From oil and gas exploration and refining to renewables and a retail network serving millions of customers, Repsol operates across the full energy spectrum in more than 20 countries. Eight years of digital transformation delivered more than a billion euros in cumulative value and over 1,000 AI initiatives in production. But as the appetite for data-driven decisions outpaced the analysts available to support them, the company needed a way to put and test that capability directly in the hands of the people closest to the work. With Genie and Genie Code on the Databricks Platform, business users now ask questions in plain language, build governed models through conversation and act on insights without waiting in a technical queue.
The data was there, but most people still couldn’t reach it
As Repsol navigated the changing landscape toward a multi-energy future that balances affordability, sustainability and reliability, it bet early that digitization would be the lever that made that balance possible on a scale. That bet paid off. The company’s digitization efforts paved the way for AI to run in production across exploration, refineries, service stations and renewables.
Algorithms guide drillers on speed and direction underground, while computer vision monitors wildlife around wind farms to keep birds safe. Two waves of digitization delivered a meaningful impact on free cash flow, and the company cataloged more than 1,000 AI initiatives. But this success exposed a more human bottleneck. In a 25,000-person organization, more people wanted to use data in the flow of work, and they could not wait for every question to become a formal project.
“If every single initiative takes years to get the data, forget it,” said Juan Jose Casado, Chief Digital Officer at Repsol. “Scaling is not going to happen.”
Repsol had built a strong data foundation on the Databricks Platform, with Unity Catalog governing who could access what and where every dataset came from. But having the data in one place did not mean everyone could use it. Most day-to-day questions still required an analyst, an IT request or someone comfortable writing code, and 80% of the company’s ad hoc reports were barely used after their first few runs.
In Repsol’s refineries, soft sensors use machine learning to predict product quality in real time rather than waiting a week for physical lab results. These sensors speed up production, improve quality control and help maximize margins. But the plant experts who understand the chemistry are not the same people who know how to build machine learning models. Repsol needed a way to let them work in their own language while still meeting the company’s standards for accuracy and governance.
Repsol opens its data foundation to the people who need it most
With a governed data foundation already in place, Genie gave Repsol a way to open it up. The team mapped everyday business context to the underlying data so that, during the exploratory stage, when an employee wants to explore and asks a question in natural language, Genie knows exactly where to look for the answer. Users receive visualizations, create ad hoc KPIs on the fly and explore the factors behind a result without specialist help.
“Our people can increasingly ask a business question and get an answer on their own, with much less reliance on analysts, SQL or waiting,” Juan Jose said. “That means our technical teams stop building reports nobody uses more than once and start focusing on the work that actually needs them.”
Genie Code takes the same idea a step further, from answering questions to building predictive models. Rather than generating open-ended code, Repsol connected Genie Code to its own data science libraries and methodology, so that a refinery expert can build a soft sensor through conversation without needing to understand the underlying code or data model. The agent walks them through variable selection, model testing and accuracy validation while enforcing the company’s standards at every step.
“We spent years defining the right way to do data science at Repsol. Genie Code lets anyone follow that process through a conversation. The citizen data scientist is no longer just a concept. It’s becoming how we work,” Juan Jose said.
Thousands of hours saved and a cultural shift toward self-service AI across the business
More than 2,000 potential users already use the Databricks Platform, and Repsol sees each of them as a candidate for self-service analytics through Genie. Most of the questions these users ask are temporary by nature, one-off explorations rather than something that belongs to a permanent dashboard. With Genie absorbing that ad hoc work, analysts and IT teams are free to focus on the governed, recurring reporting that the business actually uses again and again.
By putting model building in the hands of the people who understand the process best, Genie Code is saving an estimated 4,000 hours annually. But the bigger cultural shift is that teams that once saw AI as something built by specialists are starting to see it as something they can do themselves, guided by an agent that enforces Repsol’s standards at every step.
Genie and Genie Code also fit into a broader strategic move. Repsol is building its own agentic platform and developing dozens of agents across the business, and it sees these capabilities as part of that shift. For Juan Jose, scaling AI in an industrial company has always come down to removing the same barriers: getting data to people fast enough, making models easy to deploy and maintain, keeping governance tight and ensuring the tools actually reach the people making decisions every day. With Databricks, he believes those barriers are a thing of the past.
“Companies are potentially gaining a kind of power they have never had before,” Juan Jose said. “The barriers between our people and data-driven decisions are increasingly being erased. Now we want to scale AI further than we ever have.”