Advances in digital data and algorithms are enabling new approaches to poverty targeting at scale. Using rich data from Bangladesh and Togo, we compare an algorithmic approach based on machine learning and mobile phone data to status quo targeting with proxy means tests and community-based targeting. While proxy means tests are most accurate, algorithmic targeting is more cost effective for programs where the budget is small relative to the number of households screened. Combining our estimates with global program data, we estimate that phone-based targeting would be the welfare-maximizing approach for up to 30% of countries’ social assistance programs.

Suggested Citation: Aiken, Emily, Anik Ashraf, Joshua E. Blumenstock, Raymond P. Guiteras and Ahmed Mushfiq Mobarak, “Scalable Targeting of Social Protection: When Do Algorithms Out-Perform Surveys and Community Knowledge?” CEnREP Working Paper No. 25-001, November 2025, https://go.ncsu.edu/cenrep.wp.25.001.