Abstract
Despite growing investment in artificial intelligence (AI), many firms struggle to translate this groundbreaking new technology into consistent improvements in operational efficiency (OE). Using an unbalanced panel of 1,340 Chinese listed firms from 2019–2022, we estimate firm-level OE via stochastic frontier estimation and measure AI investment as firms' monetary commitments to AI-related fixed and intangible assets. Baseline results show a U-shaped association between AI investment and OE. Drawing on resource orchestration theory (ROT), we further show that absorbed slack—an embedded buffer that supports structuring and bundling— systematically reshapes this nonlinear payoff: at low absorbed slack, the curve resembles a localized inverted U; at moderate levels, the curvature attenuates; and at high absorbed slack, the relationship shifts to a steeper U-shape with a right-shifted turning point. In contrast, unabsorbed slack exhibits weak and inconsistent moderating effects. These results are robust to alternative specifications (including industry and year fixed effects), alternative AI measures, lagged-variable robustness tests, Lewbel type IV estimation, and alternative absorbed-slack proxies with liquidity controls. These findings clarify the role of organizational slack as a boundary condition in AI driven OE and offer managers actionable guidance for building absorbed slack and calibrating the timing and scale of AI investment to turn early drag into lasting efficiency improvements.