Project portfolio management for AI projects. Developing a framework to manage the challenges with AI portfolios

Abstract

Artificial intelligence (AI) has rapidly developed during the past decades, opening the doors to new opportunities for many organizations. The automotive industry is no exception, and AI is regularly seen as a leading technology for many disruptive trends. Thus, successfully incorporating AI as a capability in the organization is essential for future competitive advantage and growth. However, many organizations still experience challenges reaping the fruits that scaled AI initiatives present. Some of the challenges for reaching enterprise-wide implementations of AI include the difficulty of selecting which initiatives to scale, and communicating the business value of these. Similar problems have previously been tackled within the research field of project portfolio management (PPM). Indeed, previous literature has a history of adapting PPM practices to suit the needs of new project types. However, the portfolio management literature on AI projects is seemingly nonexistent up until this point. In collaboration with researchers and a newly established department at Volvo Cars, responsible for applying and diffusing AI in the organization, this study sets out to develop a new framework for the PPM practices of AI projects. To fulfill this aim, two research questions are investigated in this thesis. The first question delineates the main characteristics of AI projects from a portfolio perspective. The second research question explores how PPM practices need to be customized to support these characteristics. Regarding the first question, the study identifies significant characteristics relating to the portfolio evaluation criteria Reward and cost, Risks, and Synergies. In addition, AI projects exhibit characteristics such as heavy dependencies on data, experimentation-driven development, and high levels of unpredictability. Building on these findings, the second research question establishes that PPM practices for AI projects need to be customized accordingly. The findings point to the necessity of structured and ordered PPM practices, where projects are continuously evaluated as information is gathered. Therefore, techniques including scorecards and integrated exit criteria are proposed to reduce the projects and achieve a strategically aligned AI portfolio. In conclusion, to handle the unpredictable nature of AI projects, project selection tools from previous literature need to be customized. Furthermore, particular emphasis needs to be placed on the structure and order of the PPM to support information acquisition in a resource-efficient manner, even if AI projects often take an experimentation-driven approach to development. Ultimately, to successfully implement PPM practices for AI portfolios, adopting a perspective of project reduction rather than relying on previous notions of project selection seems essential.