Back to Bigger Picture

BIGGER PICTURE Public & everyday services

When there was no up-to-date list of families needing help

Togo needed to extend emergency cash support into rural areas quickly. Surveys, satellite images and phone metadata helped build a practical targeting system when a complete household register was unavailable.

AI-generated illustration: Mobile transfers connected to a conceptual network of rural households.
Mobile transfers connected to a conceptual network of rural households. Conceptual illustration; not a photograph or an exact depiction of the real project.AI-generated illustration of the idea, not a photograph of the actual people, equipment or event.
Rural expansion reached
>138,500 recipients
Compared with geographic targeting
Fewer poor households excluded
Payment channel
Mobile money

How the idea works

  1. Map poorer areas

    Survey labels make satellite patterns useful for targeting.

  2. Estimate household need

    Phone-survey and metadata patterns inform consumption estimates.

  3. Deliver real transfers

    Eligible recipients receive mobile-money support.

Help could not wait for a new census

During the pandemic, Togo’s Novissi programme used mobile money to deliver cash. Expanding from cities into rural areas created a difficult question: which families should receive a limited budget when detailed household information was missing? Researchers and government combined existing information with fresh surveys to estimate need quickly enough to support the response.

Sources: [1]

First locate poverty, then identify likely need

The first model connected survey-measured wealth to features visible in satellite imagery, helping identify poorer areas. A second linked phone-survey answers to anonymized call and airtime-purchase patterns to predict consumption. These were statistical proxies, not access to conversations. The two stages narrowed the search from geography to potential recipients within the selected areas.

Sources: [3]

The point was getting money to more eligible people

The rural expansion ultimately enrolled more than 138,500 vulnerable recipients. Evaluations found fewer poor people wrongly excluded than under alternative geographic targeting approaches. J-PAL’s synthesis reports a 4–21% reduction in exclusion errors for the comparisons it summarizes. The benefit came from the combination: a targeting tool, an organization delivering transfers and money that households could actually use.

Sources: [1][2]

The people absent from the data still matter

A phone-based system can miss households without a phone. It is also not automatically more accurate than a comprehensive, current social registry. The project illustrates a useful emergency tool under particular constraints, not a universal entitlement algorithm. Cash-transfer welfare gains should not be attributed to AI alone; accountable delivery and alternatives for excluded households remain essential.

Sources: [3][1]

The original sources

  1. J-PALUsing machine learning and mobile phone data to improve social protection
  2. GiveDirectlyStudy: AI targeting helped reach more of the poorest people in Togo
  3. J-PAL research projectUsing Mobile Phone and Satellite Data to Target Togo’s Emergency Cash Transfer Program

What to explore next

Traffic lights that tell the next junction what is comingA protein’s shape became something any lab could look up
Explore all 99 applications