AI for AML: where it delivers measurable value

Finextra and NICE Actimize hosted an online webinar that examined where artificial intelligence delivers measurable value across the anti-money laundering lifecycle. Speakers included Nicola Eschenburg, EMEA AML Strategy Leader at NICE Actimize, with Teresa Connors serving as moderator.

The session focused on specific use cases in customer onboarding, know-your-customer checks, transaction monitoring and investigative workflows. Panelists described applications in adverse media screening, customer risk scoring, transaction screening optimisation, automated alert triage, investigation support and suspicious activity report (SAR) preparation.

Speakers outlined how different AI methods are being used. Traditional machine learning models are applied to detect patterns in transaction data and reduce false positives in screening. Large language models are used to extract meaning from unstructured sources such as news, documents and case notes during adverse media searches and investigations. Panelists identified agentic AI as an emerging option for automating sequences of investigative tasks while keeping human reviewers involved in decision points.

The webinar addressed the shift from pilot projects to broader operational deployments. Panelists identified common obstacles, including fragmented data across systems, lack of end-to-end workflows and the difficulty of surfacing risk signals consistently across onboarding, monitoring and case management functions. They said institutions need to map where AI can produce measurable efficiency or detection improvements before integrating models into production systems.

Governance and regulatory expectations were a recurring topic. Panelists recommended documentation of model design and performance, routine testing, explainability measures for decision paths, and oversight arrangements that align AI use with supervisory requirements and an institution’s risk appetite. They advised keeping records that show how models perform and how outcomes arise from model inputs and processes.

On sourcing AI capabilities, the panel outlined trade-offs between in-house development, vendor solutions and hybrid approaches. Internal build efforts can use an institution’s specific data and processes, while vendors can offer faster deployment and specialised tools. Many institutions are combining vendor models with internal data, controls and governance to balance speed and oversight.

Panelists suggested specific metrics to demonstrate value, including reductions in alert volumes, shorter investigation time-to-resolution, higher detection rates and improved accuracy in customer risk assessments. They said tracking such metrics helps turn theoretical benefits into measurable operational outcomes.

The webinar presented regulatory background that shows growing supervisory focus on responsible AI use in financial crime controls. Panelists set out steps for scaling proven AI applications across the AML lifecycle while maintaining documentation, testing and human oversight.