AI for AML: Where it delivers measurable results

A webinar hosted in association with NICE Actimize brought together anti-money laundering experts to examine where artificial intelligence is delivering measurable value across the AML lifecycle. The panel included Nicola Eschenburg of NICE Actimize and Ashley Bostel of NatWest Group, with Teresa Connors moderating.

Panelists focused on practical applications that have moved beyond pilots into live use. They identified areas with quantifiable improvements in efficiency and detection, including adverse media research, customer risk assessment, screening optimisation, alert triage and automation of investigative steps, as well as preparation of suspicious activity reports.

The discussion covered technical approaches in current use. Machine learning models trained on transaction and behavioural data are being used to increase precision in transaction monitoring. Large language models are applied to parse unstructured sources such as news articles and case notes for adverse media and enhanced due diligence. Agentic AI, which can carry out multi-step tasks with controlled autonomy, is being deployed to automate routine investigation steps, draft SAR narratives and surface relevant documents when appropriate controls are implemented.

Panelists identified fragmented data as a barrier to measurable outcomes. Customer data, transaction histories, screening results and investigation notes often sit in separate systems. Integrating these sources and linking risk signals enables a more complete view of customer risk and supports faster, more consistent decision-making.

On sourcing AI capabilities, views varied. Some firms prefer to build models in-house to retain control and address specific data needs. Other firms buy vendor solutions to speed deployment and benefit from vendor support for regulatory requirements. Several panelists recommended a hybrid approach: use vendor platforms for standardised functions such as screening and case management, and develop proprietary models for institution-specific risk scoring and transaction monitoring nuances.

Regulatory and governance considerations were a major theme. The panel emphasised transparency, explainability and auditability for AI used in compliance workflows. Firms were advised to document model design, data inputs and performance metrics, maintain human oversight of automated decisions, and ensure models are tested for bias, validated and subject to ongoing monitoring to meet regulatory expectations.

Speakers outlined steps for scaling AI in AML programmes: start with targeted pilots on high-value problems, measure outcomes in operational terms such as reduced alert volumes or shorter case times, and expand pilots into production under formal governance frameworks. They recommended tracking business metrics rather than technical benchmarks alone to demonstrate return on investment.

Operational matters discussed included change management and skills. Firms will need data engineers, model risk specialists and investigators trained to work with AI outputs. Panelists advised integrating AI suggestions into existing workflows and user interfaces so investigators can act on model outputs without switching tools.

The panel reiterated that human judgment remains central to AML compliance. AI can triage, enrich and summarise information for investigators, while compliance officers retain responsibility for final risk decisions and attestations.