.svg)
Every Anti Money Laundering (AML) conversation I have this year seems to start in the same place: Artificial Intelligence (AI). Every vendor claims it, every roadmap includes it, and every compliance team feels pressure to move fast. I understand the instinct. But after sitting through enough of these conversations, I've landed on a simple view: AI is not a shortcut to good compliance. It is only as good as the foundations underneath it.
Too many programmes start with the technology rather than the problem it's meant to solve. A new model gets layered onto an existing process to cut alert volumes or speed up investigations, without anyone asking whether the underlying controls were sound to begin with. If they weren't, AI doesn't fix that. It just makes the same weaknesses harder to see.
Chasing zero false positives is the wrong goal
If there's one myth I'd like to retire, it's this: that the point of AI is to eliminate false positives. It isn't. An AML system throwing off zero alerts isn't efficient - it's probably taking on risk it shouldn't. The real question isn't how many alerts we can suppress. It's whether we're generating the right ones, and whether we can prioritise, review and evidence them in a way that matches our risk appetite. Regulators, from the the Financial Conduct Authority (FCA) in the UK to the Monetary Authority of Singapore (MAS), keep coming back to the same expectation: understand your risk, then apply proportionate controls. AI should help us make faster, more consistent decisions inside that framework, not replace the thinking that defines it.
Risk appetite is the real foundation
AI can't tell us which customers, products or geographies matter most. That's a compliance decision, and it has to be made before AI enters the picture. When institutions get risk-based configuration right first - thresholds aligned to genuine exposure, noise reduced at the source - AI gets to do what it's actually good at: spotting complex patterns and helping analysts work faster. Skip that step, and AI just becomes a smarter way of managing an inefficient process.
Explainability isn't a nice-to-have
In most industries, AI just needs to be right. In AML, it needs to show its workings. If a model can't tell us why an alert was closed, escalated or deprioritised, we've created a governance problem, not a productivity gain. Analysts need to trust what they're being shown, audit teams need a trail they can follow, and regulators increasingly expect exactly that kind of transparency, with the European Union AI Act being the clearest recent example. Good AI should offer reasoning, not just an output.
Embedded, not bolted on
I'm also wary of AI that sits outside the core AML platform as a separate layer. It looks like a quick win, but it usually means governing two systems instead of one, with context lost at every handoff. When AI is embedded, recommendations, feedback loops, and audit trails live in one operating model, which makes the whole thing easier to monitor, tune and explain as it evolves.
Analysts, not just agents
None of this works if we lose sight of the analyst. The best AI deployments I've seen make people more effective, not less involved. That means giving reviewers real authority to override a recommendation, enough time to do so properly, and enough understanding of the model to know when to push back. As agentic and more autonomous capabilities enter the picture, that oversight becomes more important, not less.
Readiness beats speed
The organisations that get the most from AI won't be the ones who deployed first. They'll be the ones who built the strongest foundations - risk appetite, governance, explainability, workflow and technology architecture - before they scaled up. That's the thinking behind the AI Readiness Checklist we've put together: a practical way to pressure-test your organisational readiness before you invest further.
If you want the full picture, including the complete readiness checklist across strategy, oversight, governance, technology and people, read our white paper, AI foundations for AML: building compliance-first intelligence.
