AI for AML: From pilots to operational use

An online webinar hosted with NICE Actimize brought together industry experts to review practical uses of artificial intelligence across the anti‑money laundering lifecycle. The panel included Nicola Eschenburg, EMEA AML Strategy Leader at NICE Actimize, and Ashley Bostel, Head of Economic Crime Analytics Strategy at NatWest Group, with Teresa Connors moderating.

Panelists described specific applications they consider ready for operational use. Examples cited include automated adverse media searches to identify negative news, machine‑assisted customer risk assessments that combine structured and unstructured data, screening optimisation to cut false positives, and automated alert triage that ranks higher‑risk cases for investigator review. Automation of routine investigation tasks and tools to support suspicious activity report preparation were also discussed as areas generating time savings and greater consistency.

Speakers covered differences between established machine‑learning systems and newer large language models. LLMs were described as useful for reviewing unstructured text and producing initial case summaries for investigators, while traditional numeric and pattern‑based models remain in use for transaction monitoring. The panel noted agentic AI-systems that can carry out multi‑step actions-as an emerging capability for parts of case management and workflow orchestration when paired with controls.

Panelists addressed the practical problem of connecting risk signals held in separate systems. They outlined how linking onboarding records, ongoing KYC attributes, transaction monitoring outputs and external intelligence can create a single view of customer risk. A joined view was presented as a way to reduce duplicated work across compliance teams and shorten decision times.

Governance and regulatory expectations were a major topic. The panel emphasised transparency and model explainability where feasible, robust testing and ongoing monitoring to detect model drift and bias, documented decision processes, preserved audit trails and human oversight for automated outcomes.

The debate included build‑versus‑buy decisions. Organisations were said to be weighing in‑house model development against commercial offerings or hybrid approaches that combine vendor tools with internal data and rules. Factors mentioned as influencing those decisions included scale, existing infrastructure, available talent and the need for explainability under regulatory review.

Panelists urged firms to move beyond pilot projects and run pilots that demonstrate measurable efficiency or detection improvements, alongside investment in governance frameworks that support scaled deployments without undermining compliance obligations.