The Bank of England, in collaboration with the BIS Innovation Hub's London Centre, has completed a pilot exploring how AI can be used to detect financial crime in real-time payment systems.

The Hertha project aimed to find solutions that can protect payment systems from financial crime without compromising privacy, which is often a challenging issue in the payments ecosystem.

The Hertha project takes its name from the pioneering British scientist, prolific inventor, and suffragette Hertha Ayrton, who became the first woman to read a paper before the Royal Society in 1906.

The latest experiment focused on testing the application of modern AI techniques to identify suspicious behaviour within criminal networks, which often span multiple banks and use numerous accounts to hide illicit behaviour.

It found that the analysis of payment systems could be a valuable addition to helping banks and payment service providers (PSPs) detect suspicious activities.

Because electronic payment systems operate across a wide financial network, the two organisations emphasised these systems are in a unique position to observe transaction patterns that could signal fraud.

Integrated with existing monitoring tools, the project highlighted how AI-based analytics have helped identify 12 per cent more illicit accounts than with traditional methods alone.

The research also highlighted how analysis of payment systems proved particularly valuable in detecting new patterns of financial crime, with detection rates improving by 26 per cent.

The project used a state-of-the-art synthetic simulated transaction dataset consisting of 308 million transactions and 1.8 million simulated bank accounts.

The data was generated by an AI model trained to boost the simulation of realistic transaction patterns, with the dataset designed to be representative of a retail payments ecosystem in a single jurisdiction.

While promising, the Bank of England cautioned that AI analysis is not a silver bullet, but “just one piece of the puzzle.”

“The introduction of a similar solution would also raise complex practical, legal and regulatory issues, and analysing these was beyond the scope of Project Hertha,” it said. "The results also highlight the importance of labelled training data, robust model feedback loop and explainable AI algorithms to maximise effectiveness.”

The initiative builds on the results of the previous Aurora Project, testing the potential of advanced transaction analysis to detect fraud while minimising the use of sensitive user data.

In 2023, the BIS led Project Aurora to combat cross-border money laundering by using AI and privacy-enhancing technologies and payment data to detect complex financial crime networks.

A new phase was launched in March 2024 to extend the results and develop real-world proofs of concept that could become a potential pilot project.

---