Research

working papers:

Customer Search and Product Returns
Marat Ibragimov, Siham El Kihal, John R. Hauser

Job Market Paper
Current version PDF

Abstract: Online retailers are challenged by frequent product returns with $428 Billion in merchandise returned to US retailers in 2020 (National Retail Foundation 2021). 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. We demonstrate that customer search information provides important insights about product returns. Using data from a large European apparel retailer, we propose and estimate a joint model of customer search, purchase, and return decisions. We 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, we use the proposed model to optimize the online assortment as well as the product display order on the retailer’s website.


Leveraging the Power of Images in Managing Product Return Rates
Daria Dzyabura, Siham El Kihal, John R. Hauser, Marat Ibragimov*

Minor Revision at Marketing Science
Current version PDF

Abstract: 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. In many product categories, such as the $500 billion fashion industry, experiments in real time are not feasible because the fashion season is over before sufficient return data are observed. We demonstrate that posted fashion-item images enhance return-rate selection of assortments. We develop three interconnected models: (1) a machine-learning model to predict return rates using images and other data available prelaunch. The model predicts well; robustness tests suggest it’s hard to find a better-predicting model, (2) an optimal policy to maximize profit given the imperfect predictive model, and (3) an interpretable model based on automatically-extracted image-processing features. The interpretable model provides valuable insights with which to select and design fashion items for the website. Using data from a large European retailer (over 1.2 million transactions for nearly 10,000 fashion items), we demonstrate that machine-learning methods are practical, scale to large collections and repeated fashion seasons, and improve profit relative to models using non-image data. We illustrate visually how automatically-extracted features affect return rates. Finally, we illustrate how data available postlaunch help manage return rates.


Transfer Learning for Personalized Marketing Promotions
Artem Timoshenko, Marat Ibragimov, Duncan I. Simester, Jonathan Parker, Antoinette Schoar

Abstract: Targeting policies are typically trained using data from field experiments. For example, a luxury fashion retailer can decide which customers should receive a coupon for the Valentine’s Day event using experimental data from a similar campaign implemented in the previous year. We demonstrate that firms can substantially improve targeting policies by augmenting the focal experiment with information from other marketing campaigns, even though the source campaigns may involve different marketing actions and different types of customers.




*Authors are listed in an alphabetic order