Digital Business Tranformation Through XGBoost-Based Sales Forecasting for Fast Food Franchise Inventory

Abstract

The rapid growth of the digital economy has pushed fast food franchise businesses to adopt data-driven decision-making to remain competitive, particularly in production and inventory planning. The Company, an Indonesian fast-food franchise operating more than 300 outlets, currently lacks an accurate sales forecasting system, resulting in frequent overstock (approximately 30 products per day) and understock incidents that trigger negative customer reviews and threaten business sustainability. This study develops a digital sales forecasting model to determine daily sales estimates, safety stock levels, and reorder points for two of the Company's outlets, referred to as Outlet A and Outlet B to preserve business confidentiality. Daily sales data from January 2023 to June 2024, combined with national holiday data, were processed using the XGBoost algorithm in Python and evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE); safety stock and reorder points were then calculated in Microsoft Excel. The model produced an average daily sales prediction of 435 products at Outlet A (RMSE 61.89, MAPE 10.2%) and 360 products at Outlet B (RMSE 56.2, MAPE 14.1%), both categorized as good forecasts. The resulting safety stock and reorder point values were 98 and 968 for Outlet A, and 120 and 840 for Outlet B. These findings demonstrate that digital forecasting technology can support more efficient inventory control and provide franchise businesses with a practical, replicable basis for reducing overstock and understock problems through the digital transformation of their operations