This paper studies when decentralization improves public service delivery. I analyze a Chicago reform that awarded select principals greater autonomy over budgets and operations while holding resources largely unchanged. A meta-analysis of similar reforms shows substantial heterogeneity, including both positive and negative effects. Building on insights from public finance, contract design, and psychology, I argue that the returns to autonomy depend on the capacity of local decision-makers (i.e., principals in this context) and the alignment of their objectives with those of central authorities. Event-study estimates show that, on average, increased autonomy improved achievement by about 0.1 standard deviations, effects comparable to resource-intensive interventions but achieved at minimal cost. However, deconvolution analysis reveals substantial heterogeneity, with both negative and positive effects. Design-based evidence supports the theoretical predictions; high-performing principals benefit more, reallocating resources effectively (e.g., reducing class sizes), and schools with atypical student populations benefit more and may tailor services to local needs. These results highlight that local capacity, aligned incentives, and heterogeneity are central to the success of decentralization reforms.