NITI ENTERPRISE — AI scheme integrity planning checklist

Proposed planning aid, not an operational model.
[ ] Define the scheme-specific suspected behaviour and authority for review.
[ ] Document data ownership, permitted use, quality and access controls.
[ ] Compare rules-only performance before adding ML.
[ ] Use validated investigation outcomes for labels.
[ ] Test false alerts, missed cases and subgroup effects.
[ ] Provide reason codes and a human review workflow.
[ ] Record evidence, decisions, model version and review outcomes.
[ ] Provide a correction and appeal route.
[ ] Monitor drift and reviewer capacity; define rollback.

https://www.nitienterprise.com/ai-fraud-detection