I spent time with Antonio Bravo, BBVA's Global Head of Data and AI, and Elena Alfaro, Head of Global AI Adoption, to understand how one of the world’s most innovative banks moved from experiments to enterprise-wide AI adoption. Their approach is practical, metrics-driven, and centered on people. What stands out is how they combined leadership buy-in, hands-on training, rigorous measurement, and a culture of safe experimentation to scale AI across thousands of employees and dozens of business units.
🏦 What success looks like at BBVA
BBVA’s results read like a playbook for organizations that want real, measurable impact from AI. The bank began with targeted pilots and quickly expanded use across the company. The outcomes they report are concrete:
- High engagement: 83 percent weekly recurrence among users, indicating sustained adoption rather than one-off experiments.
- Time saved: roughly three hours per employee per week on average, a meaningful productivity gain at scale.
- Efficiency improvements: measured efficiencies of around 80 percent in many workflows.
- Scale of tools: more than 20,000 internal GPTs created, with 4,000 of those being used daily by hundreds or thousands of employees.
- Leadership participation: a dedicated training program for 250 leaders including the CEO and the chairman.
When Antonio described the rollout, the numbers were deliberate, not accidental. Deployments were planned, measured, and iterated. When Elena talked about culture, she emphasized psychological safety for learning and experimenting with AI. Together, those elements made the difference between pilots and practice.
🛡️ Building a safe place to learn and use AI
One of the first priorities BBVA set was to create an atmosphere where employees felt comfortable learning and using AI. Elena put it plainly:
"We created this atmosphere of being in a safe place to learn and to use AI."
That safe environment had several components:
- Clear guardrails and governance. People need to know what is allowed, what is off limits, and how to handle sensitive data. Governance should be accessible, not obtrusive.
- Sandbox environments. Give teams a place to try out prompts and workflows without risking production systems or compliance breaches.
- Supportive learning resources. Structured learning journeys, templates, playbooks, and internal forums where people can ask questions and share wins and failures.
- Hands-on coaching for leaders. Leadership training that focuses on use cases relevant to leaders' day-to-day work and decision making.
I found that their emphasis on safety did not mean stifling experimentation. Instead, safety was the foundation that allowed experimentation to happen at scale. By making rules clear and giving people the tools to experiment responsibly, they accelerated adoption while managing risk.
🚀 From pilot to thousands of users: the phased rollout
The progression BBVA followed is instructive. They started with a controlled pilot and scaled deliberately. Antonio summarized their deployment steps:
"We have deployed the product to initially 3,000 employees across many different countries, and then very quickly we jump to 11,000."
Key elements of their phased rollout:
- Start with early adopters and high-value use cases. Identify teams where the value is obvious—customer service, compliance, analytics, or operational workflows—then demonstrate quick wins.
- Use ambassadors. Convert early adopters into champions who teach and evangelize across teams. These ambassadors reduce the learning curve and multiply impact.
- Scale horizontally and vertically. Horizontal scaling brings the same capability to more teams; vertical scaling embeds the capability deeper into business processes.
- Invest in tooling and templates. Pre-built GPTs, prompts, and playbooks accelerate adoption. BBVA reports more than 20,000 GPTs in their ecosystem, with thousands in daily use.
- Measure and iterate. Monitor usage patterns, time saved, and qualitative feedback. Use those data points to refine guidance and prioritize next rollouts.
The jump from 3,000 to 11,000 users did not happen by accident. It came from a disciplined program of targeted pilots, rigorous support, and visible executive sponsorship.
📊 Measuring impact: how BBVA tracks ROI
BBVA doesn’t treat AI adoption as an IT project. They treat it as a business transformation and measure outcomes accordingly. Several metrics stand out:
- Usage frequency: 83 percent of users come back weekly, a sign of sustained value.
- Time savings: three hours saved per employee on average, which scales into substantial labor savings across thousands of people.
- Efficiency gains: many workflows saw efficiencies of around 80 percent when AI was properly integrated.
- Number of custom GPTs: more than 20,000 internal GPTs created, with 4,000 used daily.
