A bank can describe its money-laundering risk as moderate while being unable to explain what that judgement changes. The missing link is usually operational evidence: which customers, products and payment routes create the exposure; which controls address it; and whether those controls work. A useful AML risk assessment makes those connections visible enough for management to allocate resources and challenge growth plans.

The FATF Recommendations establish international standards implemented through national frameworks. For UK firms, the FCA Financial Crime Guide connects business-wide assessment with proportionate controls. The practical method below is an implementation recommendation, rather than a prescribed regulatory scoring formula.

Begin with a decision, then assemble the evidence

Identify the decision the assessment must support. Examples include launching a payment service, entering a customer segment, retaining a correspondent relationship or increasing an investigation team’s capacity. Starting with decisions prevents the exercise becoming a lengthy description of policies that nobody uses.

Define the assessed population before assigning ratings. Record legal entities, products, customer segments, delivery channels and the observation period. Reconcile counts to business systems. If dormant accounts are excluded, explain why and how reactivation is controlled. If a new product lacks a full year of data, show the shorter observation window. An apparently stable risk score can otherwise conceal changes in the underlying population.

For each material exposure, document a short causal statement. For example: rapid onboarding of small businesses may increase the chance that ownership information remains incomplete when transactional activity begins. This statement identifies a controllable process. A label such as “digital risk” does not.

Separate the risk from the control assessment

Measure exposure using relevant facts: customer volumes, products used, ownership complexity, geographic reach and transaction patterns. Assess controls with different evidence, such as data completeness, quality reviews, monitoring coverage and unresolved exceptions. Using the existence of a policy as evidence that a control is effective creates false reassurance.

Avoid mechanically averaging away a critical defect. A serious gap in identifying beneficial owners should remain visible even when training completion is excellent. Record whether a judgement rests on verified evidence, a reasonable estimate or an unresolved data gap. Uncertainty is information for the board; it should not be converted into an artificially precise decimal.

A hypothetical example: expansion into business accounts

Suppose a bank plans to double its small-business customer base. Its annual assessment assigns the segment a medium rating. A closer review finds that a newly introduced onboarding channel produces more ownership exceptions and that those exceptions take longer to resolve than in the established channel.

The useful response is not simply to raise the segment’s score. Management can limit growth in the affected channel, improve the evidence request, assign named exception owners and sample completed files. The decision record should explain the temporary constraint, the evidence needed to remove it and who can approve that removal. These are illustrative management choices, not universal regulatory requirements.

The next assessment should test whether the intervention reduced exposure. A falling backlog is helpful only if files were corrected rather than administratively closed. Compare error rates in newly opened accounts and completed remediation, and examine whether the same defect moved into another process.

Give the board a short, testable explanation

Present material movements, their causes and the decisions required. A practical board paper can include the affected population, the weakness, the interim response, the accountable owner and the residual uncertainty. Explain changes in methodology separately from changes in business risk. Otherwise a revised scoring model may look like improving performance.

Use a small set of follow-up measures. These might include the proportion of exposed relationships covered by a functioning control, repeat exceptions and the age of serious unresolved issues. Thresholds should reflect the bank’s circumstances and be documented as management choices. Their value lies in prompting action before the next annual assessment.

A defensible assessment is therefore reproducible. Another reviewer should be able to trace a material rating to its data, challenge the judgement and see what management did because of it. That traceability makes the assessment useful both in daily management and in supervisory discussions.