Bayesian Integration of Internal and External Views in Forecasting Project Performance
Abstract
This paper focuses on the project control process and aims at improving the accuracy of the Estimate at Completion at Time Now, both in terms of cost and time. This objective requires the use of all the available information, in particular the information related to the actual performance of the current project, corresponding to the “internal view”, and the information related to the cluster of similar projects completed in the past, corresponding to the “external view”. In order to integrate both types of information, a Bayesian model has been developed, allowing for the updating of a prior estimate based on the external view by means of the data records collected during the progress of the current project, in order to obtain a posterior estimate of the final cost and duration of the project. This approach allows for the mitigation of possible biases which can affect the project control process, particularly at the project early stage. The Bayesian model has been applied to three cases in the Oil and Gas industry. Notwithstanding the great difference between the projects, the integration of the internal and external views in the Bayesian model resulted in a better accuracy compared to the traditional formulas used in the Earned Value Management approach and, moreover, a better stability of the estimates from the early stage along the entire life cycle of the project.