Itaú‘s goal is to make generative AI invisible to customers. Carlos Eduardo Mazzei, CTO of the Brazilian bank, describes the design standard as “simple, direct, and useful,” where technology is the means, not the centerpiece. A study by Itaú with Grupo Consumoteca found demand: 80% of Brazilians want better organization of their financial lives and more support for decisions, and 65% would like a financial copilot that provides ongoing guidance.

The customer-facing layer is ia.i, the next phase of the bank’s Inteligência Itaú initiative. It brings interactions for checking accounts, credit, cards, insurance, and more into a single conversational environment activated by text or voice, and can connect customers to specialized human support. OpenAI‘s models were chosen because they cover the full range of needs—from low-cost, low-latency simple requests to complex questions requiring reasoning and grounding. This range of models, rather than a single model, is what lets one product serve many use cases. At the center is Customer DNA, a representation of the financial context for roughly 70 million customers. GPT-5.x models are part of Customer DNA, and under governance and responsible AI controls they help keep interactions relevant. The product runs on three principles: interactivity, multimodality (text, voice, image, and clicks in one experience), and individualization. In its first phase, ia.i reached about 300,000 customers, which the bank is using to study the real relationship between customer and product.

Scaling AI in a regulated bank requires more than a model. Itaú built IARA, a governance and orchestration platform that gives internal teams controlled access to different model providers, including OpenAI, in line with internal policies. IARA addresses three problems: cost control per token, response accuracy, and the enforcement of responsible AI policies. The legal department uses it for document reading, analysis, and interpretation across a repository of more than 90 million documents, executing more than 2 million tasks per month with efficiency gains of 20% to 80%. Mazzei says the platform keeps information inside the bank’s context, accesses models securely and governed, and manages token-use efficiency so the solution is economically viable. Responsible AI principles guide the full lifecycle, from development to continuous monitoring, with attention to bias and cultural alignment.

The same intelligence layer is applied internally. Relationship managers and customer service teams get meeting summaries, preparation, and support for understanding customer issues, plus plans and alternatives that were previously harder to produce. The bank expects three times more time available for customer relationships, service up to twice as fast, and conversations up to three times more effective.

Reported results repeat these numbers: ia.i in first phase at 300,000 customers; IARA with governance, observability, cost control, responsible AI; legal automation with 2 million monthly tasks and 20-80% gains; and a multi-model approach covering different reasoning needs. The bank plans to make ia.i available to its entire customer base by the end of 2026. Upcoming capabilities include product simulations and sign-ups, plus recurring financial analyses of income, expenses, and habits. The leadership lessons in the article: measure the experience by simplicity rather than sophistication, choose a range of models, build governance before scaling, treat bias and cultural alignment as product requirements, and apply AI internally. Mazzei frames the initiative as responding faster to customer needs and helping people engage with financial products in a simpler, more direct way.