Production planning in the food industry faces several challenges due to the absence of current technology and appropriate forecasting techniques. This study proposes to improve forecast by comparing machine learning algorithms and time series method. Three forecasting methods were evaluated: i) Crystal Ball, using the SARIMA model; ii) Weka, using neural networks; and iii) Machine learning models, created in Python, with algorithms such as Random Forest (RF). Error metrics such as MAE, RMSE, MAPE and \(R^2\) were evaluated, in addition to a survey based on the System Usability Scale (SUS) to assess ease of use. The case study was developed in a company that uses traditional tools such as Microsoft Excel. The results revealed that Weka was the most accurate in product 001002 with a MAPE of 1%, RMSE of 8.66 and MAE of 6.58. In terms of usability, Crystal Ball achieved the highest score (70 points), outperforming the other two proposals, indicating that the software selection is based on a balance between accuracy and ease of use. The incorporation of sophisticated models enhances demand anticipation, improving inventory management and operational efficiency.