Revolut launches AI unit as proclaims virtues of proprietary AI

Revolut has launched Revolut Research, a new unit focused on AI work across financial services. The pitch is blunt: build the stack itself, lean less on outside tools, and treat proprietary models as the core of the business rather than a bolt-on feature.

The digital bank says the new unit will sit at the centre of its AI and machine learning efforts and will work with academic and tech institutions. It’s also pushing one model for finance, rather than splitting the job into separate systems for fraud, customer service and other tasks.

That model is called PRAGMA. Revolut says it was built with Nvidia and that early tests are encouraging: 65 more fraud cases caught, 2.3 times higher accuracy in spotting credit default risk, and 41 per cent more product recommendations.

The bigger argument here is scale. Revolut says it has 80m customers and billions of cross-border transactions flowing through the platform in real time, which it believes gives its models an edge that traditional banks will struggle to copy. As the dataset grows, the company says, the system gets better at fraud detection, risk evaluation and predicting what users need next.

Pavel Nesterov, Revolut’s head of AI, framed the whole thing as a rejection of third-party templates. The company wants its own foundation models trained on its global operational data, with the promise of faster feature deployment, less friction and, naturally, a safer banking experience.