Profiling Online and Physical Supermarket Customers Using Factor and Clustering Methods

Profiling of customers allows businesses to address their needs with precision and to perform effective marketing actions. Profiling methods can be applied on questionnaire-based surveys or customer history data found in databases or log files. Machine learning techniques are able to capture consumer behavior and automatically perform profiling and targeted marketing actions, while data analytics and statistical analysis methods are more suitable as decision support tools. The aim of this paper was to extract the profiles of supermarket customers from their purchase history, giving emphasis in understanding their behavior and linking data-driven findings with known profiles from marketing theory. The analysis was conducted on a dataset that derived from the purchase records of 61 K supermarket customers over a rolling year. Data from both physical stores and e-shop were integrated with demographic data available through the loyalty program of the supermarket chain. The core methods utilized were a combination of multiple correspondence analysis (MCA) and hierarchical cluster analysis on principal components (HCPC). These methods were chosen for their excellent ability to discover trends and build easily explainable profiles, as well as to identify clusters based on a large number of qualitative variables. The analysis identified six supermarket customer profiles, which were associated with product preference patterns and features such as level of spending, loyalty, and promo-hunting. The profiles extracted by our data-driven methods were associated to profiles documented in consumer behavior research, suggesting potential marketing implications.