Abstract

Targeted coupons are widely used in e-commerce, yet campaign evaluation often emphasizes conversion or gross merchandise value without distinguishing realized coupon cost from the monetary exposure created by coupon allocation. This study presents an applied policy-evaluation framework that combines established propensity-adjustment, uplift-ranking, and offline evaluation tools to compare coupon-allocation policies under recipient-coverage and full-face-value budget constraints. The analysis uses 8414 anonymized assignment records representing 8402 users within a focal campaign window. Because the platform’s accounting margin is unavailable, a 20% margin is used as the baseline scenario for an incremental net-contribution proxy. Statistical uncertainty is assessed through 2000 user-cluster bootstrap replications and five fixed user-level data partitions. At 20% coverage, PA-RF produced a less negative proxy value than response-oriented policies, although its absolute confidence interval included zero and its ordering relative to PO-RF varied across data partitions. At 30% coverage, PA-RF produced a smaller net-contribution loss than PO-RF in all five partitions, while its absolute value and ROI proxy remained uncertain. The complete monetary-budget grid further shows that value-density allocation can differ materially from score-based allocation, that economically unattractive recipients may leave part of the available budget unused, and that expanding expenditure can reduce marginal subsidy efficiency. The contribution lies in a transparent and decision-oriented evaluation structure for interpreting coupon-allocation performance under explicit margin, cost, overlap, and model assumptions.

IPC Classification

Keywords