A customer behavior-driven clustering method in the planogram design domain

The arrangement of products on supermarkets’ shelves (i.e., the planogram) has a strong impact on customer behavior, affecting their likelihood of choosing a product over another. Therefore, it is of fundamental importance, for retailers, to be able to quantitatively measure the differences between planograms. Most of the existing works in the planogram design domain are focused on compliance tasks, i.e., checking that a real-world shelf matches a given planogram. On the other hand, to our best knowledge, no work in the literature focuses on measuring the similarity between planograms. In this work, we leverage customer behavioral principles from the planogram design domain to build a data-driven clustering method, centered on customer behavior. More specifically, we achieve this by using a two-level clustering approach that, on the first level, discriminates between different shelf shapes, while on the second level, it discriminates the arrangement of products on the shelves. The proposed method allows planogram designers to efficiently modify and improve existing planograms, based on the desired customer behavior. Finally, our approach ensures domain coherence, avoiding bias propagation during the design of new planograms. The experimental results confirm that our quantitative approach correctly mirrors customer behavioral principles in the planogram design domain. • We propose a two-level agglomerative clustering in the planogram design domain. • We test our method on synthetic and real-world datasets from an Italian retailer. • We interpretable our results from a customer-centric perspective. • Our results show the effective applicability of the proposed method.