Plan-based systems attempt to achieve more flexibility by dynamically computing best-actions on the basis of available material and a library of strategies for achieving particular aims. This relies on the traditional AI notions of planning. The problem with applying this to dialogs is that the planning can be very complex and the time-constraints are high: an answer has to be produced before the user assumes that the system is broken...
One technique for producing an answer in time is to develop just-in-time algorithms. These are algorithms that can be stopped at any time to deliver a result. The result given at any time is guaranteed to be at least as good as any result produced earlier.