This study focuses on optimizing the supply chain at Company A, an industrial gas manufacturing company, by enhancing Demand Forecasting (DF) and solving the Inventory Routing Problem (IRP). Faced with significant challenges, the company seeks to improve operational efficiency and customer satisfaction through accurate forecasting and efficient inventory routing. Six methodologies, including traditional time series techniques (ARIMA, SARIMA), machine learning techniques (RF, XGBoost), and deep learning techniques (LSTM, ANN), were evaluated for Demand Forecasting, with Random Forest achieving the best performance (16.07, 2.86, and 0.53% in RMSE, MAE, MAPE, respectively). The forecasted demand, monitored over fifteen days, was then used to address the IRP through a mathematical model and an Iterated Local Search (ILS) heuristic, resulting in logistics ratio reductions of nearly 76% and 26% comparing to current state. The findings demonstrate significant improvements in forecast accuracy and potential cost savings in logistics and inventory, offering the company valuable insights for overcoming logistical challenges and contributing to the broader field of supply chain optimization in the air industry.