Tracking these metrics required BBVA to operationalize measurement. They built dashboards to monitor adoption and outcomes and used both quantitative and qualitative indicators. The combination gave leadership confidence to expand, while providing teams with clear evidence of impact.
From my perspective, the most important lesson is to measure impact on work, not just on usage. Usage can look good on a chart, but time saved, error reduction, and customer satisfaction are the signals that matter.
👥 Leadership adoption and the culture shift
BBVA’s executive team did more than sign memos. They rolled up their sleeves. Elena explained how they prioritized leader training:
"We gave a specific training for 250 leaders including CEO and chairman so they really learned hands-on how to use ChatGPT on a journey that was built for themselves."
That hands-on approach accomplished several things:
- Leaders became role models. When the executive committee used AI, it signaled that this was strategic and safe.
- Leaders understood practical possibilities. Training focused on concrete tasks leaders face—synthesizing reports, preparing presentations, scenario planning—so they could directly see the value.
- Decisions accelerated. Leaders who experience the capability firsthand can make faster and more informed decisions about investment and governance.
Antonio highlighted how leadership activity translated to contagion across the organization:
"Our executive committee is mostly composed of very heavy users of ChatGPT so we see members of the executive committee that jump into a session with the teams full of ideas on what is it that they can do."
I noticed that leadership adoption was less about top-down mandates and more about leadership modeling. The effect is cultural: curiosity and practical experimentation moved from pockets to mainstream.
🤖 Operationalizing GPTs and scaling knowledge
One of the tangible assets BBVA built was an ecosystem of internal GPTs. These ranged from simple templates to custom assistants tailored to specific workflows. The scale was notable:
"We have deployed more than 20,000 GPTs in the organization and 4,000 of them are being used on a daily basis by hundreds and thousands of employees."
How did they make that work?
- Internal marketplace and discoverability. Employees needed a way to find useful GPTs. An internal catalog with categories, ratings, and recommended use cases helped surface valuable tools.
- Ownership and maintenance. Each GPT needed an owner to keep it updated, aligned with policy, and responsive to user feedback.
- Templates and starter kits. Reusable prompt templates and starter kits lowered the barrier for teams to create their own assistants.
- Governance embedded in the workflow. Policies, data handling guidelines, and review processes were part of the lifecycle of each GPT.
- Human-in-the-loop. For use cases requiring judgment, human review points ensured quality and compliance while still capturing efficiency gains.
The result was a living library of AI helpers that employees could adopt and adapt. That kind of internal platform approach converts one-off experiments into building blocks for broader transformation.
🔑 Lessons learned and practical recommendations
BBVA’s path yielded a set of pragmatic lessons that other organizations can apply. I distilled these into a practical checklist:
- Start with a clear problem set. Pick use cases where AI can deliver immediate value and measurable outcomes.
- Make learning safe. Provide sandboxes, clear policies, and a no-blame environment for experimentation.
- Train leaders hands-on. Leadership must experience the tools directly to enable fast decisions and cultural adoption.
- Measure outcomes, not vanity metrics. Track time saved, error reductions, customer impact, and operational efficiencies.
- Build a discoverable library of assets. Encourage reuse through an internal marketplace of GPTs, templates, and best practices.
- Embed governance in workflows. Make compliance and data protection a part of development and deployment, not an afterthought.
- Empower ambassadors. Convert early adopters into trainers and coaches to multiply adoption.
- Iterate fast and often. Use feedback loops to refine prompts, assistants, and deployment strategies.
When I outlined these recommendations back to Antonio and Elena, they emphasized one more point: align AI initiatives to business priorities. Technology alone does not deliver value. The real wins happen when AI is paired with domain expertise and clear objectives.
⚠️ Common pitfalls and how to avoid them
Scaling AI at this level comes with risks. BBVA faced and mitigated common pitfalls that often trip up other organizations:
- Overemphasizing technology over process. Deploying a tool without rethinking the process can create bottlenecks instead of efficiencies.
