AI4 min read
How BBVA Scaled ChatGPT Enterprise: Lessons for Spanish Corporates
What BBVA did to grow from 3,300 to 11,000 ChatGPT Enterprise licences in a year, how other Spanish companies compare, and a practical roadmap for governance, training and hackathons.
Spanish corporates are moving fast from pilots to industrial-scale use of ChatGPT Enterprise. BBVA’s leap from 3,300 to 11,000 licences in a single year saved staff almost three hours a week on routine tasks and sparked an internal wave of innovation hackathons.
This post distils the key decisions, governance steps and cultural moves that let any large organisation replicate that success—while keeping data, compliance and people firmly in mind.
1. Why ChatGPT Enterprise (and not the public version)?
- Bullet-proof privacy: no training on your company data; SOC2 compliance; encrypted at rest and in transit.
- Unlimited, faster GPT-4 access and 128k context windows—vital for lengthy legal or technical documents.
- Admin console with SSO, usage dashboards and domain controls to manage thousands of seats.
2. BBVA’s journey in a nutshell
| Milestone | Date | What happened | Early impact |
|---|---|---|---|
| Partnership signed (3,300 licences) | May 2024 | Initial pilot across legal, risk and marketing. | 80% of users report >2 h saved weekly. |
| Expansion to 11,000 licences | May 2025 | Licences offered bank-wide; launch of AI community and “Bot Talent” idea challenge. | Avg. 2.8 h saved; 95% adoption in key units. |
What made the difference?
- Top-down mandate + grassroots community – the Global AI Adoption team set guard-rails, while a Teams channel let employees share prompts and mini-GPTs.
- Structured learning – “Data University” modules on prompt engineering and LLM ethics are mandatory before a licence is issued.
- Internal competitions – the DataRally and Bot Talent contests rewarded practical prototypes that now enter production pipelines.
3. What are other Spanish leaders doing?
| Company | Focus area | Notable move |
|---|---|---|
| Telefónica | Network ops | Built a central GenAI platform “Kernel” with Microsoft to embed GPT models in every business unit. |
| CaixaBank | Customer & employee experience | Second-wave rollout of generative AI use cases across the group, after early chatbot pilots. |
| Repsol | Legal & energy ops | First Spanish in-house legal team to use Harvey; >670 digital cases, 60% AI-enabled. |
| Santander | Training & SME focus | Open Academy MOOCs on ChatGPT open to staff and the public. |
| Inditex (Zara) | Retail & supply chain | AI chatbots in recruitment and demand forecasting to shorten collection cycles. |
4. A step-by-step roadmap to “land” ChatGPT
4.1 Set the ambition and the guard-rails
- Draft an AI policy covering PII redaction, copyright, hallucination risk and the AEPD’s guidance.
- Establish a multi-disciplinary AI Council (Tech, Risk, HR, Legal, Comms).
- Choose between Team (SMB) and Enterprise tiers; Enterprise is recommended for >250 seats or regulated data.
4.2 Secure quick-win use cases
| Function | Use case | Effort | Return |
|---|---|---|---|
| Legal | Draft NDAs & clause summaries | ★ | Hours saved/legal |
| HR | Craft inclusive job ads & screen CVs | ★★ | Faster hiring |
| Retail branches | Generate customer-friendly explanations of complex products | ★★ | NPS ↑ |
4.3 Build capability at scale
- On-board – licence request workflow tied to mandatory e-learning.
- Enable – internal Prompt Library with star-ratings and version control.
- Measure – weekly dashboard: licences active, prompts per user, time-saved estimate (BBVA tracks 2.8 h).
- Iterate – monthly showcase of best prompts and GPTs.
5. Designing a winning hackathon
| Element | Recommendation | Rationale |
|---|---|---|
| Format | 48 h hybrid sprint; teams of 4–6; mix of business, data and design profiles. | Encourages cross-pollination like BBVA’s DataRally. |
| Challenge tracks | Customer service, SME financing, fraud detection (mirrors BBVA Bot Talent). | Proven high-ROI areas. |
| Tooling | Sandbox ChatGPT Enterprise tenant + internal APIs; GitHub Copilot optional (Repsol HackIA precedent). | |
| Judging criteria | Impact (40%), Feasibility (30%), Responsible AI (20%), Demo quality (10%). | Aligns with board priorities. |
| Awards | Budget + licence top-up + seed funding to move MVP into production within 90 days. | Keeps momentum. |
6. Embedding for the long run
- Integration – connect ChatGPT via API to CRM, risk engines or knowledge bases using retrieval-augmented generation.
- Compliance & audit – log all prompts and outputs; keep data residency in EU regions if required.
- Continuous up-skilling – quarterly refresher courses as models evolve (e.g., expected GPT-5 upgrade).
- Innovation funnel – move the best hackathon ideas into a formal project portfolio.
7. Key takeaways
- Start small, think big – BBVA began with 3,300 seats but designed processes that scaled to 11,000 in 12 months.
- Treat adoption as a people programme as much as a tech rollout—train, certify and celebrate.
- Hackathons are not icing on the cake; they surface real-world problems and create champions who evangelise across the business.
- Measure relentlessly—hours saved, NPS uplift, risk reduction—to fund the next wave of licences.
Happy prompting—and buena suerte landing ChatGPT in your organisation!
Source:
BBVA amplía el acuerdo con OpenAI a 11.000 licencias de ChatGPT para sus empleados