Accuracy, consistency, efficiency: How AI strengthens AML compliance

By Neil Katkov

Legacy technology runs the risk of hampering bank operations and compliance initiatives. Today, that’s true for anti-money laundering compliance, where the operational imbalance between automated alert generation and manual alert investigation places an undue burden on bank risk professionals.

Automated systems (such as name screening and behavior detection) generate significant numbers of alerts that are dominated by false positives. These then move to human analysts for manual review as part of case investigation. With the increasing volume of alerts comes the need for more and more staff to investigate cases. In turn, costs associated with AML investigation are skyrocketing. Globally, financial institutions spent an estimated $26.4 billion on AML operations in 2021.

The manual processes—and related expenses—are not sustainable. Artificial intelligence solutions, however, can automate investigative operations. AI-supported AML investigation can reduce the burden of false positives by eliminating the imbalance between alert generation and the effort needed for investigation. The results: strengthened policy adherence, improved accuracy and consistency of results, minimized expenses for AML, greater overall efficiency of AML compliance operations and lowered regulatory risk and compliance.

Six ways AI improves AML investigation workflows

AI case management solutions can rapidly identify obvious false positives and automatically create evidence-based decisions, delivering results to analysts in a straightforward, user-friendly interface. Rather than having to manually sort through piles of false positives, human analysts can draw on the focused results delivered through AI-supported investigations.

AI delivers efficiencies for AML investigation workflow in six important ways:

1. Gathering data for entity investigation. The most time-consuming investigation activity? Gathering evidence on entities, which requires analysts to search and document multiple data sources. Automating the process of harnessing this data is now possible with natural language processing, which performs contextual searches of adverse media and other data sources. Risk scores are generated based on the findings. Results are automatically collated in dossiers that can help analysts understand the cases.

2. Performing network analysis. AI and graph databases can supercharge network analysis—even beyond a bank’s own customers, to external counterparties—by linking additional kinds of data, presenting the results through interactive visualization tools. This can reveal hidden account and customer relationships, analyzing transactions flows between accounts and generating alerts, where appropriate.

3. Streamlining beneficial owner analysis. By pulling and analyzing company datasets, media and other sources, AI can identify and compile beneficial owners, parent companies and affiliates, ownership percentages and directors/controllers. These initiatives help meet regulation like the Anti-Money Laundering Act of 2020 and the European Union’s Fourth Anti-Money Laundering Directive.

4. Generating case narratives and suspicious activity reports. Natural language generation can be used to systemically write up evidence and conclusions into alerts, case narratives or for filing as suspicious activity reports. These auto-narratives reduce the amount of time analysts need to dedicate to confirming or further examining evidence. Because standardized policies are used to inform the narratives, reporting is consistent and complete.

5. Automating alerts and case decisions. With automated case investigation, AI routines can deliver risk-ranked decisions (based on transaction, screening, beneficial owner and link/network evidence) that analysts can then review, confirm and/or investigate later. Machine learning support the continuous improvement and accuracy of alerts and case decisions by comparing results of previous decisions (generated by machines or by analysts) against new alerts.

6. Reducing false positives. Sometimes false positives are obvious; sometimes not. AI routines can identify the many varieties of false positive alerts, closing obvious false positives automatically. This may include eliminating matches made on irrelevant attributes (such as street names) or using known customer attributes to resolve the misidentification of high-risk entities. When working with behavior detection alerts, in-depth, AI-supported automated analysis (such as benchmarking transactions against historical customer or peer group patterns) may be needed to identify the false positives. The result: reduction in the remaining workload for analysts.

AI will continue to evolve in ways that support AML compliance. Currently available technologies minimize the amount of time required for analysts to investigate cases, while enabling more investigations to be completed in a single platform.

False positives are likely to linger, but AI can minimize this number drastically. The next generation of monitoring technology, AI-based detection, may reduce false positives even further. Not only do these improvement save time and expenses, AI reduces the regulatory risk that’s part and parcel of AML compliance operations.

Neil Katkov oversees the risk space at Celent, a global research and advisory firm focused on technology and business strategies in the financial services industry.