AI Anomaly Detection Cuts Verification Errors in Zambia's Social Cash Transfer — A Copperbelt Pilot

Zambia's Ministry of Community Development and Social Services (MCDSS) runs the Social Cash Transfer (SCT) programme, which reaches roughly 1.5 million beneficiaries nationally and has been progressively shifting to digital, biometrically verified payments through the Zambia Integrated Social Protection Information System (ZISPIS).
An independently published evaluation examined an AI-based anomaly-detection framework — built by Lusaka software firm Crystalised Applications — combining biometric verification, National Registration Card (NRC) optical character recognition, liveness checks, offline processing with server-side fallback, anomaly scoring and human review. Across roughly 71,000 beneficiary households processed during the September–October 2024 Emergency Cash Transfer cycle in Kitwe, Ndola, Chililabombwe and Solwezi, the framework cut average verification time from 8.5 minutes (manual NRC checks) to 3.1 minutes (offline biometric matching) and reduced the recorded error rate from 3.90% to 0.18%, while flagging 1,275 suspicious records, of which 959 (75.2%) were confirmed irregular on review.
This is a single academic study covering one emergency cycle in four Copperbelt districts — well short of the national 1.5-million-beneficiary caseload — and the authors themselves flag device heterogeneity, intermittent connectivity, data-protection concerns and limited AI governance capacity as open constraints before any nationwide rollout.

Read the full analysis: https://zenodo.org/records/21977682