This paper describes an operational model used within the scope of a budget process for a warranty extension product. The model is built on a database containing a record of the costs incurred with several thousand vehicles over a period of five years. Each component was modelled in two different ways: the classic way using linear or intrinsically linear models by means of regression and traditional distribution functions; and an alternative way based on neural networks. The transition from a calibration achieved by traditional methods to one achieved by means of neural networks does not in any way affect the model itself. On the other hand the tremendous flexibility of neural networks has significantly improved the models predictive capabilities without inasmuch adding to its complexity.