Revolut is making a direct argument that the rest of fintech is doing AI wrong. The company has launched a dedicated AI research division called Revolut Research, and alongside it came a pointed critique of competitors who rely on third-party AI tools or build separate models for separate problems. That’s a shot at practically every major bank and most neo-banks operating today.
The new unit will sit at the center of Revolut’s proprietary AI and machine learning work, with plans to collaborate with academic institutions and technology partners. The real substance here, though, is PRAGMA, Revolut’s own foundation model built specifically for finance. Unlike vertical-specific models that handle fraud in one silo and customer service in another, PRAGMA is designed to learn across the entire customer journey as a single system. Fraud detection, credit decisions, product recommendations, and customer support all run from the same underlying model.
PRAGMA was built in partnership with Nvidia, and Revolut is already citing early results: 65 more fraud cases caught, 2.3x higher accuracy on credit default risk identification, and 41 percent more product recommendations generated. Those are specific numbers, which is worth noting, because most AI announcements in financial services stay deliberately vague on performance metrics. Revolut is putting stakes in the ground here.
The scale argument is also central to what Revolut is claiming. With 80 million customers generating billions of cross-border transactions in real time, the company is betting that data volume alone creates a compounding advantage. The more transactions the model sees, the better it gets at fraud patterns and risk signals. That’s not a new idea in machine learning, but it does matter at Revolut’s scale. Few fintech companies can match that data footprint, and traditional banks sit on comparable volumes but have historically struggled to activate that data in modern AI pipelines.
The competitive framing is deliberate. Building proprietary foundation models is expensive and slow. Most companies reach for OpenAI, Google, or Anthropic APIs and call it an AI strategy. Revolut is explicitly rejecting that path, and the launch of a formal research division signals this is a long-term infrastructure investment, not a product sprint.
For developers and founders watching the fintech AI space, the key question is whether PRAGMA’s early numbers hold at production scale. But the broader signal is clear: Revolut is positioning AI as core infrastructure, not a feature layer. That distinction is going to matter more as the competitive gap between banks with proprietary models and those without starts to widen.
