Enhanced due diligence still needs a person

Enhanced due diligence is where an agent's research is most useful and its judgement least appropriate. An agent can gather source-of-wealth documents, trace ownership, check adverse media and assemble the file. Deciding whether the firm is comfortable taking on, keeping or exiting a high-risk customer is a risk decision, and it stays with a named person.

What does the agent do well in EDD?

Enhanced due diligence is the deeper level of checking applied to customers who present higher risk: politically exposed persons, customers in higher-risk jurisdictions, complex ownership structures, unusual sources of funds. It goes beyond standard customer due diligence and, in most firms, consumes far more analyst time per case.

Much of that time is spent gathering and organising. An analyst has to collect documents, read company registry extracts, map ownership chains, search news sources in several languages, cross-check what the customer has declared against what can be found independently, and write it all up. An agent can do most of that legwork at depth 1 and 2: reading, retrieving, structuring and drafting the file for review.

It can also do something analysts often lack time for: list what is missing. A good EDD agent produces not only a summary of what it found but a clear statement of the gaps, such as a declared source of wealth with no supporting evidence, or a shareholder layer it could not resolve.

Why does the decision stay with a person?

The FATF Recommendations set a risk-based approach as the core of anti-money-laundering standards. A risk-based approach means the firm decides, case by case, whether the risk a customer presents is within its appetite and whether its controls are adequate for it. That is a judgement about plausibility, proportion and appetite, not a calculation.

Consider what the decision involves. Is the declared source of wealth plausible for someone with this career history? Is an adverse media story serious, or a decade-old dispute that was resolved? Does a complex structure have a legitimate business reason? Is the firm willing to accept the reputational exposure of this relationship? Each answer depends on context, experience and the firm's own risk appetite. And for many high-risk categories, rules in most jurisdictions expect senior management approval for the relationship.

An agent can inform every one of those questions. It should answer none of them.

How should evidence be weighed by its source?

EDD files mix evidence of very different strength, and the agent should make that visible. Provenance helps here:

• Sealed facts come from the firm's own systems of record, such as the customer's transaction history on the core ledger.
• Signed facts come from a named party who answers for them, such as a verified registry extract or a bank reference.
• Declared facts are what the customer says, such as their stated source of wealth.
• External facts arrive from outside without verification, such as a news article or a web search result.

A source-of-wealth narrative supported only by declared and external facts is a very different file from one supported by signed documents. The agent should label each fact by class so the analyst can see at a glance how much of the file rests on the customer's word.

Where is the line between agent and person?

An enhanced due diligence checklist can be used as both the agent's work list and the analyst's review list. The full process is set out in how to run enhanced due diligence on high-risk customers with an AI agent.

Where do agents mislead in EDD?

In three predictable ways. First, adverse media matching: common names produce stories about other people, and an agent that is not careful about identifiers will attach them to the wrong customer. Second, absence of evidence: an agent that finds no adverse media may imply the customer is clean, when the search simply did not cover the right languages or sources. Third, fluency: a well-written draft makes thin evidence look thorough.

Each can be controlled. Require the agent to state the identifiers it matched on for every media hit. Require it to record which sources it searched and in which languages. And require the draft to separate verified facts from declared ones, so fluency cannot hide the difference.

What about ongoing monitoring?

EDD is not only an onboarding event. High-risk relationships need periodic review, and that is where agents can be especially useful: rerunning searches, comparing the customer's current activity against their profile, and flagging changes since the last review. The pattern holds. The agent prepares the refreshed file and highlights what changed; a person decides whether the relationship continues on the same terms.

The same split applies to business customers, covered in how to run business customer due diligence with an AI agent.

What should you do differently?

Use agents to make EDD files more complete and more honest about their gaps, and use the time saved to give analysts and approvers more attention per decision. Do not let a well-prepared file become a reason to approve faster. The value of the agent is better evidence for a human judgement, which is the role set out in AI compliance agents: second line, never the last word.