Clinicians often have to make treatment decisions based on the likelihood that an individual patient will benefit. In this article we consider the relevance of relative and absolute risk reductions, and draw attention to the importance of expressing the results of trials and subgroup analyses in terms of absolute risk. We describe the limitations of univariate subgroup analysis in situations in which there are several determinants of treatment effect, and review the potential for targeting treatments with risk models, especially when benefit is probably going to be dependent on the absolute risk of adverse outcomes with or without treatment. The ability to systematically take into account the characteristics of an individual patient and their interactions, to consider the risks and benefits of interventions separately if needed, and to provide patients with personalised estimates of their likelihood of benefit is shown using the example of endarterectomy for symptomatic carotid stenosis.