Governing AI-Assisted Strategy Tools: A Substitutability–Commitment Framework and the Role of Epistemic Custody

Abstract

Artificial intelligence now performs much of the analytical work that established strategy tools were designed to structure. Existing research asks which tools can be automated. This article asks a prior question: whether the authority and organisational value of a strategic instrument survive when its analytical core is machine-generated. Approach. The article is conceptual and follows a theory-construction design, synthesising the strategy-as-practice, AI-enabled decision-making and corporate governance literatures through a documented search and inclusion procedure. Framework. The Substitutability–Commitment Framework classifies strategic instruments along two independent dimensions: analytical substitutability, the extent to which a tool’s cognitive work is machine-executable, and commitment durability, the extent to which its output binds organisational resources. Crossing the dimensions yields four classes, each implying a distinct governance response. The framework is complemented by epistemic custody, defined as the explicit allocation of responsibility for producing, verifying, contesting and authorising strategic judgements. Contribution. The article separates substitutability from commitment, identifies three observable failure modes, maps thirteen widely used instruments to the framework, and advances seven refutable propositions. Epistemic custody is distinguished from accountability, human oversight and decision rights. Implications. Boards require custody arrangements rather than technical assurance alone. The withdrawal of South Africa’s Draft National Artificial Intelligence Policy in April 2026, following the discovery of fabricated references, is used solely to illustrate the framework’s plausibility. It is not evidence validating the framework, which remains untested.