Define the review question
Start with a decision the review must support. Examples include validating a new vendor, measuring analyst consistency, investigating a rise in rework, or confirming remediation after a policy change.
A broad instruction to "check KYC quality" usually produces an unfocused sample and findings that cannot be acted on.
Build the sample
Document the population, period, products, jurisdictions, customer types, outcomes, and exclusions. Use a combination of representative and risk-focused sampling where appropriate.
The sample record should explain why each case was eligible and how the final sample was selected.
Review each case
- Reconstruct the requirements and policy version in force.
- Confirm the evidence available to the original decision maker.
- Reperform the relevant checks without using hindsight as an undisclosed standard.
- Record defects against a controlled taxonomy.
- Separate factual errors, policy gaps, system defects, and judgment differences.
- Escalate material cases through the institution's approved process.
Defect taxonomy
| Class | Example | Typical response |
|---|---|---|
| Missing evidence | Required document or verification result absent | Correct case and review control design |
| Incorrect execution | Policy exists but was not applied | Training, workflow, or supervision change |
| Policy gap | Requirement is not translated into an operating rule | Policy and control redesign |
| System defect | Routing or field logic produces the wrong state | Engineering remediation and regression test |
| Documentation weakness | Decision may be reasonable but cannot be reconstructed | Improve reason and evidence capture |
Report and remediate
Report both the observed defect rate and the limitations of the sample. Assign remediation by root cause rather than sending every finding back to analysts.
Each action should have an owner, due date, validation method, and evidence location. Close an action only after the change is deployed and a follow-up sample supports effectiveness.