Introduction of AI-based sales forecasting: how to drive digital transformation in food and beverage outlets

Abstract

Abstract Food and beverage (F&B) outlets such as restaurants, delis and fast-food joins are commonly owner-operated small businesses with limited access to modern forecasting technologies. Managers often rely on experience-based forecasting heuristics, which face challenges, as demand is dependent on external factors such as weather conditions, holidays, and regional events. Introducing practical AI-based sales forecasting techniques is a way to improve operational and financial planning and automate repetitive forecasting tasks. This case study proposes an approach to work with F&B owners in creating and introducing machine learning (ML)-based sales forecasting tailored to the unique local aspects of the business. It enhances demand forecasting in the F&B domain by identifying data types and sources that improve predictive models and bolster managerial trust. The case study uses over 5 years of hourly sales data from a fast-food franchise in Germany. It trains a predictive algorithm using historical sales, promotional activities, weather conditions, regional holidays and events, as well as macroeconomic indicators. The AI model surpasses heuristic forecasts, reducing the root mean squared error by 22% to 33% and the mean average error by 19% to 31%. Although the initial implementation demands managerial involvement in selecting predictors and real-world testing, this forecasting method offers practical benefits for F&B businesses, enhancing both their operations and environmental impact.

Keywords

• Sales forecasting
• Extreme gradient boosting
• Weather data
• Time series data
• Foods and beverages
• Real-time API