Combining Data Mining and Optimization: An Application in Retail Markets

Firms with multiple branches frequently set growth targets on an ad hoc basis, and disregard branch specific constraints or advantages. This leads to skewed management-set incentives and results, sometimes leading to lowered productivity from the branches. We use a nation-wide plastics wholesaler with over $500 million revenue to show how data mining (association rule mining – ARM) helps to find global knowledge from branch sales patterns. We develop metrics and algorithms to overcome well-known ARM problems. The metrics also help to compare branch performance and provide differentiated growth projections for managerial decision making. Finally we optimize by combining global rules from data mining with local market constraints of each branch. We identify stores that are at the top of their “game”, and those that can improve revenue up to 100% using updated product assortments. Our solution offers important insights for supply chain designers and merchandising managers on product portfolio selection, including complements vs. substitutes and product bundling.