Suppliers who are better informed than purchasers, such as physicians treating insured patients, often have discretion over what to provide. This paper shows how, when the purchaser observes what is supplied but can observe neither recipient type nor the actual cost incurred, optimal provision differs from what would be efficient if the purchaser had full information, whether or not the supplier can extract informational rent. The analysis is applied to, among other things, data on tests for coronary artery disease and to Medicare diagnosisrelated groups defined by the treatment given, not just the diagnosis, illustrating the biases in provision that result.