In an increasingly demanding business environment driven by cost reduction and increased customer service through inventory management, this research develops a data-driven methodology for segmenting inventory and defining positioning strategies within a multi-tier supply chain. The study is applied to a business unit of a leading Latin American company in the manufacturing and retailing of household products. Stock keeping units (SKUs) were segmented using unsupervised machine learning algorithms, supported by dimensionality reduction techniques, to classify items according to demand behavior and supply characteristics. Based on this segmentation, an inventory positioning strategy was defined, assigning SKUs to the most appropriate tier of the network, such as regional distribution centers, urban last-mile distribution centers, and retail stores. Each tier was complemented by a hybrid push-pull strategy tailored to the predictability and variability of SKU demand. This alignment enabled product flow design that operates simultaneously within a shared physical infrastructure. The methodology was validated by applying it to the 701 most business-critical SKUs. Prior to implementation, the inventory network exhibited significant imbalances: 81% of stock was concentrated in stores, 17% in urban last-mile centers, and only 2% remained in regional distribution centers. This configuration resulted in persistent overstocking in stores, slow inventory turnover, and limited responsiveness of certain products. After implementing the proposed strategy, inventory was reallocated to 58% in regional distribution centers, 8% in urban last-mile centers, and 34% in stores. This redistribution significantly improved service levels, reduced lead times, minimized overstocking, and improved overall inventory efficiency. The results highlight the strategic value of integrating machine learning with supply chain network design to build more agile, efficient, and profitable operations.