We study a new dynamic order acceptance and assignment problem of a make-to-order manufacturing or service system that produces heterogeneous products or services using heterogeneous servers. Unlike traditional dynamic order acceptance and assignment problems, in our settings the system can strategically postpone acceptance and assignment decisions for orders in hand. The system does this while waiting for more profitable orders to come and/or more productive servers to become available, with the risk of losing orders in hand, low server utilization, and high waiting penalties. We formulate this problem as a stochastic dynamic program, characterize the structure of the optimal policy, and discuss managerial insights gained from the optimal policy. The paper also provides a methodological contribution by finding general structural properties and their preservation conditions, which can be reused in related future operations management studies.