Abstract
Implementation research often fails to have its intended impact on what programs actually do. Embedding research within target organizational systems represents an effective response to this problem. However, contradictions associated with the approach often prevent its application. We present case studies of the application of embedded implementation research in Bangladesh, Ghana, and Tanzania where initiatives to strengthen community-based health systems were conducted using the embedded science model. In 2 of the cases, implementation research standards that are typically embraced without question were abandoned to ensure pursuit of embedded science. In the third example, statistical rigor was sustained, but this feature of the design was inconsistent with embedded science. In general, rigorous statistical designs employ units of observation that are inconsistent with organizational units that managers can control. Structural contradictions impede host institution ownership of research processes and utilization of results. Moreover, principles of scientific protocol leadership are inconsistent with managerial leadership. These and other embedded implementation science attributes are reviewed together with contradictions that challenged their pursuit in each case. Based on strategies that were effectively applied to offsetting challenges, a process of merging research with management is proposed that is derived from computer science. Known as "agile science," this paradigm combines scientific rigor with management decision making. This agile embedded research approach is designed to sustain scientific rigor while optimizing the integration of learning into managerial decision making.