At BBVA, employees have built custom GPTs for work across legal, risk, customer service, finance, and marketing, according to OpenAI. The bank also created a group of advanced users called “wizards” who run workshops and help colleagues find useful applications. That is the adoption moment executives should study: an employee has a real task, an approved tool, and a nearby person who can help connect the two. [1]
When a company buys enterprise AI seats, the purchase feels like progress. Usage dashboards fill up. Information technology checks the deployment box. But a seat count is an input, not an adoption plan. Access becomes a habit only when people know which work belongs in the tool, where to get help, and what rules govern its use.
BBVA’s rollout offers a sequence worth examining, although it does not prove that the same sequence will produce the same outcome elsewhere. OpenAI says the bank began in twenty twenty-four with a bounded ChatGPT Enterprise pilot for 3,000 employees. OpenAI now reports more than 100,000 employees using the service. These are vendor-reported figures, not independently audited results. [1]
Reuters separately confirms that BBVA has a relationship with OpenAI. It reported that the bank was among European companies OpenAI said it had granted access to its latest cyber models. That reporting does not confirm the adoption figures or show that the rollout caused a business result. [2]
The operating sequence matters more than the headline count. OpenAI says BBVA aligned security, legal, compliance, and technology teams from the beginning. Starting there gave employees an approved path for experimentation instead of asking governance teams to catch up after people had already chosen their own tools. The case study does not prove that early alignment caused adoption. It does show that governance was part of the rollout design rather than a later restriction. [1]
From that foundation, BBVA built an organization-wide champions network and the smaller group of advanced users it calls wizards. These people run workshops and surface applications inside business units. A license gives an employee access. A champion gives that employee someone who understands both the tool and the local workflow. That removes a common gap between a generic training session and the moment a person needs to decide whether AI belongs in a specific task. The network also gives program owners a place to collect recurring questions, compare applications, and direct training toward actual work. [1]
Senior leaders participated as users, too. OpenAI says BBVA trained 250 leaders, including the chief executive and chairman, and reports that executive committee members are among the most active users. These are vendor-reported statements. Their value is not ceremonial. Visible leadership use can answer a practical question for managers: is this a side experiment, or is the company prepared to change how work gets done? [1]
Employee-built tools provide another signal. OpenAI reports that BBVA employees created more than 20,000 custom GPTs. It says approximately 4,000 are used frequently. Both figures are vendor-reported and not independently audited. Frequent use matters because creation alone can reflect curiosity; repeated use suggests that at least some tools found a place in recurring work. It still does not prove return on investment, quality, or risk reduction. [1]
This is not an argument that licenses are irrelevant, that BBVA’s results transfer to your company, or that OpenAI is uniquely necessary. It is an argument about operating design. Enterprise adoption requires governed access, people who translate tools into local workflows, leaders who demonstrate serious use, and a way to distinguish experiments from tools that employees return to.
At the next executive review, run that structure audit before approving more seats. Ask who owns the champions network and its budget. Ask which leaders use the system in real work. Ask which employee-built tools are used repeatedly and who reviews them. Ask whether security, legal, compliance, and technology shaped the rollout before it scaled. Fund the missing structure first. More seats can wait.