Questions to Ask Before Choosing AML Technology
A product demonstration can show what a system does. It cannot establish whether the system fits your risks, data or operating model. Start with the decisions your teams need to make, then test the supplier's claims against realistic work and evidence.
Define the problem before comparing products
Write down the control problem in operational terms. Which activity should the system support: customer due diligence, screening, transaction monitoring, investigations, reporting or a combination? Who uses the output, what decision follows, and what happens when information is missing or contradictory? Distinguish a regulatory or control need from a preference for a particular feature.
Map the current workflow, including hand-offs, manual steps, exception routes and records retained. Identify what already works and what is failing. A purchase may not solve a problem caused by unclear ownership, incomplete source data or a process that gives staff too little context.
Make data and workflow part of the evaluation
List the data needed for the intended use, where it comes from, who owns it and how quality problems are found. Ask suppliers to explain how records are received, checked, matched, transformed and retained. Test realistic cases, including incomplete data, duplicates, late records and changes to customer or transaction details. Record which populations or events are outside the system's coverage.
Use the same representative scenarios and agreed measures with each supplier. Observe how an analyst reaches an outcome, what supporting information is visible, how a decision is recorded and how a case moves to escalation. Consider integration with existing systems, access controls, audit records, migration, continuity, training and the effort needed to maintain the service. A feature list or headline price alone will not show these operating costs.
Ask for evidence, limits and accountability
For each important claim, ask what evidence supports it, under which conditions it was tested, and what limitations remain. Establish who configures rules or models, who approves changes, how performance and exceptions are monitored, and how the organisation can investigate an unexpected result. If the system includes AI, clarify whether it ranks, recommends, summarises or acts; identify what human review is required and how outputs can be challenged.
FATF's work on digital transformation describes potential benefits of new technology while emphasising enabling conditions, sound practices and applicable privacy and data-protection frameworks. The FCA and Bank of England's 2024 survey of AI in UK financial services also examines governance, accountability, monitoring and third-party providers. These publications offer questions for evaluation; neither certifies a supplier or establishes that a product is suitable for a particular firm.
Keep the decision record useful
A clear record should capture the requirement, test method, evidence considered, assumptions, limitations, unresolved dependencies and reason for the decision. Note which matters need further technical, security, legal, procurement or privacy review. Set out ownership for implementation and later monitoring. If the preferred option depends on a data improvement or process change, record that condition instead of treating it as already solved.
These steps help teams make a reasoned comparison. They do not guarantee that a system will detect financial crime, deliver a particular reduction in alerts or satisfy every regulatory requirement. Applicable obligations depend on the firm's activities and jurisdiction.
Common questions
Should we issue a request for proposal before defining requirements?
First describe the control need, users, workflow, data and decision the system should support. Those details make supplier questions and responses more comparable. Procurement format depends on the organisation.
How can we compare AML technology suppliers fairly?
Give each supplier the same requirements and representative test cases. Record the method, evidence, assumptions and limitations, and assess workflow, data, governance, integration and ongoing operating needs as well as features.
Does AI in an AML product change the evaluation?
Yes. Establish what the AI does, which data it uses, how it was tested, what people must review, who can change it and how results and incidents are monitored. The required controls depend on the use case and applicable rules.