Date
A common approach in data-driven decision-making is the predict-then-optimize (PTO) framework, where a decision-maker first estimates an unknown parameter from historical data and then uses this estimate to solve an optimization problem. While widely used for its simplicity and modularity, PTO can lead to suboptimal decisions when the estimation step does not account for the structure of the downstream optimization problem. We study a class of problems where the PTO decision introduces asymmetry in the objective with respect to the unknown parameter, creating a systematic preference for under- or over-estimation when the estimate is noisy. To address this, we present a data-driven post-estimation adjustment that improves PTO’s decision quality while remaining practical and easy to implement. We show that when the objective satisfies a curvature condition based on the ratio of the third to the second derivative, the adjustment takes a simple closed-form expression. This applies to a broad range of pricing problems, including those with commonly used linear, log-linear, and power-law demand models. Under this condition, we establish theoretical guarantees that our adjustment uniformly outperforms standard PTO asymptotically, and we precisely characterize the resulting improvement. Additionally, we extend our framework to multi-parameter optimization settings and discuss applications where estimation bias is present. Numerical experiments demonstrate that our method consistently enhances revenue performance, particularly in small-sample regimes where estimation uncertainty is most pronounced, making it especially well-suited for pricing new products or those with limited price variation.