In today’s highly competitive and volatile environment, supply chains in emerging economies face ongoing challenges related to inventory management, demand forecasting, and distribution efficiency. This research proposes a predictive approach based on machine learning, specifically using ensemble stacking techniques, to optimize key logistics processes. Real-world data from a commercial company was used to develop a predictive framework that integrates various base algorithms Random Forest, CatBoost, XGBoost, Gradient Boosting, Decision Trees, and K-Nearest Neighbors combined through a Linear Regression meta-model. Performance evaluation using metrics such as MSE, RMSE, MAE, and R² revealed significant improvements in predictive accuracy compared to individual models, particularly in indicators such as material demand, purchase profitability, sales revenue, and inventory levels. The findings confirm that stacked models not only enhance forecasting capabilities but also offer a scalable, adaptable, and cost-effective solution to support logistics decision-making in resource-constrained contexts. This approach presents a strong alternative for boosting operational efficiency in supply chains across developing regions.