This study demonstrates that mobile phone network data can enhance the targeting of humanitarian aid. Using machine-learning algorithms trained on survey data, the researchers identified patterns of poverty in mobile phone usage. This allowed them to prioritize aid to the poorest mobile subscribers. An evaluation of a cash transfer program in Togo showed that this approach reduced exclusion errors by 4–21% compared to geographic targeting, but increased them by 9–35% when compared to a hypothetical social registry approach. The findings highlight the potential of new data sources to complement traditional methods in targeting humanitarian assistance, especially in crisis situations.

Emily Aiken, Suzanne Bellue, Dean Karlan, Chris Udry, Joshua E. Blumenstock