Quantifying the effect of a treatment is a central question in epidemiology, for which the average treatment effect (ATE) is the standard measure. However, the ATE has limitations in terms of clinical interpretability, particularly because it takes account of only a single outcome, whereas medical decisions are often based on multiple criteria (typically benefit, risk and tolerability). To address this limitation, measures based on hierarchical outcomes, such as the win ratio and the net benefit, have been introduced. In order to develop doubly robust and efficient estimators of the net benefit, a crucial step is to derive its efficient influence function (EIF). An initial approach to deriving this function was proposed in the literature, but the resulting function was only partially identified, as it still depended on unobserved counterfactual outcomes. Thus, after rigorously formulating the net benefit within a causal framework, we fully identified the associated EIF, thereby removing the need to model the distribution of counterfactual outcomes parametrically and enabling the construction of a doubly robust and efficient one-step estimator based on flexible, cross-fitted methods for estimating nuisance parameters. Additionally, we showed that our one-step estimator satisfies a central limit theorem under assumptions on the nuisance parameter estimators, including some that are atypical in the literature. In some settings, our estimator outperformed two classical competing estimators in a numerical study.