Revolut puts its PRAGMA transaction foundation model into production for 80 million customers

Revolut puts its PRAGMA transaction foundation model into production for 80 million customers

Revolut is rebuilding its technology stack around PRAGMA, a proprietary foundation model that works at the level of individual transactions. The model is now live in production for an initial 80 million customers. The source describes this as a large-scale enterprise deployment of a domain-specific model, not an integration of a general-purpose third-party LLM.

This matters because most fintech AI so far has been rules engines and task-specific ML models, often with a licensed LLM added on top for customer-facing features. Revolut is going a different way. It is training a model on its own transaction data and making that model the base the rest of the stack is built on. That is a test of a contested idea in financial-services AI: whether proprietary behavioral data, at enough scale, can support a foundation model that a general model or an incumbent platform vendor can't easily copy. An initial footprint of 80 million customers makes this one of the larger production deployments of a domain-specific foundation model reported in finance, rather than a pilot.

For builders and investors, the takeaway is that the data owner may be the strongest model builder in vertical finance. Neobanks and payment companies with large transaction histories can choose to build in-house instead of buying from AI vendors. Startups selling transaction-intelligence layers to these firms could find their market shrinking to smaller institutions that lack the data volume to train their own models. The source does not disclose PRAGMA's architecture, its specific use cases, or performance metrics, so it is too early to judge how much better it is than conventional ML.

#Revolut #FoundationModels #FintechAI #DomainSpecificAI #EnterpriseAI #Neobank