Most maintenance-optimisation models assume an infinite planning horizon and suppose that the failure process is stationary. Hence, information which is not known beforehand and which beocmes available in the short term only, must be ignored. We consider in this paper a multi-component system with economically dependent components, and we compare the costs of a stationary-planning method with the costs according to an approach which can adapt this long-term plan to dynamically changing information (such as a variable use of components and the occurrence of maintenance opportunities). With numerical experiments we show that incorporating short-term information can yield considerable cost savings.