Teams are important and popular working units in our society. When a large amount of team member candidates is available, such as in large enterprises, the composition task becomes very complex. For this purpose, a number of algorithmic team recommendations have been investigated during the past several years. These approaches are, however, created for specific domains. In this paper we present a novel generic approach for recommending teams that is able to employ best practices or results derived from studies on team composition denoted as team composition models or strategies. Furthermore, the proposed approach allows for the usage of existing statistical learning algorithms to adapt and refine these strategies for improving the recommendation quality. To show the applicability of our approach, we conducted a team work experiment in the domain of computer supported creativity and evaluated our recommender with the collected data set.