Since several years, great distribution firms implement more and more complex layout and shelf allocation strategies, so as to force empirical know-how to combine with Artificial Intelligence tools. Thus simulation has become an essential tool for designing efficient article layouts. Mathematical models based on statistical observations have been replaced by agent-based models. In this paper we argue that the modelling of individual behaviors of customers in a supermarket must be done through interaction-oriented approaches, in order to allow large-scale simulations and flexible models. Our model, based on our IODA methodology, is part of a ``Serious Game'' dedicated to vendor training. It takes into account the diversity of customer preferencies, the location of articles, and the side-effects of customer moves, so as to measure the consequence of management choices on the global outcomes of the supermarket.