Objective: This study aimed to demonstrate proof of concept for an adaptive treatment strategy in Internet-delivered cognitive-behavioral therapy (ICBT), where risk of treatment failure is assessed early in treatment and treatment for at-risk patients is adapted to prevent treatment failure. Methods: A semiautomated algorithm assessed risk of treatment failure early in treatment in 251 patients undergoing ICBT for insomnia with therapist guidance. At-risk patients were randomly assigned to continue standard ICBT or to receive adapted ICBT. The primary outcome was self-rated insomnia symptoms using the Insomnia Severity Index in a linear mixed-effects model. The main secondary outcome was treatment failure (having neither responded nor remitted at the posttreatment assessment). Results: A total of 102 patients were classified as at risk and randomly assigned to receive adapted ICBT (N=51) or standard ICBT (N=51); 149 patients were classified as not at risk. Patients not at risk had significantly greater score reductions on the Insomnia Severity Index than at-risk patients given standard ICBT. Adapted ICBT for at-risk patients was significantly more successful in reducing symptoms compared with standard ICBT, and it decreased the risk of failing treatment (odds ratio= 0.33). At-risk patients receiving adapted ICBT were not more likely to experience treatment failure than those not at risk (odds ratio= 0.51), though they were less likely to experience remission. Adapted treatment required, on average, 14 more minutes of therapist-patient time per remaining week. Conclusions: An adaptive treatment strategy can increase treatment effects for at-risk patients and reduce the number of failed treatments. Future studies should improve accuracy in classification algorithms and identify key factors that boost the effect of adapted treatments.