Store segmentation using SAS clustering Baofu Ma Merchandising AUTOZONE ANALYST,MERCH RESEARCH
2 Introduction Motivation: Need to create similar business model for stores with either similar product sales or customer GBB brand preference. And explain these clusters in terms of demographic variables. Challenges: Business rule requires that the store cluster size has to be greater than certain number. Enforce a minimum cluster size with proc cluster. Explore the relationship between the clusters and demographic variables.
3 Overview of hierarchical clustering Each observation begins in a cluster by itself. The two closest clusters are merged to form a new cluster. Using Proc cluster to get the tree. Using Proc tree to get the desired cluster. 6
4 Solution Get the history of the clustering process using ODS. ods output ClusterHistory=history; proc cluster data=indatatemp METHOD=ward outtree=Tree; Search clusters which satisfy the minimum size criteria from top to bottom.
5 Example Classify autozone stores based on market share of 2 oil brands, high mileage and blends. Business rule requires minimum cluster size is 300.
6 Example Even borders. 1. Find centers of each cluster. 2. Calculate distance between store and each cluster center. 3. Reassign store to the closest cluster.
7 Example Relationship between clusters and demographic variables. Blue-positive Orange- negative
8 Thank you!