We address the problem of building a decision model for a specific bidding phase in the game of Bridge. We propose the following multi-step methodology. We first build a representative example set for the decision problem at hand and use simulations with a black-box solver to associate a bidding decision to each example. Then, supervised relational learning builds an explicit interpretable model fitting the blackbox model as well as possible. Interactions between game and machine learning experts that evaluate and gradually improve the successive models produced make this methodology successful.