Skip to content

Global

Artificial intelligence at service stations

How Repsol uses artificial intelligence at its service stations

From demand forecasting to automatic delivery note reading: applications that help improve product availability and customer service.

Repsol uses predictive, analytical, and generative artificial intelligence to anticipate store demand, propose orders, automate delivery note reading and simplify customer service management. The suggested order system is already operational in over 800 stations, it reaches 85% accuracy and generates more than 100,000 product suggestions per month. These tools are not a substitute for teams: they help them decide better and spend more time with the customer.

You stop to refuel and enter the service station store to get food. And there it is, on the shelf, just the one you were looking for. There is no shortage of items or excess product that is about to expire. What's behind this ‘small’ success? How does a service station anticipate what it will sell every day? The answer lies in different applications of predictive, analytical, and generative artificial intelligence that help anticipate demand, simplify processes and spend more time on what really matters: better customer service. 

In retail, AI can add value in sales forecasting, order management, and customer support. Repsol already applies these capabilities in projects with varying degrees of maturity; some are deployed at scale and others remain in the pilot phase.

What AI applications does Repsol use at service stations?

The included initiatives combine different AI approaches. Suggested ordering uses predictive and optimization algorithms to calculate replenishment needs, while automatic delivery note reading uses generative AI to interpret documents and convert their content into structured data.

Solution

Type of AI

What it doesScope or Result
Suggested orderPredictive AI and optimization.Proposes what items to restock and how many.More than 800 stations; 85% accuracy and 14% acceptance with adjustments.
Automatic delivery note readingGenerative AI and document processing.Extract date, supplier, items, and units from a photograph.Testing phase in some stations; the operator validates the result.

Predictive AI to anticipate demand and optimize orders

The project is active  in over 800 stations participating in the program, sharing the same goal as suggested fuel ordering: freeing up time to deliver greater value to customers. 

The system does not replace the manager's criteria: it offers a recommendation that can be accepted or adjusted depending on the specific situation of the station. 

  • Around 1,000 items per station, with orders placed two or three times a week. 
  • About 900 forecasting models and another 900 optimization models run every day. 
  • More than 100,000 product suggestions generated per month.

Like any predictive model, the system improves as it incorporates new data and learns from decisions made at each station.

A total of 85% of recommendations are accepted unchanged, and an additional 14% are accepted with quantity adjustments only.

For the customer, the goal is to reduce stock outs and increase product availability. It allows orders to better match actual sales and reduces waste, especially on fresh food.

Artificial intelligence makes it possible to anticipate patterns and improve decision-making, but it does not eliminate uncertainty. Unexpected changes in demand, logistical incidents, or extreme weather events continue to require the expertise and judgment of the teams.

Generative AI to read delivery notes and reduce administrative tasks

Repsol is testing a generative AI system that streamlines delivery note registration in select stations. The employee takes a photo of the document with their mobile phone and the tool identifies the date, supplier, items, and quantities to upload the information into the system.

The goal is to reduce manual data entry, decrease transcription errors, and free up time for higher value customer tasks.

An operator reviews and validates the result. This human supervision allows combining the speed of automation with the operational know-how of the teams.

The project remains in the testing phase at some stations. Therefore, its results should be interpreted as those of a pilot and not as a widespread deployment.

Benefits, limits, and human oversight

The applications described offer different benefits, but share a common goal: using data and automation to improve service without replacing human judgment. 

  • For the customer: higher product availability, faster service processes, and more relevant communications. 
  • For employees: less time spent on data entry or performing repetitive tasks. 
  • For the station: orders better aligned with demand and a lower risk of waste. 
  • For decision making: recommendations based on data that must always be validated with operational context. 

AI does not eliminate uncertainty or replace team expertise. Its role is to support decision-making, automate repetitive tasks, and allow staff to dedicate more time to what truly makes a difference. 

The next time someone enters a station looking for a cold drink or a specific product, they will probably only see a well-stocked shelf. What they won’t see is the artificial intelligence behind making that small everyday gesture so seamless.

FAQs about artificial intelligence at service stations

Repsol uses predictive and optimization models to anticipate demand and suggest orders, as well as generative AI to extract information from delivery notes. It also uses AI analytics to study interactions and adapt communications or promotions.

No. The suggested ordering system proposes what to restock and in what quantity, but the manager can accept or adjust the recommendation. Human judgement remains essential for unforeseen changes or local situations.

As of July 2026, the system was deployed in over 900 stations participating in the program.

No. The generative AI system is currently in the testing phase in some stations, and an operator always validates the extracted information.

Greater product availability, fewer stock-outs, simpler transactions, and communications or promotions that are more relevant to their context.