Revolut's Proprietary AI Model PRAGMA Unifies Banking Intelligence

A unified approach to banking AI

Revolut is building its own foundation model for financial services rather than deploying separate AI tools for different functions. The proprietary system, called PRAGMA, treats every customer interaction—transactions, app navigation, investments, bill payments, and support requests—as interconnected signals within a single learning framework.

This architectural choice distinguishes Revolut from competitors that typically use off-the-shelf AI solutions or develop isolated models for specific tasks like fraud detection or credit scoring. PRAGMA is designed to understand financial behavior holistically, allowing insights from one area to improve performance across the entire customer experience.

The model runs on NVIDIA GPU infrastructure and currently serves Revolut's 70 million users across multiple markets.

Measurable performance gains

Revolut reports concrete improvements from the unified model approach. Fraud detection accuracy increased 64.7 percent compared to previous systems. Credit risk prediction improved 16 percent, and product recommendation relevance rose 41 percent.

These gains stem from PRAGMA's ability to recognize patterns that span multiple banking functions. A transaction flagged as unusual gains additional context from the customer's app usage history, investment behavior, and past support interactions. This cross-functional learning creates a more complete picture than models trained on isolated data sets.

The system also enables what Revolut describes as "agentic AI"—automation that can take actions on behalf of customers based on learned preferences and behavioral patterns, though the source material does not detail specific implementations.

Why it matters

Financial services generate enormous volumes of structured data across clearly defined processes, making them ideal testing grounds for enterprise AI. Revolut's approach demonstrates that proprietary foundation models trained on domain-specific data can outperform general-purpose tools adapted for banking. For technology leaders in regulated industries, this represents a strategic choice: build specialized AI infrastructure or rely on vendor solutions. The performance differentials Revolut reports suggest that investment in custom models may deliver competitive advantages that generic tools cannot match, particularly in environments where cross-functional data integration creates compounding value.

Setting an enterprise AI standard

By consolidating AI capabilities into a single foundation model, Revolut gains organizational agility. Updates and improvements to PRAGMA benefit all banking functions simultaneously rather than requiring separate development cycles for each use case. This unified architecture also simplifies compliance and governance, as the company maintains one model to audit and explain rather than dozens of disconnected systems.

The approach positions Revolut to deliver increasingly personalized services as the model continues learning from customer interactions. As financial institutions worldwide evaluate their AI strategies, PRAGMA offers a blueprint for how proprietary models can create systemic advantages beyond point solutions.

These details were first reported by Bernard Marr in Forbes.