A project by the Bank of England and the London BIS Innovation Hub to use AI to spot unfolding and novel financial crime patterns in real-time retail payment systems showed promise but threw up a number of limitations to its efficacy.
To evade detection, criminals operate in complex networks which include many accounts across multiple financial institutions. Electronic payment systems process transactions across many participants, which gives them a network-wide view. The bank of England's Project Hertha tested the application of modern artificial intelligence (AI) techniques to help spot complex and coordinated criminal activity in payment system data.
The experiments were conducted using a state-of-the-art simulated synthetic transaction dataset, developed as part of the project. It includes data for 1.8 million bank accounts and 308 million transactions. The dataset was built by using an advanced AI model trained to simulate realistic transaction patterns.
It found that payment system analytics could be a valuable "supplementary tool" to help banks and payment service providers (PSPs) spot suspicious activity.
Banks and PSPs participating in the project uncovered 12% more illicit accounts than they would otherwise have found. The experiment also proved particularly valuable for spotting novel financial crime patterns. When trying to spot previously unseen behaviours, it helped achieve a 26% improvement.
"The results demonstrate promise but also show there are limits to the application and effectiveness of system analytics," states the BofE. "It is just one piece of the puzzle. The introduction of a similar solution would also raise complex practical, legal and regulatory issues. Analysing these was beyond the scope of Project Hertha."
The central bank says the results also highlight the importance of labelled training data, robust model feedback loop and explainable AI algorithms to maximise effectiveness.