Future AI Pulse
Itaú’s next AI expansion is a rollout goal, not a completed launch to every customer. An OpenAI case study published October 16 says the Brazilian bank’s conversational experience, ia.i, reached roughly 300,000 customers in its first phase. Availability across the customer base is a goal for the end of 2026.
The distinction matters for anyone assessing enterprise AI: an initial deployment, a wider customer context system and a future availability target describe different things.
What OpenAI’s case study reports
According to the official account, ia.i combines text and voice interactions across banking services, with access to human support. Behind it, the bank’s IARA platform governs access to multiple model providers, including OpenAI, and supports monitoring and cost controls.
The source also describes a financial context layer called Customer DNA covering approximately 70 million customers. That figure is not the number of people reported to have received ia.i in its first phase.
For an internal legal solution, the article reports a repository exceeding 90 million documents, more than two million tasks monthly and efficiency gains ranging from 20% to 80% in process stages. These are figures in a vendor customer story, not independently audited results presented by Future AI Pulse.
Why the denominators matter
A bank can have a large customer information system while exposing a new conversational product to a smaller group. Combining those numbers into a headline about tens of millions already using the assistant would erase the rollout boundary.
The same care applies to efficiency. A gain in one document processing stage does not establish an equivalent reduction in an entire department’s workload. Reading, checking, escalation and final decisions can consume different amounts of time. Without a disclosed baseline and measurement method, the reported range is evidence of a claimed improvement, not a forecast another organization can copy.
Our coverage of Stack Overflow’s plugin growth makes a related distinction between entering a product and completing useful work. Both stories benefit from keeping the unit of measurement visible.
A useful rollout brief for product teams
The following is an editorial framework for evaluating an assistant deployment, not a description of undisclosed Itaú controls.
- Define availability. Record who can access the experience, which channels work and which tasks are included. Separate an invitation, an active user and a completed task.
- Measure the whole journey. Track time to a correct resolution, repeat contacts, failed attempts and human handoffs alongside model response speed.
- State the action boundary. Make clear whether the assistant explains information, prepares an action or executes one after confirmation. Users should understand the result before committing.
- Test exceptions. Include ambiguous requests, missing context and cases requiring a specialist. A fluent response alone is not a successful resolution.
- Keep the rollout reversible. Define who can pause a capability and what users see if it is unavailable. A fallback should remain understandable during an incident.
What to look for next
The meaningful follow up is evidence of broader availability and measured customer outcomes as the rollout progresses. Useful disclosures would specify the population, observation period, task mix and rate of successful resolution. Those details would help distinguish scale of access from quality of service.
For now, the story is a substantial initial deployment with a stated expansion ambition. It does not demonstrate that every customer already has the experience, or that reported internal efficiency gains predict customer outcomes.
Sources and image credit
OpenAI: How Itaú is building generative AI for its entire customer base, published and consulted October 16, 2026. Reporting above is attributed to this primary vendor case study; the evaluation framework is Future AI Pulse analysis.
Cover: Kevin Ache on Unsplash, under the Unsplash License. Illustrative server photograph, not an Itaú facility or product screenshot.