Warranty modelling with incomplete data is a major issue in reliability analysis. The incomplete failure region characterized by warranty field data may be classified into several domains representing failures from manufacturing/assembly defects, usage or fatigue. In the present paper a data driven approach has been suggested to demark the regions optimally through estimation of the change point in a hazard function. In the perspective of bivariate warranty analysis, as relevant in automobiles, we have assumed the lifetime distribution to be a mixture of distributions corresponding to the burn-in period and the useful life period. The proportions of observations in different regions demarketed by the warranty policy in the bivariate plane have been estimated by considering mileage along with age. The estimation scheme has been verified and validated through extensive simulation studies. The utilities of the results have been demonstrated by addressing several issues through a real life synthetic warranty data set from a large automobile company. © 2014 Elsevier Ltd. All rights reserved.