Product Choice with Large Assortments: A Scalable Deep-Learning Model

Abstract

Personalized marketing in retail requires a model to predict how different marketing actions affect product choices by individual customers. Large retailers often handle millions of transactions daily, involving thousands of products in hundreds of categories. Product choice models thus need to scale to large product assortments and customer bases, without extensive product attribute information. To address these challenges, we propose a custom deep neural network model. The model incorporates bottleneck layers to encode cross-product relationships, calibrates time-series filters to capture purchase dynamics for products with different interpurchase times, and relies on weight sharing between the products to improve convergence and scale to large assortments. The model applies to loyalty card transaction data without predefined categories or product attributes to predict customer-specific purchase probabilities in response to marketing actions. In a simulation, the proposed product choice model predicts purchase decisions better than baseline methods by adjusting the predicted probabilities for the effects of recent purchases and price discounts. The improved predictions lead to substantially higher revenue gains in a simulated coupon personalization problem. We verify predictive performance using transaction data from a large retailer with experimental variation in price discounts. This paper was accepted by Gui Liberali, Management Science Special Issue on Data-Driven Prescriptive Analytics.

Live shopping promotions: which categories should a retailer discount to shoppers already in the store?

“…However, important recent contributions (see Table 1) in the field of machine learning (ML) even allow for considering cross-product effects in a sparse form. ML approaches for product choice modeling show an exceptional performance and good scaling properties for targeted price promotions (Gabel and Timoshenko 2021;Jacobs et al 2016;Ruiz et al 2020). Ruiz et al (2020), for instance, model how customers select products and consider interactions between products and price changes.…”

“…Recently, Gabel and Timoshenko (2021) developed a deep-learning model at the product level to optimize the assignment of personalized in-store coupons. They find that their model performs particularly well when cross-category promotional effects are pronounced.…”

“…In particular, online marketers have the opportunity to deliver the right promotion (or content) to the right customer at the right point in time (Johnson et al 2013;Wamsler 2021;Zhang and Krishnamurthi 2004) due to customer-specific data collection. Instore coupon systems that provide customers with personalized price promotions for their current shopping trip-such as the CVS coupon center 1 -enable marketers to do the same offline (e.g., (Breugelmans et al 2015;Gabel and Guhl 2021;Gabel and Timoshenko 2021;Gedenk et al 2010)). Specific to both, online and instant redemption in-store coupons, is that the time since the last purchase in a category is a valuable source of information for regularly purchased categories, since it can be used to provide customers with well-timed promotions.…”

Transparent Model of Unabridged Data (TMUD)

“…Alternatively, one can employ rapidly advancing deep learning models that are capable of accepting these rich and unstructured data to generate meaningful managerial insights, such as predictive tasks. Unlike the traditional statistical approach, however, deep learning models operate as black boxes (Urban et al 2020, Xia et al 2019, Hartmann et al 2021, Gabel and Timoshenko 2021, as they are not designed to test theoretical insights. This problem is compounded by the sheer number of parameters that typically number in the millions or more.…”

“…Even if generating theoretical insights is not the goal, the inability to understand how they work reduces confidence and trust in these models, further limiting the adoption of otherwise fruitful applications. As a result, few publications in marketing journals report deep learning models as the primary methodology (Burnap et al 2020, Dew et al 2019, Gabel and Timoshenko 2021, Gabel et al 2019, Guan et al 2020, Hartmann et al 2021, Hu et al 2019, Li et al 2019, Liu et al 2019, Liu et al 2020, Malik et al 2019, Shin et al 2020, Timoshenko and Hauser 2019, Tkachenko and Jedidi 2020, Troncoso and Luo 2020, Xia et al 2019, Zhang and Luo 2018, Zhang et al 2021, Zhang et al, 2017 Here, we propose the transparent model of unabridged data (TMUD), which resolves this dilemma by integrating several existing analytic tools. A model of unabridged data (MUD) uses raw data as input which is typically rich and unstructured.…”