Public sector organizations often rely on reports by local monitors that are costly to verify and that serve twin objectives: to incentivize agent performance, and to provide information for planning purposes. Received wisdom has it that pay for locally monitored performance (P4LMP) will result in collusion and undermine both objectives. But simple Coasian logic suggests the reverse: P4LMP puts transferable money on the table and may enable interested parties to bargain to a more efficient outcome. This paper develops a theoretical model that shows why, and for which parameters, the welfare-enhancing Coasian scenario exists. Focusing on education, we model how the preferences of a teacher (agent) and head-teacher (local monitor) affect actual and reported teacher attendance, and how these equilibrium outcomes depend on the financial stakes attached to reports. To capture the value of information, we also consider the welfare of a bureaucracy that makes a costly policy mistake when holding the wrong belief about teacher performance. We test the model and estimate the predicted effects using data from a field experiment in Ugandan primary schools, randomly varying whether head teachers' reports of teacher attendance are tied to bonus payments or not. Consistent with Coasian logic, P4LMP increased actual and reported teacher attendance (by 9 and 15 percentage points respectively) and reduced policy mistakes (by 7 percentage points) relative to unincentivized local monitoring. We use these experimental impacts to undertake a detailed cost-benefit analysis and conclude, even under conservative assumptions, that welfare improved when paying for locally monitored performance.