Generative AI and Organizational Decision-Making: A Systematic Review of Performance Effects
El Mehdi Nassiri, Amal El Mzabi · Procedia Computer Science · 2026
Recent empirical studies on generative AI in organizations present a puzzling picture: the same technology appears to improve performance for some workers while degrading it for others. This systematic review synthesizes 15 studies (N=15,752) to develop an explanation for this heterogeneity. We argue that asymmetric reliability signals—AI’s tendency to produce uniformly confident outputs regardless of underlying accuracy—constitutes a key mechanism disrupting human calibration of when to delegate tasks to AI. Our analysis suggests that effects vary systematically by decision phase: positive in intelligence tasks (d=+0.41), mixed in design tasks, and negative in choice tasks (d=-0.28). We formalize seven testable propositions specifying boundary conditions for when AI helps versus harms organizational decisions.