How Itaú is building generative AI for its entire customer base

The challenge is to make the technology disappear

Creating a close, personalized experience for millions of customers in a country like Brazil is, first and foremost, a question of simplicity rather than of models.

“Much of it is about making something simple, direct, and useful, delivering a direct benefit to the customer, and making the technology almost unnoticeable,” says Carlos Eduardo Mazzei, CTO of Itaú.

The standard he describes is demanding: the solution must solve a real everyday problem and be direct, simple, and memorable. Technology serves as the means, never as the centerpiece of the experience.

There is demand for this. A study by Itaú in partnership with Grupo Consumoteca shows that 80% of Brazilians want to better organize their financial lives and seek more effective support for decisions, and that 65% would like technologies that act as a financial copilot, providing ongoing guidance.

“OpenAI’s models were very important in delivering this experience, because they were able to operate across this range of solutions.”

—Carlos Eduardo Mazzei, CTO, Itaú

ia.i: from convenience to advice, in one place

The journey began in 2024, when the bank introduced the first Inteligência Itaú capabilities for specific financial needs. ia.i is the next stage: a complete conversational layer, activated by text or voice, bringing interactions related to checking accounts, credit, cards, insurance, and other services into one environment — and connecting customers to specialized human support whenever needed.

This breadth was precisely what guided the choice of models.

“OpenAI’s models were very important in delivering this experience, because they were able to operate across this range of solutions,” Mazzei explains. “From the simplest ones, which need to be efficient in cost and latency, to more complex solutions, where reasoning and all the necessary grounding are required to provide an appropriate answer.”

Having a range of models — rather than a single model — made it possible to deliver the complete solution in one product. At the center of this experience is Customer DNA, a layer of financial context that helps the bank understand the needs, habits, and life stages of approximately 70 million customers. Under governance and responsible AI controls, GPT models are among the components that help maintain and enrich this representation, making interactions progressively more relevant.

The solution rests on three principles: interactivity, multimodality — text, voice, image, and clicks in the same experience — and individualization.

With approximately 300,000 customers in the first phase, the bank entered a period of learning about the real relationship between customer and product.

“We can understand how customers begin to use this product more deeply and see how it can directly benefit each person’s daily financial life,” says Mazzei. Its presence extends across the journey: from a question about an expense in an account statement to investment advice and convenient transactions on WhatsApp.

“The financial world often has products that are not so straightforward and need this context, this simpler relationship with money.”

IARA: the layer that makes scale possible

Behind the visible experiences is an architectural decision. To operate AI at the scale of a bank, Itaú built IARA — a governance and orchestration platform that gives teams controlled access to different models in line with internal policies. OpenAI is among the model providers available in this environment.

This layer supports solutions such as the legal department’s, where the platform automates document reading, analysis, and interpretation. The solution operates on a repository of more than 90 million documents and executes more than 2 million tasks per month, with efficiency gains of 20% to 80% in stages of the department’s processes.

“It is essential that this information be preserved, remain within the bank’s context, and access models in a secure, governed way, with an understanding of token-use efficiency so that the entire solution is economically viable for the bank,” Mazzei says.

IARA addresses three problems at once: cost control per token, accuracy of responses, and application of responsible AI policies. In such a regulated sector, this makes it possible to move quickly while maintaining control.

“When we talk about Responsible AI at the bank, it is a discipline we take great care with,” he says. “We need customers to interact with the bank and our responses to align with our values and culture.” This includes reducing any risk of bias that may arise in frontier models.

Responsible AI principles guide the entire lifecycle of solutions, from development to continuous monitoring, aiming to ensure safety, transparency, ethics, and regulatory compliance as the technology scales.

“IARA is a fundamental platform, because it ensures we can have governance and observability in a highly regulated environment.”