Exploring Opportunities for Cross-Selling and Demand Forecasting in Retail Using Explanatory Graph Analytics

Abstract

Understanding the factors that drive the co-purchasing of products is a key challenge in the retail industry. Traditional methods for quantifying product complementarity are often inadequate, as they fail to capture the dynamic nature of customer purchasing behavior. To address this problem, we propose Co-Purchasing Exponential Random Graph Model (CPERGM), an explanatory graph model for capturing co-purchase behavior in a retail context. CPERGM is trained using a co-purchase network (CPN) trained on a large corpus of market basket data. We use CPERGM to explain changes in product co-purchase behavior over time in terms of known measurable variables. Thereafter, we propose complementarity propensity scores (CPS) to quantify non-traditional complementarity relationships. We evaluate our proposed graph modeling artifacts using a quarter's sales history of a large brick-and-mortar store in Canada and show that our graph analytics methods effectively explain co-purchase patterns and significantly improve the accuracy of demand forecasting models to inform merchandising decisions.

Keywords: Demand Forecasting; Retail Analytics; Social Network Analysis; Network Modeling; Retail Intelligence; Statistical Application; ERGMs