There is growing evidence that inappropriate prescribing is harming patients and raising costs in the US health care system. Through a partnership between the federal government and academics, we seek to develop evidence on reducing this prescribing. We conduct several randomized letter interventions targeting high-volume prescribers of drugs that can harm patients. We take a continuous improvement approach, rapidly evaluating each round and using the results to inform subsequent work. The first round of letters yielded no effects, and we responded with new interventions that are now under evaluation. We discuss lessons our work provides for future government-academic partnerships.