Machine Learning-Driven Energy Demand Optimization Reduces Costs by 70% and Credit Back Losses by 45% for Leading US Energy Provider