Revolut launches AI research unit for banking models

Revolut has launched Revolut Research, a dedicated artificial intelligence division within its broader AI Department.

The new unit is intended to support machine learning across financial services and serves as the institutional base for PRAGMA, a proprietary foundation model developed with NVIDIA.

PRAGMA is designed to handle tasks including risk assessment, platform operations and product recommendations. Revolut is building the model on data generated from its retail and business operations in more than 40 markets.

The company says it serves more than 80 million retail customers worldwide, who make more than a billion transactions a month. That gives the business a large pool of financial and behavioural data for model training.

Initial internal results released by the fintech suggest the model has improved performance in several core banking functions when tested on historical data. Early deployments identified credit default risk with 2.3 times higher accuracy than legacy baselines, caught 65% more fraud cases and generated 41% more relevant product recommendations.

Those figures highlight a broader push by fintechs and banks to apply machine learning to underwriting, fraud detection and customer engagement. Many financial groups have adopted third-party generative AI tools, but some are now trying to build systems trained on their own data in areas where precision and speed affect both costs and customer outcomes.

Internal build

Revolut presented the new unit as part of a broader strategy to develop in-house systems rather than rely on external software layers. The division will provide a central structure for proprietary AI deployments and machine learning initiatives across the business.

PRAGMA is intended to create a single model that can interpret financial behaviour across different products and markets. That contrasts with the more common approach of using separate models for narrower tasks such as fraud screening, credit scoring and customer support.

Beyond PRAGMA, Revolut says its security models already review nearly one billion transactions each month to detect fraudulent activity. It also pointed to AIR, its in-app assistant in the UK, as an example of machine learning in customer-facing products.

The launch comes as financial institutions face pressure to modernise ageing technology stacks while meeting stricter demands on fraud controls, service quality and efficiency. For challenger banks and fintechs, internal AI development can also help distinguish their systems from rivals using the same external tools.

Leadership comments

Pavel Nesterov, Head of AI at Revolut, outlined the rationale behind the move.

"To lead the future of intelligent banking, you cannot rely on third-party blueprints," said Nesterov. "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."

Anton Repushko, who leads the new division, said the aim was to build a broader model for finance rather than use a series of smaller systems.

"Revolut Research has been established to responsibly build financial intelligence at its deepest layer, rather than patching together narrow, specialised models," said Repushko. "In PRAGMA, we are developing a single, unified foundation model capable of understanding the true nuance of financial behaviour in real time. Technology is in Revolut's DNA, and by collaborating with global tech leaders, this division is engineering proprietary capabilities that set us apart from traditional banks."

Data advantage

A central part of Revolut's argument is scale. Its customer base, cross-border activity and range of products provide a broad set of transaction and behavioural data that can be used to refine models over time.

That approach reflects a wider belief in finance that data collected through day-to-day operations can become a competitive asset if companies can turn it into better decisions on lending, security and product targeting. It also raises familiar questions about governance, oversight and the extent to which automated systems should influence consumer financial outcomes.

Revolut did not disclose financial details for the new division or its work with NVIDIA, but the announcement shows the company expanding its investment in internal research as competition intensifies among digital banks, incumbents and technology providers seeking a bigger role in financial services.

Early benchmark claims will be watched closely by peers, particularly in fraud and credit, where even small changes in model performance can materially affect losses and approval decisions. Revolut says early PRAGMA deployments on historical data delivered 17% greater precision in fraud alerts alongside the increase in fraud cases caught.