The Bank for International Settlements (BIS) Innovation Hub, in collaboration with the Bank of England, has carried out a study known as Project Hertha, testing AI techniques to identify fraud within payment system data.  

It studies the potential of transaction analytics in detecting financial crime within real-time retail payment systems.  

The project leverages AI to pinpoint “complex” criminal networks that operate across multiple financial institutions. 

Project Hertha’s findings suggest that payment system analytics can aid banks and payment service providers in spotting illicit activities and the use of these analytics resulted in a 12% increase in the identification of illegal accounts.  

Moreover, the system proved detecting new patterns of financial crime, with a 26% improvement in recognising previously unknown behaviours. 

The research was conducted using a synthetic transaction dataset, representing 1.8 million bank accounts and 308 million transactions.  

This dataset was generated by an AI model designed to simulate realistic transaction patterns, ensuring that no real customer data was involved. 

The results also indicate that the application of system analytics has its limitations and is not a solution.  

The introduction of such analytics would involve “practical, legal, and regulatory” considerations, which were not explored within the project’s scope. 

The project also highlights the importance of access to labelled training data, a feedback loop for the AI model, and the need for algorithms that can be explained to “maximise” the effectiveness of such systems. 

Project Hertha draws its name from the British scientist Hertha Ayrton, known for her work in the physical sciences and her historic presentation to the Royal Society in 1904.