Administrative data is key to many government functions; but generating and maintaining it is costly and challenging in low-income countries. We study an overhaul of public assistance in Pakistan that created a national database of household assets and used the data to means-test cash transfers, eliminating discretion in their allocation. We use difference-in-differences and regression discontinuity approaches to quantify the effect of this reform. Favoritism and transfers to wealthy households dropped; we estimate that the welfare benefits of the reform were seven times as large as its costs. The reform improved public perceptions of social assistance and helped create a robust institution that survived political transitions.