Recently, Process-Aware Information Systems (PAIS) were introduced, which allow for dynamic process and service changes. This, in turn, has led to a large number of process model variants, which are difficult to maintain and expensive to configure. This paper deals with goals and issues related to the mining of process model variants. Our overall goal is to learn from process changes and to "merge" the resulting model variants into a generic process model in the best possible way. By adopting this generic process model in the PAIS, future cost of process change and need for process adaptations will decrease.