Abstract:
Problem: Retailers capture extensive products and store sales data but still manually group stores and design planograms for shelf space, relying on subjective, outdated methods. These department-by-department processes often miss real-time sales patterns and similarities in product movement, leading to inconsistent clustering, inefficient planograms, and lost opportunities to optimize inventory and promotions. An automated, data-driven approach is needed to enhance store clustering and planogram recommendations using both historical and forecasted sales.
Methodology: This research introduces a modular analytics system that integrates advanced time-series forecasting (Prophet) with automated, department-level store clustering using detailed sales patterns. The system combines forecasts and historical sales to guide product allocation, while its automated clustering analyzes granular sales data to group stores by department, eliminating subjectivity and improving local planogram relevance. Managed through a scalable, cloud-based Python workflow, it delivers actionable recommendations for product and supplier hierarchies across all key levels.
Initial Results: When applied to large-scale retail data (~60 million records), the time-series models achieved mean absolute error rates of 20–25% for stores and products. Experts confirmed that the automated clusters were clear and useful, and the new planogram recommendations consistently outperformed manual methods in accuracy and efficiency. Planners experienced less manual work and quicker decisions, and stakeholders validated the system as practical and business ready.
