Applied agricultural research institutes play different roles in complex agricultural innovation processes, contributing to them with other actors. To foster learning and usable knowledge on how research actions influence such lasting innovation processes, there is a need to identify the causal mechanisms linking these actions and the effects of the changes they enable. A participatory, theory-driven, ex-post evaluation method, ImpresS, was developed by the French Agricultural Research Center for International Development (Cirad). ImpresS reconstructs the innovation history and its impact pathway by analyzing behavioral mechanisms linked to stakeholders' individual reactions and responses, and underlying process mechanisms at a group level. ImpresS relies on iterative updating and refinement and on triangulating data sources and collection methods to ensure internal validity and to increase credibility by enabling different actors to express their viewpoints. Drawing on an in-depth case study, we discuss how ImpresS makes it possible to draw robust conclusions on causal mechanisms while posing challenges linked to the group dynamics and power imbalances commonly encountered in participatory methods. As demonstrated by the case study, ImpresS generates policy-relevant knowledge for future research projects. It also demonstrates how research actions can help coconstruct lasting dynamics that can survive fluctuating institutional support. Distinguishing between behavioral and process mechanisms benefits knowledge use as it makes it possible to disentangle the conditions that trigger changes in a given context while generating research questions concerning the external validity of mechanism hypotheses.