Using payments data, privacy-enhancing technologies, AI and enhanced cooperation for a behavioural-based analysis approach to detecting money laundering networks is more effective than the current rules-based approach, concludes new research from the BIS Innovation Hub.
According to the Financial Action Task Force, almost all large money laundering schemes are cross-border and involve different business sectors. Meanwhile, financial institutions often face limitations in detecting potential suspicious networks and transactions due to their reliance on fragmented data and systems.
A LexisNexis Risk Solutions study on the costs of financial crime compliance underscores the financial burden placed by AML efforts on financial institutions. Between 2020 and 2022, these costs surged by around $60 billion to approximately $274 billion.
The BIS Innovation Hub's Nordic Centre has been investigating new approaches to the problem, carrying out a proof of concept - called Project Aurora - in partnership with Lucinity, an Icelandic AI software-as-a-service company.
The PoC used a comprehensive synthetic data set that represents real-world domestic and international payments data. To ensure the protection of sensitive information, privacy-enhancing technologies were employed, drawing on machine learning and other analytical tools while the data remain encrypted.
Subsequently, algorithms were trained on this synthetic data set to detect various patterns, known as "typologies," associated with money laundering activities across institutions and countries.
The project involved exploring different views on the synthetic data to represent various monitoring scenarios, including siloed, national and cross-border. Additionally, different approaches to collaborative analysis, including centralised, decentralised and hybrid models, were considered at both national and cross-border levels.
BIS says that the results highlight the "effectiveness of employing advanced analytics and technologies that adopt a behavioural-based analysis approach, which focuses on understanding the relationships between different individuals and businesses and identifying anomalies from normal behaviour. The results demonstrated that such methods are more effective in detecting money laundering networks than are the current rules-based approach, which is limited by its siloed nature."