Improving Replenishment for Retail: Utilizing Planogram Information

Replenishment is the process of restocking inventory to maintain sufficient stock levels of products for sale. This involves analyzing sales data, monitoring stock levels and determining the optimal quantity of products to order from suppliers in order to meet customer demand. Each retail store needs to determine the ideal replenishment amount for each product, which can be challenging as it requires continuous manual effort and strategic decision-making expertise. In this paper, we present an end-to-end replenishment recommendation system specifically designed for frozen products in a retail company. Our approach combines historical sales data with computer vision based real-time product recognition on store shelves to recommend the optimal replenishment amount for a specific product at a specific store. We begin by forecasting the demand for the upcoming period using time-series forecasting methods. Then, we determine the initial replenishment quantity for each product in every store using a stochastic optimization model based on expected values. This model is designed to maximize profitability. This initial quantity is further adjusted by comparing planogram information with real-time product recognition from freezer images. Thanks to the high accuracy of over 90% in product recognition, our replenishment results have demonstrated an 18.77% improvement in total profitability for the selected products. Our end-to-end replenishment recommendation system’s outcomes are provided to sales representatives as suggestions in the order-taking phase and more than 70% of our automatic replenishment values are used without any adjustments.