Causal machine learning methods could be used to predict treatment outcomes for subgroups and even individual patients; this Perspective outlines the potential benefits and limitations of the approach, offering practical guidance for appropriate clinical use.