Retailers frequently face stockouts and overstocking due to inaccurate demand forecasting, leading to financial losses and reduced customer satisfaction. This study proposes a data-driven framework to improve weekly sales forecasting at both aggregate and store levels using Walmart’s historical sales data. A hybrid methodology integrating time series models, regression techniques, deep learning, and a hierarchical structure is developed to capture temporal patterns and external demand factors. The proposed approach achieves high predictive accuracy, with a Mean Absolute Error (MAE) of 306,361.11, Root Mean Square Error (RMSE) of 528,096.34, and an R² of 0.99, outperforming traditional models. Beyond accuracy, the study emphasizes the role of forecasting as a decision-support tool. The results demonstrate that improved forecasts enable better operational decisions such as replenishment planning and safety stock optimization, while also supporting tactical and strategic decisions related to distribution, workforce planning, and supply chain design. Overall, the findings highlight that integrating hybrid forecasting models with decision-making processes can reduce inventory costs, enhance service levels, and support more efficient and sustainable retail operations.