Identification of new causal relationships among knowledge components is a critical but understudied element of recombination search. The more novel an innovation, the more causal relationships among knowledge components will differ from previous innovations. Difficulties in working out causal relationships will lead to many failures. We suggest that learning from different types of causal failures can assist firms in diagnosing a rugged underlying technological landscape associated with knowledge components, improving the likelihood of novel outcomes. We consider two types of causal failures, those dispersed within a knowledge domain and those from other knowledge domains. Drawing on a unique database that allows us to assess such failures in more than 40,000 drug innovation projects, we find that both types of causal failures in the search process help to improve the novelty of subsequent innovations. Even so, the inclusion of more varied knowledge components in the search process diminishes the ability to derive insights from causal failures. These results not only highlight the importance of causal failures in the innovation search process, they shed light on the contradictory nature of search associated with knowledge components and causal mechanisms.
