Predicting project trouble using machine learning, artificial intelligence and historical project status reports

Abstract

The discipline of project management has evolved over the years, yet projects still run into trouble, failing entirely, running late, or not delivering expected benefits. It is challenging to predict troubled projects, especially within organisations that are running many concurrent projects. Program and portfolio managers may not have the skills, tools, processes, or bandwidth needed to identify and predict risks before they reach their full negative potential which reduces the time they might have to bring those projects back on track. This research sets out to explore mitigations that may improve this situation through the design, development and experimental deployment of an AI-based portfolio risk emergence prediction capability intended to proactively monitor the health of individual projects within large project portfolios and send alerts when trouble is predicted. This research is located at the complex nexus of risk management, project portfolio management, artificial intelligence, human-computer interfaces (HCI) and a range of human phenomena including trust, and response to warning messages.

This research contributes to knowledge by firstly confirming the utility of project status reports as input for predicting project trouble, providing a new starting point for the design of future AI-based systems intended to monitor project health in real-time. Secondly, this work extends the risk management and early warning signs literature with the introduction of automated in vivo project health prediction. Thirdly, this work contributes empirical research of optimal time series algorithms for a project health prediction use case, which may be of use to future AI and project management researchers. Finally, it joins the Human AI Interaction (HAII) discourse around the adoption and use of AI-based monitoring systems.