OPTIMIZING INVENTORY COSTS FOR FOOD PRODUCTS IN METRO KAZAKHSTAN THROUGH EFFECTIVE FORECASTING AND TARGET STOCK LEVEL EVALUATION

Abstract

Metro Cash and Carry Kazakhstan LLP, one of the biggest wholesale companies in Central Asia, needs robust and reliable demand forecasting techniques to maintain its position as a B2B leader in the market. Accurate forecasting algorithms are essential to plan inventory, manage supply chain operations, optimize sales, and ultimately improve customer satisfaction. By applying advanced demand forecasting techniques (machine learning and advanced statistical methods) the company can acquire valuable insights into market trends, empowering the business to take well-informed decisions and stay competitive in the market. The report demonstrates the results of a study that is aimed to improve demand forecasting for Metro Cash and Carry Kazakhstan. The study contrasted the company's current demand forecasting algorithm with statistical and machine learning approaches for different products and stores. In terms of RMSE for both products and stores, the statistical approach WMA, and the machine learning method LightGBM yielded the best results. Also, the study assessed the performance of WMA and LightGBM by forecasting demand on a daily and weekly basis for five products and three stores additionally. The findings showed that WMA was more accurate at forecasting weekly demand values than LightGBM but could not outperform the machine learning method on a daily forecasting basis in terms of RMSE. For the validation purpose of the forecasting results, the actual sales data was compared with the forecasted results by RMSE at every step of the project. Additionally, the forecasted demand data was used in the Simulation to calculate inventory costs in comparison with the company's forecasting method. Both WMA and LightGBM significantly saved inventory costs (ordering, holding, and stock-out costs). Moreover, the company's Target Stock Level (TSL) is compared with suggested TSLs while analyzing target stock levels for two product categories. As a result, the analysis showed that the proposed TSL was smaller than the company's TSL, which decreased the total inventory cost. Moreover, the proposed TSL was validated with actual sales of categories to ensure that the proposed TSL meets all customer demands. The project concludes that the suggested strategies can improve the accuracy of demand forecasts, hence increasing product availability levels, and decreasing total inventory costs, potentially enhancing competitiveness and economic returns. For future work, it was suggested to develop a general guideline for using proposed forecasting techniques for various food product categories. Similarly, it was recommended to expand the calculation of the proposed TSL to other local food categories.