Optimal control techniques can actively maximize buildings efficiency by a smarter HVAC (Heating, Ventilation and AirConditioning) systems operation. All these techniques involve a model of the controlled system and an optimization process based on a cost function. The main difficulty with these methods is to find an accurate model of a system with the right data for training and on-line operation. Our work introduces an ad-joint based method aimed at efficiently computing an optimal predictive command law using a descriptive thermal model of a test case. Our approach has the advantage to enable fast analysis in order to identify the most influential inputs and parameters with respect to the final performance. In this paper, we present numerical results concerning a simple two-room test case.