How Revolut PRAGMA AI is Dismantling Fragmented Banking Tech

For years, traditional banks have relied on fragmented AI tools that treat fraud detection, credit risk, and customer service as isolated silos, leaving massive gaps in security and personalization. The UK-based digital bank Revolut is dismantling this outdated approach with Revolut PRAGMA AI, a proprietary foundation model designed to understand financial behavior holistically. By treating every transaction, app tap, and support request as a connected signal, this unified system is setting a new benchmark for enterprise AI in finance.

Instead of applying a generic language model to financial data, Revolut built a system that learns patterns across the entire customer journey. Trained on 40 billion events and interactions from 25 million users, PRAGMA eliminates the need for separate engineering pipelines. This means the AI can instantly connect a user's bill payment habits with their investment activity to form a complete behavioral profile.

According to a recent report, this unified architecture is powered by 200 NVIDIA H100 GPUs. This hardware backbone has allowed the platform to scale massively, supporting a jump from 38 million users in 2023 to over 70 million today. Because learnings from one task automatically improve performance across the board, the model adapts in real-time to shifting economic and behavioral trends.

The shift from isolated tools to a single shared foundation model has yielded concrete performance leaps. PRAGMA has delivered a 64.7 percent improvement in detecting and stopping fraudulent activity. By analyzing the entire ecosystem of a user's behavior rather than just isolated transactions, the system identifies anomalies that traditional fraud models miss.

Beyond security, the model has driven a 16 percent increase in credit risk prediction accuracy and made the bank 41 percent more effective at recommending relevant financial products. This compounding intelligence means that a pattern detected in a user's investment activity can instantly inform their credit profile or trigger a proactive fraud alert.

This unified data infrastructure also powers Revolut's customer service AI assistant, which currently resolves 75 percent of support requests without any human intervention. By tapping into the same foundation model used for fraud and credit, the assistant possesses deep context regarding the user's financial state.

Taking automation a step further, the company recently launched its first customer-facing agentic AI system, AIR (AI by Revolut), for its UK customers. Designed as a comprehensive lifestyle companion, AIR goes beyond answering questions to take autonomous actions - such as managing subscriptions, canceling lost cards, assisting with budgeting, and even organizing travel arrangements.

The deployment of Revolut PRAGMA AI exposes a critical vulnerability in legacy banking: maintaining separate data pipelines for specific tasks is rapidly becoming a financial and operational liability. While traditional institutions spend heavily on maintaining disjointed AI stacks, Revolut’s unified model creates a flywheel effect where every customer interaction makes the entire system smarter.

Looking ahead, the 64.7 percent leap in fraud detection alone will likely force incumbent banks to rethink their enterprise AI strategies. As agentic tools like AIR become the standard for consumer expectations, banks that fail to integrate their data into a single foundation model won't just lose out on operational efficiency - they will fundamentally lose the ability to compete on personalized customer experience.