SEGMENTASI KATEGORI PRODUK BERDASARKAN RESPONSIVITAS PROMOSI PADA BISNIS FASHION RETAIL MENGGUNAKAN AGGLOMERATIVE HIERARCHICAL CLUSTERING

Abstract

Fashion retail companies often repeat identical promotion types without proportional sales gains because no empirical mapping shows which product categories respond to which promotional mechanisms. This study segments product categories by promotional responsiveness using Agglomerative Hierarchical Clustering (AHC) with Ward's Linkage. The data comprise 103.9 million sales transaction rows from 2023 at PT Matahari Department Store, Tbk., which, after cleaning, department-level aggregation, label encoding, and Min-Max normalization, yielded 6,089 records covering 790 departments, 14 promotion names, and six promotion periods specific to the Indonesian market. We determined the optimal number of clusters using the Silhouette Coefficient, Davies-Bouldin Index, and SSE reduction, and cross-validated it against the merge-distance jump in the dendrogram. Ward's Linkage consistently outperformed Average Linkage across all tested values of k, with the best configuration at four clusters (Silhouette 0.474; DBI 0.049). The four clusters exhibit sharply different responsiveness: one cluster of 317 departments contributes 99.11% of promotion-related units sold with a PromoHitRate of 0.986, while another cluster of 316 departments records no promotional sales at all. The near-uniform distribution of promotion types and periods across clusters indicates that the performance gap stems not from exposure bias but from each department's internal effectiveness. These segmentation results provide a basis for more targeted promotional budget allocation and more efficient inventory management.