Algorithms and Bureaucrats: Evidence from Tax Audit Selection in Senegal

Can algorithms enhance bureaucrats’ work in developing countries? In data-poor environments, bureaucrats often exercise discretion over key decisions, such as audit selection. Exploiting newly digitized micro-data, we conduct an at-scale field experiment whereby half of Senegal’s annual audit program is selected by tax inspectors and the other half by a transparent risk-scoring algorithm. Algorithm-selected audits are 18 ppt less likely to be conducted, detect 89% less evasion, are less cost-effective, and don’t reduce corruption. Moreover, even a machine-learning algorithm would only have moderately raised detected evasion. These results are consistent with bureaucrats’ expertise, the task complexity, and inherent data limitations.