Data Science in Retail: Pricing Optimization and Customer Engagement

Data science is widely applied across multiple operational domains of retail companies, including merchandising, supply chain optimization, logistics, quality control, store site selection, etc. This talk will focus on the use of data science for two core challenges of a retail business: pricing products and engaging customers.



Pricing products is guided by multiple considerations, including optimizing profit margins, cash flow, and inventory sell-through rate, setting brand perception, and avoiding cannibalization. To permit taking these diverse goals into consideration, each individual product must be priced in conjunction with the entire range of products carried. Such multi-objective optimization for a wide range of products is a computationally intensive task. We propose a two-level approach to this problem. First, we create a sensitivity model that predicts the demand for each product under a specific set of conditions, and the price elasticity for that product. Then, a Bayesian optimizer utilizes the sensitivity model for setting the optimal price for each product under a set of constraints that is selected based on the business goals. The computational efficiency of the optimizer is improved by splitting the optimization into a warm-up phase to find better initial estimates for product prices and a price refinement phase.



Customer engagement relies on the knowledge of individual customer preferences that is never exhaustive. Using a synthetic dataset, we explore approaches for optimally targeting customers when the information that is available about them is incomplete. We examine how observing customer responses to promotional offers may be used for extracting additional information about customer preferences and for grouping customers into cohorts for marketing campaigns.