The financial services industry is undergoing digital transformation accelerated by technology and data. This has disrupted how banks interact with data, processes, technology, and their customers. This has been catalysed by the increased use of digital electronic devices, social media networks and big data analytics.
Amid the disruption, today’s customer is demanding more trust and accountability. This has increased the scope of risks that range from, fraud, cybercrime, money laundering, and other financial-related crimes.
Money laundering particularly is a key priority area for banks considering regulatory obligations, and reputational and financial implications for non-compliance. A 2018 International Monetary Fund (IMF) report estimates that $1.5 to $4 trillion is laundered annually through the banking system.
Established banks have managed to implement machine learning (ML) monitoring at an individual institutional level, though reliance on legacy systems remains prevalent. Rule-based AML detection systems are not reliable in the digital era and have led to missed or low detection of laundering activities and high false positive alerts, pushing up investigation costs and efforts.
These systems are heavily relied upon to monitor and assess risks of transaction data across the banking ecosystem in real-time or near-real-time basis. The systems function based on developed algorithms found in expert systems. Detection scenarios are defined by way of facts and rules. The system pairs a transaction to an equivalent set detection scenario. Typically, detection scenarios are hard-coded by human experts based on their experience
Considering the huge gap between effort, costs and results of legacy rule-based systems, practical improvement of the systems is an issue to be addressed. With the complexity of a deepening financial sector, banks need to reassess their Artificial Intelligence AI/Machine learning and machine-people strategy.
Banks can leverage other emerging technologies including advanced neural analytics to triangulate relationships within complex transaction nodes.
Advanced risk-based systems can be complemented with legacy rule-based systems which can process and isolate the high volume of alerts from rule-based systems results on a more vertical level such as geographical locations, value, product, type of transactions and channel.
AI systems can effectively assign weighted risk scores to transactions. It can then categorise them as either higher, medium, and lower risk. Based on the allocated risk profile, the human analyst can manually review high and medium-risk exceptions. Business intelligence capabilities such as analytics dashboards of transactional relationships can improve the identification and assessment of suspicious patterns improving accuracy and time efficiency.
Human analysts will continue to play a significant role in AML analysis and reviews. However, their role must evolve to increase their effectiveness. With more accurate transaction monitoring systems, analysts can be freed-up to focus on higher-valued tasks such review and refining potential machine biases and study the ever-evolving strategies of money launderers. This way, they can craft more agile recommendations and long-term resilience can be achieved.
Khulabe is asenior manager, Data Analytics and Digital Forensics at PwC, Matogo is Associate, Deals-Transaction Advisory at PwC and Kerubo is an Associate, Data Analytics and Digital Forensics at PwC
