AI as a cognitive multiplier: explaining performance variance in organizational decision-making

Purpose This paper aims to propose an explanation for why identical AI systems produce divergent decision-making outcomes across users, and what happens to the distribution of decision quality over time across repeated AI-supported decision-making episodes. Design/methodology/approach This paper is conceptual. The proposed framework was built using abductive reasoning and a synthesis of existing research in the fields of human–AI interfaces, decision-making, cognitive psychology and management research. Findings The cognitive multiplier framework posits that AI systems increase the sensitivity of decision quality to users' baseline cognitive engagement, defined as the level of critical thinking and metacognitive monitoring deployed during a discrete decision-making episode. Across repeated decision-making episodes, cognitive engagement patterns stabilize, moving users along a continuum ranging from augmentation to atrophy. Perceived ownership, feedback responsiveness and need for cognition are proposed as moderating factors. To enable empirical testing of the framework, six testable propositions are derived. Originality/value The proposed framework acts as a mechanism-level explanation for the observed variance in AI-supported decision-making. It offers a new perspective that bridges the existing debate over whether implementing AI systems causes better or worse aggregate results and reconciles contradictory empirical findings by reframing both positive and negative outcomes as poles on a continuum from cognitive augmentation to cognitive atrophy. The implications of the framework apply to organizational workflow design, governance, capability development and business education.