We develop a demand postponement mechanism to improve the performance of a single item, periodic review inventory system with advance demand. The focus in the literature has been on how to stimulate customers towards advance demand. Predicting how demand will shift can be problematic, however, and backorders may still occur. We focus on how a firm can address backorders under a given advance demand pattern by a mechanism of compensation from which both the firm and the customers will benefit: the firm may offer a discount to customers for accepting later deliveries at a promised delivery date. Delivery postponement offers are made selectively, i.e. in some periods and to some customers only when there is a benefit for the firm to do so. Customers may decline the offer, but then face the probability of a backorder. In each period, the firm has to decide whether to make delivery postponement offers and for how long, and whether to order from its supplier and how much. We formulate the problem as a Markov Decision Process and solve it by backward induction. Numerical examples illustrate the properties of the state-dependent policies obtained for both uncapacitated and capacitated inventory systems. The postponement mechanism in capacitated systems leads to policies that differ from the threshold policy identified as optimal in the literature. Overall, the approach shows promise to improve system performance more efficiently compared to strategies aiming to increase advance demand in the system.