Financial crime investigators are employed to investigate. In practice, much of their working day goes on something else: trawling through systems, gathering data, compiling evidence, writing up findings and clearing large numbers of alerts that turn out to be harmless.

With alert volumes climbing, deadlines tightening and criminals becoming more sophisticated, already stretched teams in both Transaction Monitoring (TM) and Fraud are under growing pressure.

WorkFusion, a UiPath company, argues that AI can change this operating model, but not by handing everything over to Large Language Models (LLMs). Its AI Agent, Isaac, is built on what the firm calls a “Goldilocks” principle. It combines AI, deterministic automation and human judgement in the right proportions and orchestrates them across the investigation.

The aim is to shift where investigators spend their time, not to remove them from the process. After a TM alert fires, an analyst may need to pull customer records, review transactions, research counterparties, check case history, compare expected and actual behaviour, map relationships and document it all. Fraud cases are similar. An account takeover (ATO) alert can require evidence from fraud engines, authentication services, card networks, digital identity platforms, behavioural intelligence tools and third-party sources.

Isaac takes on much of this “hunter-gatherer” work. In Fraud Alert Review, it connects to existing fraud engines and data sources, pulls together relevant context, analyses transactions, drafts a narrative report and passes the finished work to investigators for human-in-the-loop review.

The firm stresses that not every task needs an LLM. Language models are well suited to unstructured information, such as synthesising historic SAR narratives, reconciling past dispositions against a current alert or interpreting messy counterparty data.

Source link