversight, and decision-making distinct allows Nubank to innovate faster without weakening accountability.

The End State is Not Adding AI to Banking, But Rebuilding Banking Around AI: Rohan framed AI investment across three dimensions: AI as a building mechanism, AI as a driver of better decisions, and AI as a delivery mechanism. The opportunity ultimately extends beyond productivity toward redesigning how financial services are delivered. Instead of waiting for customers to identify a need, navigate an app, and request a product, AI could enable banks to proactively identify opportunities to improve customers’ financial lives. But that future depends on solving the trust problem. Specifically, combining increasingly capable AI with human accountability, explainability, and controls strong enough to make proactive financial assistance reliable at scale.

Explore the ranking