Understanding and predicting online product return behavior: An interpretable machine learning approach

Product return is a costly phenomenon, which could be mitigated by examining how intrinsic (e.g., durability, reliability) and extrinsic (e.g., services, packaging) product attributes trigger online product return behavior (PRB). This study develops an interpretable machine learning predictive approach for PRB by extracting product attributes from customer reviews across five product categories. Our results suggest that, among extrinsic attributes, product returns management, packaging and customer services are key PRB drivers. For intrinsic attributes, the impact is rather distinctive to each product category. To reduce PRB, sellers should encourage customer feedback on primary features for books, and secondary features for electronics. For food, home appliances and electronics, sellers should improve product appearances and provide estimated product lifespans with warranties to cover premature failures. Surprisingly, for fashion, durability is imperative as dissatisfied consumers may keep the product if it is durable. Regarding PRB prediction, the optimal random forest model can accurately flag reviews with high risk of return intention regardless explicit or implicit customers’ expressions. This helps sellers selectively prevent product returns in a cost-effective manner. The research contributes to the marketing-operations interface by supporting retailers in tailoring the online marketing strategy and post-purchase services, manufacturers in identifying and improving shortcomings of product design, and product return operators in quickly selecting the best treatment pathway for returned products.