Managerial AI reliance in multi-decision contexts: Navigating the paradox of cognitive dissonance and anticipated workload in AI-augmented strategic decision-making

ABSTRACT Managers are increasingly using artificial intelligence (AI) to enhance organizational decision-making, especially in complex multi-decision contexts. Yet, current AI-augmentation theories paradoxically consistently view AI reliance in multi-decision contexts as shaped by single decisions in isolation, ignoring the interconnected nature of multi-decision tasks. This oversight is critical because overall cognitive experiences with the task and the AI tool collectively shape reliance, a dynamic unexplained by existing theories. To address this gap, I connect two crucial but previously unlinked frameworks explaining decision-making in multi-decision contexts—cognitive dissonance and workload anticipation—and draw on paradox theory to explain how managers navigate the tension between discounting AI when they disagree with it and relying on it under high anticipated workload. I propose that higher disagreement with AI throughout the tasks reduces AI reliance in single decisions beyond the (dis)agreement with AI for a single decision, and that this effect is less pronounced when managers anticipate high task-level workload. I empirically tested this model through a field experiment with 88 expert managers who evaluated 540 innovation decisions with AI assistance, complemented by 11 interviews with the experiment participants. This study advances AI-augmented decision-making research by introducing a path-dependent model linking cognitive dissonance, anticipated workload, and paradox theory, and by uncovering novel individual-level cognitive factors shaping reliance across multi-decision tasks, offering both theoretical and practical guidance for sustaining effective AI integration in decision-making.