Novissi — AI-targeted emergency cash transfers in Togo

Novissi is Togo's COVID-19 emergency cash-transfer programme, run by the Togolese government together with GiveDirectly, UC Berkeley's CEGA and IPA, created to reach the country's poorest residents quickly when no reliable income registry existed.

The partnership used machine-learning models trained on high-resolution satellite imagery and mobile-phone call-detail records to estimate household wealth at fine geographic and individual levels, then used these estimates to decide who received emergency cash payments.

The programme enrolled and paid more than 138,500 of the poorest Togolese around US$10 million in emergency cash; the ML-based targeting is reported to have been about 42% more precise than simple geographic targeting of the poorest cantons.

The page describes this as a landmark of data-driven social protection, while noting that phone-based targeting can exclude people without phones, and that inferring wealth from personal data raises privacy questions the research team says it partly addressed through differential-privacy methods.