Segmentation of store customers to increase sales using ABC-XYZ-analysis and clustering methods

Abstract

To manage interaction with customers, it is necessary to divide them into groups depending on their buying activity. For this, the author uses segmentation (ABC and XYZ analysis) and clustering (K-means, X-means, Expectation-Maximization) methods in the intelligent software platform RapidMiner Studio. For calculations, data on sales of a store selling auto parts are used. The author notes such features of the trade in truck parts as the seasonality of maintenance and the irregularity of purchases in comparison to general-purpose goods, which affect the choice of the time interval in sales data. The ABC-XY-analysis showed that it is better to use only the results of the ABC analysis, taking into account the volume of purchases, and the XYZ-analysis based on the analysis of the frequency of purchases gives the wrong result due to the seasonality of purchases. The results of clustering using K-means and X-means methods are almost the same, but their disadvantage is that 95% of customers are classified as the cluster with the worst purchasing activity, and this separation cannot be used for marketing purposes. The Expectation-Maximization method gave the best division of the client base into clusters.