AI-Powered Store Layout, Product Positioning, and Promotion Recommender Based on Operational Data

Abstract

Grocery stores operate in one of the most competitive and operationally demanding segments of commerce. With high product turnover, frequent replenishment cycles, and razor-thin margins, even small inefficiencies in shelf space allocation or product positioning can translate into meaningful revenue loss and inventory imbalance. The optimal arrangement of products within a store, which implies determining which items occupy which shelf, how much space each is allocated, and where they are positioned relative to one another, is a critical lever for maximising sales performance and streamlining inventory management. Traditionally, these decisions have relied on manual planning and the experience of category managers, leaving significant room for improvement through data-driven approaches. This dissertation presents an Artificial Intelligence-powered system that leverages a store's operational data, including sales forecasts, stock levels, and retailer-defined business constraints, to automatically generate accurate and actionable recommendations for shelf space allocation and product positioning. At its core, the system models the store layout as a structured planogram, representing shelves, rows, and individual product slots, and uses this representation as the foundation for an optimisation pipeline. The allocation engine operates in two phases. The first phase determines ideal facing counts for each product in isolation, using demand forecasts and stock data to identify products that are under or over-represented on the shelf. The second phase reconciles these targets against the physical constraints of the store layout (row widths, family restrictions, and locked spaces) formulating the problem as a Bounded Integer Linear Programme solved with the CBC optimisation solver. The objective function maximises demand-saturating utility across all shelf spaces, ensuring that high-demand products are systematically prioritised when competing for limited shelf real estate. The result is a system capable of producing shelf layout recommendations that are both commercially optimal and operationally feasible, directly supporting retailers in one of their most impactful, yet historically underserved, decisions.