There are delivery regions that may experience high order quantities but insufficient courier availability in the process of the crowdsourced on-demand delivery due to varying order densities across delivery regions. Ant colony optimization algorithm based on robust evaluation and intelligent decision-making is proposed to address the delivery capacity uncertainty in crowdsourced on-demand delivery. First, the K-means clustering algorithm is used to divide the delivery regions. And order quantities are predicted for each sub-region using eXtreme Gradient Boosting. Second, a robust optimization model for courier dispatching is designed for crowdsourced on-demand delivery. The objective functions aim to optimize the ratio of orders to couriers, the dispatch distance and the number of couriers to be dispatched across the divided regions. Finally, an ant colony optimization algorithm based on robust evaluation (ACO-RE) is proposed to resolve the robust optimization model. Intelligent decision-making is performed on the obtained solutions to select the solutions with better robustness and optimality. The ant colony pheromone updating is guided according to solution robustness during the evolutionary process. Experimental results demonstrate that this method effectively improves the stability and service quality of the crowdsourced on-demand delivery system.