This paper investigates a dynamic food delivery problem, in which orders arrive dynamically and must be delivered to customers within specified time windows. A fleet of capacitated vehicles is used to deliver the orders. The goal is to minimize the total transportation distances associated with fulfilling customer demands on time. The authors propose a methodology for solving this problem, which involves two steps: 1) Order grouping: the orders are grouped into sets based on their arrival time and delivery time windows. 2) Route optimization: for each set of orders, an optimal delivery route is found using a CVRPTW model. We evaluate the methodology using experimental results. We found that the decision timeframe (the length of time over which orders are grouped) has a significant impact on the performance of the methodology. When the decision timeframe is narrower, it tends to require a higher number of vehicles compared to wider time frames. However, it also results in a shorter average waiting time for customers. With a wider timeframe, more options for orders can be combined in one delivery routing, which can reduce the travel cost/distance. However, too wide a timeframe may lead to increased travel distances due to a small gap between the time of decision making and the guaranteed time of delivery.