Revolut Launches 'Revolut Research' AI Division to Pioneer Proprietary PRAGMA Foundation Model

Revolut has launched Revolut Research, a dedicated machine learning division designed to move the fintech beyond standard software integrations toward proprietary foundation models. For fintech professionals, this signals a shift from using generic AI wrappers to building sovereign financial intelligence, potentially redefining how institutions manage credit risk and fraud detection at scale.

What was announced

Revolut Research is a new unit within Revolut’s AI Department, tasked with developing next-generation machine learning architecture. The division’s primary output is PRAGMA, a proprietary foundation model built in partnership with NVIDIA. Unlike traditional banking setups that rely on third-party software, PRAGMA is engineered to decode financial behaviours natively within Revolut’s core engine.

The unit aims to provide a unified behavioural intelligence layer across Revolut’s global operations, which currently serve 80 million retail customers in over 40 markets. By processing billions of cross-border transactions, the models are designed to improve real-time risk assessment and product recommendations. Early testing of PRAGMA on historical data has yielded significant performance benchmarks:

• Credit Risk: A 2.3x increase in accuracy for identifying default risks.
• Fraud Prevention: A 65% increase in caught fraud cases, paired with a 17% improvement in alert precision.
• Personalisation: A 41% rise in the relevance of product recommendations for both retail and business users.

Revolut Research plans to share its findings with the wider scientific community, including upcoming appearances at NVIDIA GTC Berlin in October and ICAIF in November. The division will also host quarterly meet-ups at Revolut’s offices and intends to open-source specific technical frameworks.

"To lead the future of intelligent banking, you cannot rely on third-party blueprints. We have launched Revolut Research to institutionalise our 'build, don't bolt on' philosophy. 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."

Pavel Nesterov, Head of AI at Revolut.

The companies involved

Revolut is a global financial leader that has scaled rapidly from its origins as a digital-only challenger to a platform serving 80 million customers. The company operates in a highly regulated environment, overseen by bodies such as the Financial Conduct Authority (FCA) and the Prudential Regulation Authority (PRA) in the UK. Its market position is defined by a high-velocity product release cycle and a focus on cross-border financial services.

Nvidia, the partner for the PRAGMA model, is the dominant force in the global semiconductor and AI hardware market. By collaborating with Nvidia, Revolut is tapping into the high-performance computing power necessary to train large-scale foundation models on massive datasets. This partnership places Revolut in a distinct category compared to traditional retail banks, as it adopts the infrastructure strategies typically seen in pure-play technology giants rather than legacy financial institutions.

What this means

This move highlights a growing divide in the industry between "tech-first" firms and legacy banks that remain tethered to vendor-supplied AI solutions. By building a proprietary foundation model, Revolut is attempting to bypass the "technological wall" that occurs when off-the-shelf AI is retrofitted onto aging core banking systems. This puts significant pressure on traditional competitors who may struggle to match the speed and precision of a native intelligence layer.

The success of this initiative raises critical questions for the sector: will proprietary models become a prerequisite for maintaining competitive margins in fraud and credit? Furthermore, as Revolut begins to open-source its frameworks, the industry must decide whether to adopt these new standards or continue developing fragmented, specialised models that may lack the same cross-market scale.