Improving the Customer Grouping Algorithm for Spare-Parts Distribution Using k-Means Tools

Customer classification using k-means algorithm for optimizing the transportation plans is one of the most interesting subjects in the Customer Relationship Management context. In this paper, the real-world data and information for a spare-parts distribution company (ISACO) during the past 36 months has been investigated and these figures have been evaluated using k-means tool developed for spare-part demand similarity function in different regions of the country. Similarity function for customer behavior in different regions has been defined. Based on this function and with help of k-means algorithm, customers have been grouped and similar customers were put in the same groups. Customer similarity function has been developed through 5 steps and has been defined individually based on each factors of Euclidean distance, customer's order time and bulk value of the order. Then, these three factors have been combined and DCB function has been defined. In the final step, different weights have been allocated to different years and seasons and BCD function has been improved. The grouping process has been improved for three functions of Euclidean distance, DCB and BCD. This process was executed using the R software and the improved BCD function was recognized as the optimum grouping function. Then, using DMT model, customer behavior has been analyzed at each part and the proper distribution policies have been defined. Results indicate a significant cost reduction (32%) in spare-parts distribution costs for ISACO