Revolut launches AI research unit to build models in-house

Fintech giant Revolut has announced the establishment of a dedicated AI research division called Revolut Research, aimed at developing proprietary machine learning models for financial services. This move comes as European digital banks intensify competition to build in-house models rather than simply adding consumer chatbots to existing products.

The new unit will collaborate with academic and technology partners to pioneer advanced financial services AI technology, with the flagship project being PRAGMA, a foundation model developed with Nvidia. Early tests of PRAGMA have shown impressive results, including 2.3 times better credit risk accuracy, 65% more fraud detection, and 41% more relevant product recommendations.

Revolut's head of AI, Pavel Nesterov, emphasized the importance of building their own models instead of relying on third-party solutions. The company's vast dataset from 80 million customers will fuel the development of smarter features, creating a "self-reinforcing loop" as the dataset grows and models become more sophisticated.

Revolut plans to share its scientific findings and technical frameworks, engaging with the broader scientific community through conferences and meetups. The AI race is heating up among Europe's neobanks, with competitors such as Starling, Bunq, and Revolut themselves launching AI-powered assistants to enhance customer experiences and streamline operations.