AI agents put data quality at centre of AML and KYC debate

Stephen Platt, executive adviser for financial crime and AML at KY360, part of Experian, has argued that AI agents are changing how banks approach anti-money laundering and know-your-customer controls. In an external opinion published by Finextra, Platt cited financial crime costs of more than $500bn a year globally and said many institutions still face transaction-monitoring false positive rates of 90% or higher.

The operational question for banks is whether agentic systems can reduce manual compliance workload without weakening accountability. Platt said the answer depends less on the sophistication of the AI than on the quality of the data and controls around it.

How agentic compliance systems work

Platt drew a distinction between machine-learning models already used by banks and AI agents. A model may assign a risk score to a transaction or customer file. An agent can use that signal to gather more information, compare records across data sets, review adverse media, examine customer history, prepare Suspicious Activity Reports and route cases to analysts with supporting evidence, according to Platt.

In KYC, he said such systems could coordinate tasks that often require multiple manual steps, including identity checks, sanctions and politically exposed person screening, beneficial ownership analysis and jurisdiction risk assessment. The systems can work across structured and unstructured information, including material in multiple languages, Platt said.

For AML teams, Platt argued that agents could screen alerts before they reach human reviewers by placing a flagged transaction in the context of a wider customer relationship. That could allow analysts to spend more time on higher-quality cases, although the claim rests on the reliability of the systems and data used.

Potential gains and constraints

Platt said banks spend billions each year on compliance operations, while large KYC remediation exercises can cost hundreds of millions and recur every few years. He argued that agentic AI could shorten due diligence processes that take days into minutes where the technology architecture is suitable.

He also identified customer onboarding as a commercial pressure point. KYC checks can cause customers to abandon applications, and Platt said faster identity verification, risk screening and due diligence could improve both compliance efficiency and customer experience.

Regulatory acceptance remains conditional. Platt said supervisors are becoming more receptive to AI in compliance where banks can show explainability and auditability. In practice, that means a system must record what information it used, how it reached a conclusion and why a decision can be defended.

Data and governance risks

Platt warned that agentic AI introduces material risk in financial crime controls. An agent using flawed logic or stale information could fail to trigger a Suspicious Activity Report or wrongly remove a customer, creating regulatory, legal and operational consequences.

Data quality is the central vulnerability, he said. Incomplete or inaccurate information on identity, ownership, transactions or sanctions status can lead an AI agent to produce a confident but wrong assessment. Platt said that risk may be worse than a system that clearly signals uncertainty.

He also pointed to third-party and concentration risk as banks adopt external AI platforms and data providers. As compliance infrastructure relies more heavily on vendors, Platt said supply-chain oversight becomes a board-level concern.

Platt concluded that human judgment will remain central to financial crime compliance, while AI agents may change where that judgment is applied. His argument places data governance, explainability and audit trails at the core of any bank deployment of agentic AML and KYC systems.

This story draws on original reporting from Finextra Research.