The Bank of England, in collaboration with the BIS Innovation Hub in London, has conducted a promising trial using artificial intelligence to detect fraud in real-time retail payment systems. Dubbed Project Hertha, the initiative tested AI’s ability to identify complex and coordinated criminal activity through payment system data.
Using a highly realistic synthetic dataset of 1.8 million accounts and 308 million transactions, generated by a cutting-edge AI model, the experiment simulated how fraudsters operate across multiple banks using intricate networks. The findings revealed that AI-driven payment system analytics could serve as a valuable supplementary tool for banks and payment service providers.
Participants were able to detect 12% more illicit accounts than they would have using traditional methods, and the AI system was especially effective in identifying novel types of financial crime, with a 26% improvement in spotting previously unseen behaviors.
However, the Bank noted the limitations of AI in this context. While the results were encouraging, real-world implementation would raise practical, legal, and regulatory challenges, which were not explored within the scope of Project Hertha. The findings also stressed the importance of using labelled training data, a robust feedback loop for model learning, and explainable AI for transparency and effectiveness.