Classification and Clustering Approaches to Understanding Customer Ordering by Customers of a Fresh Food Supplier

Purpose: This paper looks at characterization of B2B customers of a fresh food wholesale company supplying SME clients in terms of their weekly orders of a variety of fresh products. Customers whose orders can be predicted (days of the week order is placed, size of order) can easily be supplied without risk of waste due to the wholesaler ordering stock that is not sold to customers before it must be disposed of. Greater understanding of customer order patterns is necessary to improve demand prediction and reduce waste. Research Approach: Extensive real-world data from a fresh food wholesaler has been analysed in bulk. Customers’ weekly orders have been classified into one of nine classes depending on how each week’s order compares to the previous week. Equal order amounts on the same day (or days) of the week as the previous week are the most predictable class. Varying order amounts for orders placed on different days of the week are a much less predictable class. Other classes represent customers who either cease ordering after having made previous orders, or who place an order after not ordering in previous weeks. K-means clustering has also been used to extract clusters of customers showing similar ordering patterns from the customer base. These functions have been integrated into a data visualization tool which displays the clusters in terms of the frequency of occurrence of order classes, and their standard deviation within the clusters.