Reaching a Consensus Among Construction Stakeholders: Defining Success and Benchmarking Performance

Abstract

Executive Summary Despite significant and increasing figures for expenditure and employment which are demonstrative of a successful and growing economic sector, construction still has an industrial struggle with the precise definition of a successful project, and lacks a universally accepted methodology for project performance assessment. It is not unheard of for a project to be simultaneously assessed as successful by one stakeholder, but less than successful or outright failure by another. Thus, there is a need for contractors, owners and other project stakeholders to assess and benchmark project performance to understand relative success and learn lessons which can be applied to future projects. To address this need, the research discussed in this thesis established a unified project performance score (PPS) to facilitate the comparison and assessment of projects, and also defined thresholds of success for various performance metrics that comprise the PPS. Data was collected from owner and contractor members of the Construction Industry Institute (CII) for various projects which encompassed four different delivery systems, three different project types, and three different compensation types. The data collected primarily consisted of key performance indicators (KPI) for the projects, from which six discrete performance metrics were derived. Additionally, project stakeholders’ degree of success for each project was assessed. To facilitate comparison of projects across cities, states, and time, a logistic regression model was developed and subsequently validated that returns the PPS. The model revealed that the two most predictive factors were Construction Cost Growth and Construction Schedule Growth – that is, they had the most direct correlation with the performance of the PPS. It was further found that in terms of the PPS, those projects executed under Integrated Project Delivery (IPD) outperformed all other delivery types, and those projects executed under a Target Value (TV) compensation model outperformed all other compensation models. To determine success status thresholds for construction projects, the modeling process consisted of two Classification and Regression Tree (CART) models. The first was developed using four-fold randomized cross validation to assess the threshold values of both successful and unsuccessful projects for the purpose of explanation. This first model has a classification accuracy of 81%, and determined that successful projects are those with less than 10.4% Construction Schedule Growth and less than 9.5% Construction Cost Growth. An additional recommendation regarding Change Order per Million Dollars with a maximum limit of 0.39, was found from the CART (1) model that separates high performing successful projects from other low performing but still successful ones. A second CART model, with selective tuning, was created for the purpose of prediction, with a classification accuracy of 85%, and to assess the three metrics not included in the first model. The model success thresholds were found to be nearly the same as CART (1) model for 2 out of 6 metrics, which are 10.4% for Construction Schedule Growth and 9.8% for Construction Cost Growth (as compared to 9.5% in CART [1] model). While, RFIs per Million Dollars maximum limit is 8.6 and that for RFIs processing time is 7.5 days (~1 week). Second category of results which separates high performing successful projects from low performing successful ones include values of Rework %. A high performing successful project will have a rework % less than 1.5%. These thresholds can be used as a standard to define and measure success on construction projects. One such method may be by using these as linchpins of risk and profit sharing (such as under an IPD model). Additionally, the bidding selection system can be revolutionized from a cost-based to a performance based one using the PPS model to improve the outcomes of the buyout process.