OpenAI published a BBVA customer story on June 11, 2026 showing ChatGPT Enterprise at regulated-bank scale. The headline figure is more than 100,000 BBVA employees using ChatGPT Enterprise globally, after an initial 2024 deployment to 3,000 employees.
The more interesting detail is how BBVA scaled it. OpenAI says the bank wrapped adoption in governance, security, legal review, structured training, an AI champions network, and senior-leader enablement. That is the operational pattern regulated companies care about: broad access only works when the institution can control how people use it.
The bank scaled through governance first
BBVA’s adoption story starts with a constraint that many AI pilots skip: the company is a global financial institution operating under strict risk, privacy, compliance, and audit expectations. OpenAI says BBVA aligned security, legal, compliance, and technology teams from the beginning.
That framing matters because the alternative in large companies is often unmanaged consumer-tool usage. BBVA’s approach was to provide an approved enterprise environment, then train people into it. OpenAI says the bank created an organization-wide AI champions network and advanced users called AI “wizards” who run workshops, help teams integrate ChatGPT into workflows, and identify use cases.
This is a distribution story as much as a model story. The model creates the capability, but the operating model decides whether the bank can actually use it.
Internal GPTs are the expertise layer
OpenAI says BBVA employees have created more than 20,000 custom GPTs across legal, risk, customer service, finance, and marketing, with about 4,000 used frequently. That gives the deployment a different shape from a simple seat-count rollout.
The examples are practical. OpenAI says BBVA built a Credit Analysis Pro GPT for extracting and analyzing unstructured information from annual reports, ESG disclosures, and media coverage. In legal services, the bank created a Retail Banking Legal Assistant GPT to help answer roughly 40,000 annual client-related legal inquiries from branch managers. In Mexico, a client-experience assistant analyzes open-ended survey responses.
Those numbers also show the management problem. Thousands of internal assistants can create local productivity gains, but they also need ownership, review, retirement, and access control. A bank cannot treat every internal GPT as a harmless document.
The productivity numbers need careful reading
OpenAI’s results section says BBVA has more than 70% weekly active usage across deployed employees, saves about 3 hours per employee per week, and sees up to 80% efficiency gains in selected workflows. The clearest example is in Peru, where OpenAI says more than 3,000 employees use an internal AI assistant that reduced average query handling time from about 7.5 minutes to about 1 minute.
That is useful evidence, with a caveat. The “up to 80%” figure is not a bank-wide productivity number. It is a selected-workflow metric. The distinction matters because regulated enterprise AI will be judged by task-level proof first, then broader business outcomes later.
The bank’s public story is still meaningful. It suggests that ChatGPT Enterprise can clear security and governance thresholds in a large financial institution, and that employees will build domain-specific assistants when they are given a sanctioned environment.
The Eight is the strategic frame
OpenAI says BBVA’s broader AI roadmap is called “The Eight,” covering customer experience, commercial banking, risk, operations, software development, and employee productivity. The name matters less than the shift in scope: BBVA is framing AI as bank redesign, not as a collection of isolated productivity pilots.
That ambition raises the bar. A writing assistant can be measured by time saved. AI inside credit risk, customer engagement, legal response, or operations has to be measured by quality, compliance, customer outcomes, and auditability.
What to watch next
The next checkpoint is whether BBVA publishes outcomes beyond usage and selected efficiency gains. Useful evidence would include error rates, customer-service quality, risk-analysis cycle time, audit results, or software-delivery metrics.
The second checkpoint is how the bank manages its GPT estate. Twenty thousand internal GPTs can be an innovation signal, but the durable advantage will come from the smaller set that is maintained, measured, secured, and tied to real workflows.
For broader model context, see our AI model leaderboard. For related company coverage, see OpenAI.
