Employees’ fear, not technology, is the main brake on enterprise AI adoption: perceived threat—working through job insecurity and anxiety—significantly lowers workers’ willingness to adopt AI, and these psychological barriers vary markedly by firm size, industry and prior experience.

Main Finding

A robust "aspiration–reality" gap exists in enterprise AI adoption: management’s technology-empowerment vision often clashes with frontline employees’ low willingness to adopt. The strongest psychological barrier is perceived threat; job insecurity partially mediates the effect of perceived threat on adoption intention, forming a full transmission chain from threat cognition → affective response → behavioral intention. The intensity of these psychological barriers varies significantly by enterprise size, industry type, and employees’ prior AI experience. The paper integrates four psychological-barrier dimensions into a unified model, advancing innovation-resistance theory for enterprise AI.

Key Points

• Four psychological-barrier dimensions examined:
  • Perceived threat (strongest inhibitor)
  • Technology anxiety
  • Job insecurity (partial mediator between perceived threat and adoption intention)
  • Perceived complexity
• Mechanism identified: perceived threat leads to affective responses (e.g., job insecurity, anxiety), which reduce intention to adopt AI—i.e., a complete psychological transmission chain.
• Heterogeneity: barrier strength differs by organizational context and employee background (enterprise size, industry, individual AI experience).
• The study addresses fragmentation in prior literature by proposing a four-dimension integrated model applicable to enterprise AI settings.
• Managerial implication emphasized: move from pure technology-push strategies to combined technology + psychological interventions tailored to context.

Data & Methods

• Design: Cross-sectional survey across three enterprises at different AI-application stages.
• Sample: Valid respondent samples collected from the three enterprises (study summary does not report specific n here).
• Analysis: Structural equation modeling (SEM) to test relationships among barrier dimensions and adoption intention; multi-group comparison analyses to assess heterogeneity across enterprise/employee subgroups.
• Theoretical framing: Innovation resistance theory adapted to enterprise AI adoption through psychological barriers.

Implications for AI Economics

• Adoption friction and diffusion models: Psychological barriers (especially perceived threat and job insecurity) are measurable frictions that slow AI adoption and should be integrated into diffusion and adoption models as behavioral-cost parameters or segmented-adopter latent classes.
• Heterogeneous uptake and productivity effects: Variation across firm size, industry, and employee AI experience implies uneven productivity gains and adoption timing across the economy; macro forecasts should account for this heterogeneity to avoid overestimating near-term AI-driven growth.
• Labor-market impacts and adjustment costs: Job insecurity both reduces adoption willingness and signals potential real adjustment costs (retraining, redeployment). Economic assessments of AI investments should include these psychological/transition costs and employer investments in mitigation (training, redeployment programs).
• Policy design: Public policies aiming to accelerate beneficial AI adoption should go beyond subsidies for technology and include measures to reduce psychological barriers—e.g., incentivized retraining, unemployment-smoothing programs, certification of safe/augmentative AI use—to lower perceived threat and job insecurity.
• Managerial strategy and ROI: Firms should implement dual-track investments—technical deployment plus targeted psychological interventions (communication, upskilling, participatory deployment, simplified UX)—to convert managerial vision into actual adoption, reduce rollout failure risk, and improve realized ROI.
• Measurement and evaluation: Empirical economic evaluations of AI should incorporate survey-based measures of employee attitudes as leading indicators of adoption success and as moderators of expected productivity impacts.
• Research recommendations for economic modeling: Use longitudinal and experimental designs to estimate causal effects of psychological interventions on adoption and productivity; include dynamic feedback where adoption changes perceptions over time (learning effects that can reduce barriers).

Limitations to keep in mind: cross-sectional design restricts causal claims; data drawn from three enterprises may limit generalizability; potential self-report/common-method bias—future work should use longitudinal, multi-site, and objective adoption-measure approaches.