Triage objective
Triage should determine the next accountable action using sufficient context. It should not optimise only for queue closure or treat a low model score as proof that an alert is immaterial.
Minimum case context
- Alert rule, scenario, threshold, and generation time.
- Customer identity, business profile, and risk classification.
- Relevant transaction and counterparty history.
- Screening or intelligence matches and their provenance.
- Previous alerts, reviews, and reported activity where permitted.
- Policy version and decision options available to the analyst.
Review sequence
- Confirm that the alert and referenced transaction are correctly identified.
- Validate the customer and counterparty context used by the alert.
- Assess the behaviour against the scenario and available history.
- Record material supporting and contradictory evidence.
- Select an approved disposition and reason code.
- Escalate where the evidence, policy, or authority requires it.
- Preserve the decision, reviewer, time, and evidence references.
Assisted review
AI assistance can assemble context, identify inconsistencies, draft a review memo, and recommend reason codes. The interface should distinguish extracted facts from analytical inferences and show the evidence behind each material statement.
Quality measures
Measure more than handling time. Useful measures include re-open rate, escalation precision, missing evidence, reason-code consistency, overdue cases, repeat alerts, and the proportion of decisions changed during quality review.
Escalation
Escalation criteria should be explicit, testable, and connected to authority. The workflow must not rely on an analyst knowing an undocumented expectation or contacting a specific person outside the case record.