Development actors facing pressure to provide more rigorous assessments of their impact on policy and practice need new methods to deliver them. There is now a broad consensus that the traditional counterfactual analysis leading to the assessment of the net effect of an intervention is incapable of capturing the complexity of factors at play in any particular policy change. We suggest that evaluations focus instead on establishing whether a clearly-defined process of change has taken place, and improve the validity and credibility of qualitative impact statements. IIED research in Uganda shows that the methods of process tracing and Bayesian updating facilitate a dialogue between theory and evidence that allows us to assess our degree of confidence in ‘contribution claims’ in a transparent and replicable way.