Research Information System
Journal of Retailing and Consumer Services, vol.92, 2026 (SSCI, Scopus)
The escalating incidence of product returns poses a critical threat to profitability and sustainability in the apparel retail industry, yet firms continue to lack actionable, customer-level foresight to pre-empt such behavior across omnichannel settings. In collaboration with a leading European fashion retailer, this study analyzes a fully anonymized panel comprising 58,412 shoppers who generated 1.27 million purchases and 243,000 returns over twelve months. The dataset integrates e-commerce clickstream records, in-store radio-frequency identification (RFID) fitting room scans, mobile application geolocation events, and detailed product metadata. Using 142 theory-informed features grounded in cognitive dissonance, expectation-disconfirmation, information search, and omnichannel integration frameworks, we develop explainable machine learning models to predict return behavior. Our gradient-boosted decision tree model achieves an area under the receiver operating characteristic curve (AUC-ROC) of 0.879, representing a 39% reduction in prediction error compared with baseline heuristics and marking a significant improvement over previous fashion return studies. Integrating omnichannel data yields an additional 90 basis points of predictive gain, with RFID fitting room scans contributing 45 basis points and geolocation signals adding 29 basis points beyond online-only predictors. Shapley additive explanation (SHAP) analysis indicates that theory-driven behavioral features account for 78% of the model's predictive power. Size-preference mismatches, multi-size basket behavior, and fit-query intensity emerge as dominant predictors, while demographic variables exert negligible influence—validating that cognitive dissonance and expectation-disconfirmation mechanisms are more influential than product attributes in driving returns. To test managerial applicability, a six-week randomized field experiment was conducted on 112,473 high-risk transactions, deploying smart-fit nudging interventions. The intervention achieved a 3.0 percentage point absolute and 15.6% relative reduction in return rates without compromising conversion, average order value, or customer satisfaction. This outcome translates into an estimated annual value exceeding €136,000 and a 353% first-year return on investment. Heterogeneous treatment effects further substantiate the behavioral mechanisms, revealing the strongest impacts among multi-size baskets and fit-complex product categories, with effects remaining stable over time and persisting post-experiment. Overall, this study demonstrates that integrating comprehensive omnichannel behavioral data with theory-grounded explainable machine learning and empirically validated interventions enables retailers to proactively mitigate product returns by addressing size uncertainty at checkout—without constraining customer autonomy.