This paper is devoted to preference-based recommendation or configuration in the context of multiagent (or multicriteria) decision making. More precisely, we study the use of decomposable utility functions in the search for Choquet-optimal solutions on combinatorial domains. We consider problems where the alternatives (feasible solutions) are represented as elements of a product set of finite domains and evaluated according to different points of view (agents or criteria) leading to different objectives. Assuming that objectives take the form of GAI-utility functions over attributes, we investigate the use of GAI networks to determine efficiently an element maximizing an overall utility function defined by a Choquet integral.