A Togo cash-transfer program used machine learning and mobile network data to identify people most likely to need assistance. Its results show the potential of alternative data—and the importance of current training data, inclusion safeguards and privacy controls.