Suppose early trials suggest both efficacy of a new drug and the presence of a predictive biomarker. In such a situation, it may make sense to conduct a randomized phase II / III trial to confirm both findings. The phase II portion of the phase II / III trial uses a Bayesian design to confirm efficacy of the drug in the biomarker(+) stratum, and lack of efficacy (or potential harm) in the biomarker(-) stratum. Depending on the posterior probabilities of treatment effects considered clinically worthwhile and futile, either stratum can be continued or discontinued, leading to stopping further development, continuing the phase III trial with enrichment in biomarker(+) patients, or continuing the phase III trial in all patients. The advantages and limitations of this design are explored.