<p>The electric vehicles routing problem (EVRP) is a highly studied topic due to the “mileage anxiety” of electric vehicles (EVs). Unfortunately, the demands analysis for EVRP are hardly addressed. In this work, the energy-saving demand is discussed and the energy-consuming model is proposed. Based on the strategy of time disturbance, integrating static and dynamic data, the EVRP process is modified into a three-stage routing planning to meet the dynamic customer demand. Furthermore, the energy-consuming model and its objective function to analyze the cost-saving demand are improved. In the optimized heuristic algorithm based on demand analysis, the Q-learning algorithm is used to obtain a high-quality initial population and the two-step genetic algorithm (GA) is proposed to optimize the routing. The first step of the two-step GA is to use improved GA to meet the energy-saving demand and the cost-saving demand, the second step is to use the traditional GA to complete the dynamic routing of the three-stage routing with the dynamic demand. Compared with the traditional GA, the combination of the Q-learning initialization and the two-step GA reduces 36.68% in energy consumption. The two-step GA can quickly output different dynamic routes for dynamic customer demand at different times. In cost-saving demand, the optimization curve of the two-step GA versus the traditional GA converges 4 times faster and saves 2.84% in cost. This research highlights the necessity of the demand analysis for routing planning and demonstrates the critical role of the two-step GA in demand optimization, providing inspiration for responsive and sustainable EVRP.</p>