We provide a systematic framework to diagnose underlying problems and predict intervention effectiveness ex-ante. For this, we developed a parsimonious and generalizable survey tool (anamnesis). Our anamnesis classifies underlying problems along three fundamental diagnoses: awareness, intention, and implementation problems. We validate the framework in an online experiment with 7,500 subjects. We find that (i) intervention effectiveness is heterogeneous across different settings, and (ii) our diagnosis accurately predicts this heterogeneity. On average, predicting a 10%-effect corresponds to an actual effectiveness of 8.92%. We further demonstrate the applicability of our framework to predict heterogeneities in the setting of COVID booster take-up.