AI in AML: The Three Honest Questions Every Buyer Should Ask

Question 1: What model decides what, and what model only suggests?

There is a meaningful difference between a system that uses AI to *suggest* a typology to an investigator, and a system that uses AI to *decide* whether an alert ever reaches a human.

The first is decision support. It is generally safe, generally explainable, and generally regulator-friendly. The second is autonomous triage. It is significantly more powerful, and significantly more dangerous, because the first time a missed alert turns into a regulatory event, "the model deprioritized it" is not an answer that holds up.

Many vendors use the same word — "AI" — to describe both. Force them to draw the line on the whiteboard. The vendors who can are the ones worth trusting.

Question 2: How was the model trained, and how is it monitored after the contract is signed?

The interesting question isn't "what does your model do today." It's "what happens when a new typology emerges six months from now, and your model has never seen it?"

Push for specifics:

Question 3: Can you defend this to my regulator without me in the room?

This is the question that quietly separates serious AML AI vendors from the rest. It is also the question I most often see compliance officers forget to ask in the excitement of the demo.

A good vendor will have:

The honest pitch I would write

If I had to write a vendor pitch I would actually believe, it would say something like:

*"We use machine learning to surface patterns that traditional rules miss, in places where the cost of missing them is high. Our models support — but do not replace — investigator judgment. We retrain on a defined cadence, we monitor drift, and our model risk documentation has been reviewed by clients with three different lead regulators. We are happy to walk your second-line team through every decision the model makes."*

That is a less exciting paragraph than "AI-powered platform that detects financial crime in real time." It is also the only kind of paragraph a serious buyer should sign on.

Where this leaves the marketer in me

I am paid to make AML AI sound compelling. I genuinely believe in this category — it is one of the few places where AI makes a measurable, defensible difference. But the way to win this market is not to lean into the hype. It is to be the team that answers these three questions clearly, in writing, before the customer asks them.

That is the marketing job worth doing.