Diminishing returns to calibrated reliance on AI: effects on deliverable quality and rework effort, moderated by task verifiability
Junjie Sun, Yifan Yang, Jiangnan Cheng, Weichao DingPurpose
Artificial intelligence (AI) tools can function as both “performance accelerators” and “risk propagators” in organizations. This study investigates how AI-enabled work capability, particularly employees' decomposition and evaluation skills, is associated with deliverable quality and rework effort through the behavioral pathway of calibrated reliance on AI. Additionally, it examines task verifiability as a critical boundary condition that shapes the effectiveness of human-AI collaboration.
Design/methodology/approach
Guided by appropriate reliance and automation trust theories, data from 380 workers were analyzed using SmartPLS 4 (partial least squares structural equation modeling) and hierarchical polynomial regression to test mediation, moderation, and quadratic effects. Procedural remedies and a latent method factor approach were employed to assess common method bias.
Findings
AI-enabled work capability was positively associated with calibrated reliance on AI. Calibrated reliance on AI was associated with higher deliverable quality and lower rework effort. A double diminishing marginal pattern emerged: as reliance calibration increased, quality gains and rework reductions leveled off, indicating a “too-much-of-a-good-thing” pattern. Task verifiability strengthened the positive association between calibrated reliance and deliverable quality and also strengthened its negative association with rework effort.
Originality/value
This study shifts the focus from reliance intensity to reliance calibration in explaining AI-enabled work outcomes. It documents a double diminishing marginal pattern, showing that the benefits of calibrated reliance are meaningful but not unlimited. It also identifies task verifiability as a boundary condition that helps explain when calibrated reliance is more strongly associated with improved deliverable quality and reduced rework effort. These findings provide focused guidance for capability development, reliance calibration, and task-specific AI governance.