With the blossoming of online shopping, a lot of goods need be distributed to customers. For a logistics company, how to improve the delivery efficiency, reduce the logistics cost and satisfy the real logistical constraints including vehicle's capacity and customer's available time window is an important issue. This kind of problem is called Vehicle Routing Problem with Time Windows (VRPTW). Although a number of algorithms based on artificial intelligence have been proposed, most of them cannot efficiently solve VRPTW while the number of goods increases rapidly. In this paper, we propose a novel approach named Pool-based Recursive Constructor (PRC) to efficiently find a set of logistics routes by considering real logistics constraints. In PRC, an urgent value measurement, several customer selection strategies and a pool-based mechanism are proposed to evaluate the cost of each customer and select the most suitable customer for route constructing, recursively. Through the experimental evaluation based on two semi-real logistics datasets and comparison with two greedy strategies used by Kerry TJ Logistics, PRC shows an excellent performance in terms of route quality.