Software Project Estimation has been one of the hot topics of research in the software engineering industry for a long time. Solutions for estimation are in great demand. By knowing the estimates early in the software project life cycle, project managers can manage resources efficiently. The objective of this paper is to investigate the estimation practices within an individual software company and to assess their reliability. We perform a methodical review of predictions from a within-company model, based on our analysis of their historical project data. We analyze their estimation practices and compute prediction accuracies and thereby, suggest improvements or modifications. The data analysis revealed that the company used expert judgment in the early years but gradually switched to parametric approaches (calibrated COCOMO II hybrid model). We describe our systematic review of the estimation process, perform experiments and analyze the results. Our findings suggest that these methods should be employed.