Predictive thinking is an integral part of good decision making. However, predictions require plausible assumptions for extrapolating from the present to the future, the notoriously difficult problem of induction. Considerations of probability of success (PoS), or predictive power, comprise qualitative (e.g. choice of endpoint) and quantitative aspects. Well calibrated PoS calculations should include all relevant sources of uncertainty. These comprise sampling uncertainty, parameter uncertainty, as well as between-trial and between-endpoint heterogeneity (PFS vs. OS). The most informative phase II design is a randomized trial with the same endpoint as in phase III. If such a trial results in promising effect estimate and between-trial heterogeneity is small, PoS for phase III will be approximately 60 to 70%, but can be much smaller if endpoints in phase II and III differ (PFS vs. OS).