In the current economic context, optimizing projects' cost is an obligation for a company to remain competitive in its market. Introducing statistical uncertainty in cost estimation is a good way to tackle the risk of going too far while minimizing the project budget: it allows the company to determine the best possible trade-off between estimated cost and acceptable risk. In this paper, we present new statistical estimators derived from the way IT companies estimate the projects' costs. In the current practice, the software to develop is progressively divided into smaller pieces until it becomes easy to estimate the associated development workload and the workloads of the usual additionnal activities (documentation, test, project management,...) are deduced from the development workload by applying ratios. Finally, the total cost is derived from the resulting workload by applying a daily rate. This way, the overall workload cannot be calculated nor estimated analytically. We thus propose to use Monte-Carlo simulations on PERT and dependency graphs to obtain the cost distribution of the project.
