Search

Existing research which has attempted to determine differences between the costs of the different procurement routes has consistently aimed to determine a single figure for the difference for projects as a whole. No attempt has been made to provide a difference which is project specific (Duff et al., 1998). Furthermore, no previous research has determined the cost to the client using any objective method. The absence of such a technique is significant. It means that the client’s advisors have no means of providing an objective measure of the cost of following different procurement routes. The client must depend upon the judgement of the advisors, which is based on their own perception of both the project and the different procurement routes, and is hence subject to their opinions and prejudices. This paper reports on the development of a neural network model (ProCost) which is able to determine the total cost to the client of a project, it functions as a decision support tool by enabling the project specific comparison of alternative procurement routes and other strategic decisions.

Neural networks are information processing systems that can automate semi-structured knowledge domains. Neural networks complement and enhance the capabilities of conventional systems in order to increase productivity in a given domain. Research into the application of neural networks in the construction industry indicates the maturity of the technology and the system development approach. In order to have practical application and contribute to the notion of wealth creation in society, research must be focused on transferring the know-how acquired through research to the industry. The paper recommends involving the end users throughout the system development process, and creating an environment within which direct and indirect learning can be achieved by practitioners. An example of this approach, which develops a decision support system for the decision to bid process, is described and evaluated.

Existing research that has attempted to determine differences between the costs of following different procurement routes has consistently aimed to determine a single blanket figure, such as “design and build is 15% cheaper than traditional”. No attempt has been made to provide a difference which is project specific (Duff et al., 1998). Furthermore, no previous research has determined the total cost to the client using any objective method. The lack of data defining client costs and the absence of suitable modelling techniques have prevented such an objective evaluation being made (Masterman, 1994). These factors prompted research into the cost of procurement at UMIST. This research has required the collection of a substantial database of the total cost, to a client, of past projects, and the subsequent creation of a neural network model of these costs. Results of the first phase of development of this model are presented, including regression analysis, preliminary neural network models and sensitivity analysis. An assessment of how these results will inform future development of the model is also made.

Existing research which has attempted to determine differences between the costs of the different procurement routes has consistently aimed to determine a single figure for the difference for projects as a whole. No attempt has been made to provide a difference which is project specific (Duff et al., 1998). Furthermore, no previous research has determined the cost to the client using any objective method. The absence of such a technique is significant. It means that the client’s advisors have no means of providing an objective measure of the cost of following different procurement routes. The client must depend upon the judgement of the advisors, which is based on their own perception of both the project and the different procurement routes, and is hence subject to their opinions and prejudices. This paper reports on the development of a neural network model which is able to determine the total cost to the client of a project, as well enabling the project specific comparison of alternative procurement routes.

The choice of procurement system for a building project is a very significant one. However, there is currently very little comparative cost data to inform the selection of procurement system, especially the total cost to the client. This paper reports on a model that will make such a comparison possible. This model, which is currently under development at UMIST, is designed to consider 39 project variables, including the choice of procurement system, and estimate the final cost of the project using a neural network. The suitability of a neural network to model this problem has already been established by a pilot study. The advantage of using this type of model is that it permits comparisons between the different procurement methods to be made within the context of that particular project, rather than within projects as a whole. The classification and representation of the different variables to be considered within this model are discussed. In addition to this, the implementation of factor/cluster analysis to reduce the number of variables required by the neural network, and hence increase its accuracy, is explained. Furthermore, the level of confidence of the model and its implications for the implementation of a “What if?” analysis are discussed. This analysis would allow the client to assess how changing certain variables, including procurement, might affect the final cost of the project.

The impact of technology on all areas of science and industry in the latter half of the twentieth century has been enormous, and the building industry has been no exception. However, early stage cost estimation has been little affected by these advances in comparison to most of the rest of the industry, and is still largely based on a combination of simple models and professional judgement. This paper reports how research currently being undertaken at UMIST could change this practice significantly. A neural network model which is able to estimate the total cost of construction to the client is being developed. This model considers forty variables which define the project at this early stage. Some of these variables would not usually be considered within the estimation of the cost. Perhaps most significantly, by including procurement it is able to evaluate the cost of different procurement routes inclusive of the client’s administrative costs. How the neural network improves on existing modelling techniques by its use of existing project data is explained, and its role in early stage cost estimation in the 21st century outlined.

There is a dearth of information on the comparative costs of projects carried out using the main procurement building systems. This paper reports the feasibility study of a research programme to produce a computer-based neural network cost model to show the effect on client costs of ‘inter alia’ using different procurement approaches. A literature search identified 39 cost-significant project variables. Data were collected from collaborating QS practices, resulting in 46 project data-sets with which to test various modelling approaches. Evaluation of the data and model objectives identified multiple regression and neural networks as potential model forms. Investigation and trials of both have shown that regression and neural networks can provide effective representation of the client costs model but neural networks, due to their greater ability in modelling interdependencies between input variables, modelling non-linear relationships, and handling incomplete data sets, will probably be the better choice with which to analyse the very much larger volume of data planned for the next phase of the research. The results have demonstrated that such a model can be developed, that data to support it can be obtained and, additionally, that the utility of the model may be significantly greater than had been envisaged at the start of the study.