Data-driven personalized assortment optimization by considering customers’ value and their risk of churning: Case of online grocery shopping

• Published in: Computers & industrial engineering 2023-08, Vol.182, p.109328, Article 109328
• Main Authors: Saberi, Zahra, K. Hussain, Omar, Saberi, Morteza
• Format: Article
• Language: English
• Subjects: Customized assortment planning; Dynamic programming; E-tailing; Mixed multinomial logit (MNL); Survival analysis
• ISSN: 0360-8352; 1879-0550
• DOI: 10.1016/j.cie.2023.109328

•Developing dynamic assortment customization to increase the e-tailer’s profit.•Reducing the risk of customer churn in the case of an imbalanced inventory.•Using the survival analysis technique to find at-risk customers.•Elaborating the applicably of the approach using a real case study. Online shopping has enabled customers (shoppers) to access a wide variety of products from different online retailers (e-tailers). While beneficial to customers, it puts an extra burden on e-tailers to accurately predict the demand and optimize their inventory. Failure to do so will result in the e-tailer having an imbalanced inventory which could result in customer dissatisfaction and churn. In this paper, we propose a solution to this problem by using the notion of an e-tailer using personalized assortment planning to manage expected customer demand. Our proposed method (termed as Customized Assortment by considering customer Churn – CA&C) assists the e-tailer to identify customers who are at risk of churning, determine the assortment of products they need, and if these products are low in stock, then prevent them from being shown to the regular customers. Doing so, ensures that the assortment which is in limited stock is available for high valued at-risk customers since they could not replenish during the selling periods. CA&C uses mixed multinomial logit (MNL), survival analysis and dynamic programming (DP) techniques to estimate customer preferences, determine the risk of customer churn, the value of each customer and personalized assortment planning, respectively. Customers are grouped based on demographic information. In this study, we discuss how to use Customized Assortment as a lever not only to increase the revenue but also to maintain customers who are at risk of churning. We demonstrate the superiority of the CA&C policy, in saving at-risk customers while maintaining profitability compared to the two other policies, namely, the customized assortment (CA) and myopic decision policy by using a real-world data set.