| dc.description.abstract | In Chapter 1, I and coauthors study the problem of predicting the product return rate using the products’ visual information. In online channels, products are returned at high rates. Shipping, processing, and refurbishing are so costly that a retailer’s profit is extremely sensitive to return rates. Using a large dataset from a European apparel retailer, we observe that return rates for fashion items bought online range from 13% to 96%, with an average of 53% – many items are not profitable. Because fashion seasons are over before sufficient data on return rates are observed, retailers need to anticipate each item’s return rate prior to launch. We use product images and traditional measures available prelaunch to predict individual item return rates and decide whether to include the item in the retailer’s assortment. We complement machine-based prediction with automatically extracted image-based interpretable features. Insights suggest how to select and design fashion items that are less likely to be returned. Our illustrative machine-learning models predict well and provide face-valid interpretations – the focal retailer can improve profit by 8.3% and identify items with features less likely to be returned. We demonstrate that other machine-learning models do almost as well, reinforcing the value of using prelaunch images to manage returns.
In Chapter 2, I consider customer search and product returns on the individual level. Previous research has focused on linking customers’ purchase and return decisions. However, online retailers have access to the information which precedes the purchase decision – customer search. I demonstrate that customer search information provides important insights about product returns. Using data from a large European apparel retailer, I propose and estimate a joint model of customer search, purchase, and return decisions. I then provide theory and data indicating that using search filters, viewing multiple colors of a product, spending more time, and purchasing the last item searched are negatively associated with the probability of a return. Finally, I use the proposed model to optimize the product display order on the retailer’s website.
Chapter 3 extends and reinforces the results obtained from previous Chapters. In the paper, I study the assortment planning problem in presence of frequent product returns. I develop a deep-learning model of customer search, purchase, and return. The model is based on a transformer framework and allows the recovery of important relations in the data. I use the estimated model to demonstrate that retailers could identify successful and unsuccessful products and modify the assortment. The modified assortment would increase the retailer’s sales and at the same time decrease returns. Lastly, I provide qualitative insights on which products are most likely to be unsuccessful in online retail. | |
