We study innovation diffusion and ask how to optimally balance between incentives and compromising social learning. Motivated by the Covid-19 crisis we consider an individual’s decision to adopt the innovation (e.g., a vaccine) depending on a known  cost and an unknown benefit, which is learned over time through social learning. The population is heterogeneous with respect to the perceived value of innovation and the acquisition of information. We consider policy interventions that add incentives towards adoption, and (thus) compromise the perceived reliability of social learning. We derive a macro-level adoption dynamics and show that if a single-crossing assumption holds, there is a unique optimal time for any intervention (possibly never). When considering a set of interventions, we show that an optimal sequence exhibits increasing incentives. Our findings can inform policy, as we identify a local, testable condition that identifies the optimal intervention timing. Finally, using vaccination data from France’s 96 mainland départements, before and after the introduction of Covid certificates, we are able to validate the key assumptions of our model. We then calibrate the model to several countries that implemented Covid certificates during the pandemic, and assess the health impact of non-optimal timing.