Single-Leg Revenue Management with Advice

Machine learning algorithms are becoming increasingly powerful, but with that power comes greater complexity and opacity. As these models become more sophisticated, they become increasingly difficult to understand—and, crucially, harder to anticipate when and how they might fail. This makes it essential to incorporate their predictions in ways that remain robust to errors. In “Single-Leg Revenue Management with Advice,” S. Balseiro, C. Kroer, and R. Kumar develop an algorithm for the classical single-leg revenue management problem that robustly incorporates predictions. They uncover a fundamental tradeoff: Placing greater trust in predictive models can yield high performance when predictions are accurate but also makes algorithms vulnerable when predictions are off. The proposed algorithm achieves the optimal tradeoff between these goals, allowing decision makers to leverage machine learning predictions while guarding against their potential inaccuracies. By doing so, this work provides a principled approach to integrating powerful yet imperfect forecasts into real-world decision making.