- Lack of measurable goals. If you cannot quantify impact, it is easy for initiatives to lose momentum and funding.
- Poor governance. Without clear rules for data access and model use, organizations expose themselves to compliance and reputational risk.
- Insufficient training. People need more than a how-to guide; they need contextualized, role-specific training and support.
- Underestimating change management. Introducing AI changes roles and workflows; plan for the human side of change.
I recommend addressing these pitfalls with straightforward actions: invest in change management, tie AI projects to KPIs, require data protection checks before deployment, and create role-based training programs that go beyond generic tutorials.
🔮 The future: what this scale enables
BBVA’s work shows that when adoption is broad and governed, the next phase is not simply more automation. It is deeper human-AI collaboration. With thousands of GPTs and heavy executive involvement, the bank is positioned to do more than save time. It can:
- Augment decision making. Leaders can synthesize scenarios faster and test strategies with AI-assisted modeling.
- Personalize customer experiences. With secure and compliant assistants, frontline employees can deliver more relevant services at scale.
- Unlock institutional knowledge. GPTs can codify processes, playbooks, and tacit knowledge so new hires and cross-functional teams can onboard faster.
- Continuously improve operations. Data from usage and feedback loops lets teams refine assistants and workflows in real time.
Antonio captured the sense of partnership that supports this future:
"You're serving hundreds of millions of users all over the world, a lot of companies. And then I feel, really I feel that you are there every day for us."
That sentiment reflects more than vendor satisfaction. It highlights the importance of ongoing collaboration between platform providers and adopters to maintain security, improve models, and surface enterprise-specific capabilities.
How to get started: an actionable 90-day plan
If you want a practical path inspired by BBVA’s rollout, here is a concise 90-day plan I would follow:
- Days 1–15: Assess and prioritize
- Inventory high-impact processes and pain points.
- Choose 3 to 5 pilot use cases with clear metrics.
- Days 16–45: Pilot and enable
- Launch pilots with cross-functional teams in sandboxed environments.
- Provide role-specific templates and hands-on training for pilots.
- Days 46–75: Measure and iterate
- Collect quantitative and qualitative metrics: time saved, error rates, user satisfaction.
- Refine prompts, workflows, and guardrails based on feedback.
- Days 76–90: Scale and formalize
- Onboard ambassadors and create an internal catalog of assets.
- Train leaders and embed governance into the deployment lifecycle.
- Communicate wins and next steps to maintain momentum.
This plan balances speed and prudence. It gets you into production quickly while preserving controls and learning from early deployments.
Small wins that compound
One of the things that struck me about BBVA’s approach is the focus on many small, compounding wins rather than a single monolithic transformation. When teams saved a few hours each week, those savings multiplied across departments. When leaders used AI to prepare more insightful briefings, decisions became faster and better informed. Those incremental changes accumulate into strategic advantage.
Creating a culture where experimentation is safe, outcomes are measured, and leadership participates creates a feedback loop. That loop accelerates improvement and builds internal momentum. The 20,000 GPTs BBVA amassed are evidence of that compounding effect—each assistant encodes a piece of institutional knowledge and productivity.
Final thoughts
Scaling AI across an organization is as much an organizational design problem as it is a technical challenge. BBVA’s model combines governance, leadership engagement, hands-on training, measurable outcomes, and an internal library of reusable assets. The result is not just faster work but transformed work: smarter decisions, better customer interactions, and knowledge that travels across the company.
If you are leading an AI adoption program, consider the following concise checklist inspired by BBVA:
- Make learning safe: sandboxes, clear policies, and supportive culture.
- Measure what matters: time saved, quality improvements, and customer value.
- Train leaders: hands-on journeys for executives and managers.
- Build reusable assets: templates, GPTs, and an internal marketplace.
- Embed governance: integrate compliance into workflows, not afterthoughts.
What impressed me most was how BBVA balanced ambition with discipline. They did not treat AI as a novelty. They treated it as a capability to be grown, measured, and governed. The result is a scalable, sustainable approach that other organizations can learn from and adapt.
