Banks Deploy AI in AML: From Pilots to Live Systems

Several banks have moved artificial intelligence tools from experimental pilots into active anti-money laundering operations, deploying machine learning and large language models across onboarding, transaction monitoring, alert triage and suspicious activity report preparation. Institutions are applying AI to specific AML tasks where results can be measured, including adverse media searches, customer risk scoring, screening optimisation and investigation automation.

Banks report using machine learning to refine transaction-monitoring models and reduce false positives. Large language models are being used to process unstructured documents and draft narrative sections of suspicious activity reports. Firms are integrating these capabilities into know-your-customer and onboarding workflows to accelerate identity checks and risk assessments, and into case management systems to link alerts and enrich investigations.

Compliance teams are working to connect signals from screening, transaction monitoring, customer due diligence and external media so analysts can view a single, consolidated customer risk profile. Aggregating data in this way supports faster and more consistent decision-making and reduces repetitive manual work for investigators.

Some organisations are testing agentic AI, which can run sequences of actions or coordinate multiple tools, for routine triage tasks such as gathering documents and surfacing context for analysts. Institutions continue to require human review and final sign-off on regulatory filings and high-risk determinations.

Governance and regulatory expectations are influencing deployment. Firms are prioritising explainability, model validation, audit trails and documentation. They are also focusing on data quality, ongoing performance monitoring and the ability to show clear decision paths to meet compliance requirements.

When choosing how to obtain AI capabilities, banks weigh build, buy or hybrid approaches. Decisions hinge on the volume and quality of internal data, the availability of AI specialists, the complexity of integrating with legacy systems and the need for transparency from vendors. Many organisations combine third-party detection models with internal controls and customised workflows.

Industry attention is on measurable operational results. Reported metrics used to judge success include reductions in alert volumes, faster case resolution times, improved report quality and lower compliance costs. A recent industry webinar assembled experts to discuss where AI is producing measurable outcomes across the AML lifecycle and how institutions can scale those applications while meeting governance requirements.