Itaú Unibanco Deploys Generative AI Assistant to 70 Million Customers via IARA Governance Platform
Itaú Unibanco is expanding its generative AI conversational assistant, ia.i, to its entire base of 70 million customers by the end of 2026, according to agent5.news. The deployment follows a phased launch that has already engaged over 300,000 users, marking a structural shift in how Latin America’s largest private bank processes retail banking interactions through a unified text and voice interface.
How ia.i integrates transactional and conversational banking
The ia.i assistant represents the next phase of the bank’s digital strategy, which initially launched in November 2024 with specific financial use cases under the Inteligência Itaú banner. Unlike traditional banking chatbots, ia.i functions as a comprehensive conversational layer that consolidates checking accounts, credit, cards, and insurance into a single environment.
The system also enables direct transactional capabilities, such as executing Pix payments via WhatsApp without requiring users to open the bank’s Superapp. To power these varied interactions, the bank utilizes a range of OpenAI models rather than a single architecture.
"OpenAI’s models were very important in delivering this experience, because they were able to operate across this range of solutions," said Carlos Eduardo Mazzei, CTO of Itaú, in an OpenAI case study. He noted that this multi-model approach balances the low latency required for simple tasks with the complex reasoning needed for grounded financial advice.
What the IARA platform does for enterprise AI governance
Scaling generative AI across 70 million accounts requires strict regulatory compliance and cost management. To address this, Itaú built the IARA governance platform, an internal orchestration layer that gives development teams controlled access to different models while enforcing internal policies. IARA manages three critical variables:
- Token cost control to ensure economic viability at scale
- Response accuracy for complex financial queries
- Application of responsible AI policies to mitigate bias
This architecture is already driving internal productivity. The bank’s legal department uses the platform to automate document analysis across a massive repository, executing 2 million tasks per month and achieving efficiency gains of 20% to 80%.
"IARA is a fundamental platform, because it ensures we can have governance and observability in a highly regulated environment." — Carlos Eduardo Mazzei, CTO, Itaú
Why Itaú accelerated its cloud and ML infrastructure migration
The ability to deploy IARA and ia.i at this scale is the direct result of a broader infrastructure overhaul that began in 2020. Facing bottlenecks with on-premises servers that took months to configure, Itaú migrated its machine learning workloads to the cloud.
By adopting Amazon SageMaker Studio, the bank standardized its environment for over 3,200 machine learning users, reducing model development time from six months to five days according to AWS. This strategic shift created the agile MLOps foundation necessary to iterate rapidly on generative AI applications. The timing of the current GenAI rollout aligns with a broader industry trend, as leading Brazilian banks are heavily investing in automation and cloud infrastructure to maintain competitive parity in digital banking.
What the ia.i rollout means for global banking AI strategies
Itaú’s strategy highlights a critical evolution in enterprise AI: the transition from isolated productivity tools to context-aware financial copilots. A study by Itaú and Grupo Consumoteca found that 80% of Brazilians want to better organize their financial lives, and 65% specifically want technology to act as an ongoing financial guide.
For IT professionals and business leaders in regulated industries, the primary takeaway is that model capability is secondary to orchestration. The success of ia.i relies less on the underlying large language models and more on the "Customer DNA" context layer and the IARA governance framework that safely applies those models. Enterprises attempting to scale GenAI must prioritize building similar orchestration layers to manage token economics and compliance before attempting mass consumer deployment.
As Itaú moves toward its target of full customer availability by late 2026, the integration of its internal ML pipeline with external GenAI models will serve as a benchmark for the financial sector. The transition from siloed chatbots to context-aware financial copilots demonstrates that infrastructure agility and strict governance are the true prerequisites for enterprise AI adoption.
