Distributors of contemporary coffee equipment must control inventory costs while juggling erratic demand and unpredictable supplier lead times. In order to enhance the ordering of replacement parts for commercial coffee makers, this study suggests a data-driven approach that combines machine-learning predictions with traditional inventory optimization. Historical sales and purchase-receipt records were combined into a single dataset after being cleaned up and enhanced with calendar and event characteristics. TensorFlow was used in Google Colab to train two feed-forward neural networks: one for daily demand and one for supplier lead time. The demand model's Root Mean Square Error (RMSE) on held-out data was approximately1.8 units with R² value of 0.75, and the lead-time model's was 15 days with R² value of 0.81. Forecast results are directly entered into the calculations for safety-stock, reorder-point, and economic-order-quantity. Planners were able to rapidly see the effects of changing service levels, warehouse restrictions, and cost criteria on order size, inventory value, and overall yearly cost using an interactive dashboard. In comparison to the company's previous policy, the results indicated an anticipated 18% reduction in working capital tied up in stock and a 12% reduction in stock-out instances.