Abstract:

This study examines several time-series approaches to forecast a food company’s monthly sales in Mexico for 2025. A total of 28 models were tested and compared using different error measures to determine which provided the most accurate results. Producing dependable forecasts is important for supporting financial decisions and for identifying unusual patterns in a timely manner. Among the methods reviewed are ARIMA and Croston, both implemented within the CRISP-DM methodology. This approach enabled selecting the most suitable model for each branch based on historical data. Results show that the ARIMA model applied to Branch 7 achieved the best relative performance, with a -14% relative improvement compared to the other evaluated models. This model demonstrated a strong ability to capture patterns in historical sales and generate consistent estimates. The findings highlight the importance of employing robust tools for inventory management and strategic planning. Proper selection of the forecasting model not only improves the accuracy of the estimates but also enhances their usefulness for business decision-making.