Revolut launches AI research unit to build proprietary models

At the center of the unit's work is PRAGMA, a proprietary foundation model developed with Nvidia. Revolut says PRAGMA is designed to understand complex financial behavior and powers functions including fraud detection, credit-risk assessment, platform operations and personalized product recommendations. Early deployments, the company said, delivered 2.3 times better credit-risk accuracy, detected 65 percent more fraud cases and generated 41 percent more relevant product recommendations.

Sifted reported that the AI arms race among Europe's neobanks has been heating up over the past year. The announcement follows UK digital bank Starling's launch last week of smart tools partly inspired by continuous model rollouts by frontier large-language-model companies. Dutch rival Bunq responded in December by upgrading its AI customer-support assistant, and by March Starling had launched what it described as the UK's first agentic AI assistant. Revolut followed with its own assistant in April.

Pavel Nesterov, Revolut's head of AI, was quoted by Sifted as saying that to lead the future of intelligent banking the company could not rely on third-party blueprints. "We have launched Revolut Research to institutionalise our 'build, don't bolt on' philosophy," he said. "By training native foundation models on our global operational data, we are giving our engineering teams an unprecedented engine to deploy smarter features faster, eliminate systemic friction, and give our customers a safer, radically better financial experience."

Revolut's data from 80 million customers feeds into the model, creating what the company described as a self-reinforcing loop in which larger datasets make the models smarter. The research unit plans to regularly publish scientific findings and open-source technical frameworks, and to engage with the wider scientific community through international conferences and meetups at Revolut's offices.