Yemen's Social Fund Pilots In-House AI to Police a 9.6-Million-Person Cash Programme

Yemen's national unconditional cash transfer programme, delivered with the government's Social Fund for Development and historically funded in part by the World Bank, supports an estimated 9.6 million people — around a third of the country's population — across roughly 1.5 million households.
With UNICEF Venture Fund backing, UNICEF's Yemen country office built an anomaly-detection model in-house, rather than through an external contractor, to flag irregular patterns — duplicated records, missed or unusual payments — in registration and enrolment data. The stated aim is to move fraud and error review from periodic, rules-based manual checks to continuous, individual-level monitoring, while keeping human review for sensitive cases.
An initial pilot phase concluded in late 2024, after which the Yemen Service Centre began refining the model further through 2025; UNICEF is also exploring AI-assisted triage of the programme's complaint and feedback mechanism.
Evidence is currently self-reported through UNICEF's own channels and describes potential and process change rather than published, independently verified results — no external audit, accuracy figure, or amount of fraud caught has yet been made public, so this is best read as an early, promising pilot rather than a proven model.

Read the full analysis: https://www.unicef.org/innovation/stories/ai-accountability