According to a recent LinkedIn post from Revolut, nine of the company’s researchers, working with NVIDIA, contributed to a scientific paper on applying artificial intelligence to banking data. The post highlights that, unlike most financial AI models focused on text, this work targets long and complex sequences of banking events such as transactions, app actions, and trades.
Claim 55% Off TipRanks
- Unlock hedge fund-level data and powerful investing tools for smarter, sharper decisions
- Discover top-performing stock ideas and upgrade to a portfolio of market leaders with Smart Investor Picks
The post suggests that Revolut’s team developed a purpose-built foundation model, named PRAGMA, designed specifically for banking event sequences rather than adapting generic models. Trained on 24 billion events, PRAGMA is described as enabling improvements in credit scoring, fraud detection, and product recommendations within a single architecture.
As shared in the LinkedIn post, the paper detailing PRAGMA has been made open to the broader industry, while its creators are employed at Revolut. For investors, this may indicate Revolut’s intent to position itself as a technical leader in AI-driven banking infrastructure, potentially enhancing risk management, personalization capabilities, and operational efficiency relative to digital and traditional banking peers.
If PRAGMA delivers materially better fraud detection and credit risk assessment, Revolut could see lower credit losses and fraud-related costs over time, which would be supportive of unit economics. In addition, more accurate and timely product recommendations could increase customer engagement and cross-sell rates, strengthening revenue per user and improving the platform’s competitive differentiation in a crowded fintech market.
The collaboration with NVIDIA, as referenced in the post, may also signal deeper strategic ties with a leading AI hardware and software provider, which could help Revolut scale and optimize its AI workloads. However, any financial impact will depend on real-world performance, regulatory acceptance of AI-driven decisioning, and whether competitors adopt similar or superior models, especially given that the underlying research is being shared openly with the industry.