Managing AI Projects Under Deep Uncertainty: An Adaptive Governance Framework for Project Managers

Abstract

Artificial intelligence (AI) projects are frequently managed as if uncertainty were concentrated in execution: schedule, cost, resources, and delivery risk. Yet AI initiatives can remain uncertain at more fundamental levels, including whether the problem is correctly framed, whether the technical approach can perform sufficiently, whether data are adequate, whether a viable solution will create sustained organizational value, whether its use is governable and acceptable, and whether today's assumptions will remain valid as models, vendors, costs, regulation, and stakeholder expectations change. This integrative review develops a project-manager-centered framework for governing AI projects under deep uncertainty. Drawing on project uncertainty research, experimentation and innovation literature, agile and Stage-Gate approaches, machine-learning engineering evidence, and AI governance frameworks, the paper proposes six interacting uncertainty dimensions: problem, technical/capability, data, value/adoption, governance/context, and temporal uncertainty. It argues that conventional risk and control mechanisms remain necessary but become insufficient when important uncertainties cannot yet be represented as stable requirements or identifiable probabilistic events. The proposed adaptive governance framework therefore links experimentation, evidence assessment, decision gates, and progressive commitment. Its central principle is that commitment should increase only as uncertainty is converted into evidence. The framework gives project managers explicit options to continue, modify, pivot, pause, or terminate initiatives as evidence evolves and positions monitoring and reassessment as ongoing governance rather than post-deployment activities. The paper contributes a conceptual bridge between project uncertainty theory and emerging AI-project management research, while offering a practical structure for balancing learning, accountability, and delivery discipline.