A project initiated by the Bank of England in collaboration with the London BIS Innovation Hub aimed to utilize AI for identifying emerging and novel patterns of financial crime in real-time retail payment systems. While the initiative showed promise, it also revealed several limitations.
Criminals often operate within complex networks that involve numerous accounts across various financial institutions. Electronic payment systems, which process transactions among many participants, provide a network-wide perspective. Project Hertha explored the use of advanced artificial intelligence (AI) techniques to detect complex and coordinated criminal activities in payment system data.
The project utilized a cutting-edge simulated synthetic transaction dataset, comprising information from 1.8 million bank accounts and 308 million transactions. This dataset was generated using a sophisticated AI model designed to replicate realistic transaction patterns.
The findings indicated that payment system analytics could serve as a valuable supplementary tool for banks and payment service providers (PSPs) in identifying suspicious activities. Participants in the project discovered 12% more illicit accounts compared to traditional methods. Furthermore, the AI system significantly enhanced the detection of novel financial crime patterns, yielding a 26% improvement in recognizing previously unseen behaviors.
While the results are promising, they also highlight limitations in the application and effectiveness of system analytics. The Bank of England noted that this approach is just one aspect of a broader solution and emphasized that implementing similar systems would involve complex practical, legal, and regulatory challenges, which were beyond the project’s scope.
Additionally, the central bank underlined the critical role of labeled training data, robust feedback mechanisms, and explainable AI algorithms to optimize effectiveness in combating financial crime.