The combination of broadcast and on-demand data delivery services is an economic way to build a highly scalable wireless information system with limited bandwidth. The use of data broadcasting should be adaptive so that the system response time can always be minimized. A traditional approach requires the development of a system response time equation in order to find the optimal solution. However, obtaining such an equation is not always possible.We observe that by maintaining a certain level of on-demand request arrival rate, a close approximation to the optimal solution can be obtained. Using this approach, a real-time adaptive data delivery algorithm is developed. Our algorithm does not require the access information of the data items to be known exactly, which is needed normally for this kind of optimization problems. A simple and low overhead bit vector mechanism is able to capture the relative popularities of the data items. With this information, our algorithm can give a performance comparable to the ideal case in which the access information for each data item is known exactly.