We study a configure-to-order assembly system consisting of multiple parallel subassembly stages and one final assembly stage. Each stage has a stochastic leadtime. The system is controlled by planned leadtimes at each stage. Planned leadtimes are used to plan start and finish times at all stages. The system incurs holding cost for each stage from the planned start time of the stage until the final product is delivered to the customer. In addition, a penalty cost is incurred if the final assembly stage is late. The objective is to set the planned leadtimes for each stage such that the total expected cost is minimized. We prove that the optimal planned leadtimes satisfy a set of Newsvendor equations and show that the probability that a stage is “blamed” for the lateness of the system equals a Newsvendor fractile. We derive structural properties of the optimal solution and compare it with the optimal solution under an alternative holding cost regime.