Decision problems are often characterized by complex criteria dependencies, which can hamper the development of an efficient and theoretically accurate multicriteria decision aid model. These criteria interactions have the form of either a redundancy or synergistic effect and require arduous and demanding preference statements for their quantification. This paper investigates interactions between pairs of criteria in decision models and addresses them with the proposition of an MCDA framework, coupling the elicitation protocol of the method of cards and the 2-additive Choquet integral preference model. An interactive robustness control algorithm ensures the concurrent acquisition of a stable decision model and satisfactory evaluation results. Robustness is assessed with a portfolio of robustness indicators, spanning from the variability of the preference parameters to the reduction of the model's feasible space and rank acceptability indices. At the core of the algorithm, a heuristic module generates pairwise elicitation questions and selects those delivering the highest expected information gain. The whole framework is stress-tested with a small-scale decision problem, where three versions of the heuristics are automatically applied, with the machine randomly answering the questions. Subsequently, the same problem is approached with the involvement of a real decision maker, with a view to appraising the required cognitive effort and receiving valuable feedback.