Anticipating customer substitution: a data-driven, distance-based approach for out-of-stock product configurations

Purpose When facing capacity bottlenecks, manufacturers of configurable, multi-variant products may adjust the product mix to uphold the scheduled output. However, maintaining market attractiveness by choosing the right product configurations as substitutes is a non-trivial task as it involves anticipating the substitution behaviour of customers. Substitution behaviour models currently used in quantitative production planning models for configurable products are either based on domain knowledge of experts, which makes them bias-prone, or they require extensive market research. The purpose of this study is to present a data-driven approach. Design/methodology/approach Based on data science concepts, distance measures are applied to derive distances between different product configurations from historical order data. Different design options for such a distance measure are discussed regarding configurable products and tested with automotive industry data. Furthermore, the study shows ways to validate the distance results. Findings The experiments show that the presented distance measure represents the expected customer substitution behaviour quite well. A context-sensitive distance measure including rank information of ordinal product features is most suitable for the automotive data sets. Originality/value This study presents a new approach to model the substitution behaviour of customers. The attractiveness of a potential substitute is represented by a distance from the customer’s first-choice configuration. The presented distance measure provides an inexpensive tool using existing data instead of expensive market research. Thus, it supports the integration of substitution into quantitative production planning models that deal with a large variety of configurable products.