If our goal in education research is to produce knowledge that is useful and used, we need to design impact evaluations that value understanding variation as much as understanding average impact. The context in which an intervention is implemented and the characteristics of participants are integral components of the intervention, and we need to study those components just as seriously as we do the intervention itself. Better understanding variation requires building a framework for what contextual information is likely to predict effectiveness, so researchers know what data to collect; measuring inputs into education systems, not just outputs; and increasing collaboration across disciplines and modes of inquiry. It will also require creating more space for practitioners to participate equitably in research so their insights can inform our research questions and findings. Most challenging, researchers will need to let go of some old ways of working in service of producing research that better serves the field of education.