Managers trust AI-assisted performance reviews when interactions feel high-quality, allow human oversight and seem human; transparency drives trust most strongly. Firms aiming to increase uptake should prioritise explainability, clear human review and traceable accountability.

Main Finding

Managers’ judgements of fairness, accountability and transparency (FAT) in AI-supported performance evaluations are shaped primarily by interaction quality, human agency and perceived humanness; these FAT dimensions in turn are positively associated with trust—especially transparency. FAT dimensions contribute both distinctly and jointly (as an overall FAT appraisal) to managers’ trust in AI-supported evaluations.

Key Points

• Sample: 289 Australian managers with AI experience, evaluated using a standardized performance-evaluation scenario.
• Predictors positively associated with all three FAT dimensions: interaction quality, human agency, perceived humanness.
• Voice (opportunity to be heard/appeal) was associated only with transparency, not equally with fairness or accountability.
• All three FAT dimensions (fairness, accountability, transparency) were positively associated with trust; transparency had the largest effect on trust.
• Most specific indirect effects (predictor → FAT dimension → trust) were significant, indicating mediation by FAT.
• Moderation tests: task complexity and perceived uncanniness did not moderate the FAT→trust relationships.
• Human agency had slightly greater unique explanatory power than voice.
• Exploratory model: an overall FAT appraisal (combining the three dimensions) was strongly associated with trust.
• Limitations: cross-sectional design (no causal inference) and a geographically/experience-constrained sample (Australian, AI-experienced managers), limiting generalisability.
• Practical recommendations from the authors: combine understandable interaction with contextual sensitivity, human review and documented override authority, clear traceable responsibility, and credible appeal mechanisms.

Data & Methods

• Design: Cross-sectional survey using a standardized performance-evaluation scenario.
• Sample: 289 managers in Australia with AI experience.
• Analysis: Partial least squares structural equation modeling (PLS-SEM) to test:
  • the hypothesized structural model,
  • indirect (mediation) effects,
  • moderation effects (task complexity, perceived uncanniness),
  • alternative model specifications,
  • out-of-sample prediction performance.
• Key constructs: interaction quality, human agency, perceived humanness, voice; FAT dimensions (fairness, accountability, transparency); outcome: trust.
• Findings are associative (correlational); PLS-SEM used for variance explanation and prediction rather than causal identification.

Implications for AI Economics

• Modeling AI adoption and productivity:
  • Trust in AI-mediated HR processes depends on perceived transparency, fairness and accountability. Economic models of AI adoption should include FAT as determinants of organizational uptake and effective deployment.
  • Transparency carries particularly large influence on trust; investments in explainability/interpretability may yield outsized returns in adoption and cooperation.
• Design and deployment incentives:
  • Human agency (ability for human oversight and override) materially affects FAT and thus trust—organisations may need to internalize the benefits of human-in-the-loop design despite automation efficiencies.
  • Voice mechanisms matter mainly for perceived transparency; appeal and feedback processes influence acceptability and legitimacy, with implications for retention, morale and dispute costs.
• Labor market and incentive effects:
  • Management trust in AI-supported evaluations can influence performance-pay contracts, monitoring intensity, and promotion decisions—affecting wages, effort and allocation of tasks.
  • Lack of moderation by task complexity suggests FAT→trust relationships are robust across simple and complex evaluation tasks; policy and firm-level interventions can generalize across many contexts.
• Policy and regulation:
  • Regulations mandating human review, override rights, traceability of responsibility, and accessible appeal processes align with features that build FAT and trust—potentially improving compliance and reducing litigation/turnover costs.
  • Policymakers evaluating the social welfare effects of AI in workplaces should account for trust-mediated channels (e.g., acceptance, compliance, productivity).
• Empirical research directions:
  • Incorporate FAT measures into empirical studies linking AI adoption to firm performance, labor outcomes, or productivity to capture omitted-variable bias.
  • Use longitudinal or experimental designs to identify causal effects (given current cross-sectional limitation).
  • Explore heterogeneity beyond the Australian, AI-experienced managerial population to assess external validity across sectors, countries and worker types.

Summary: For economists studying AI’s organizational and labor-market impacts, FAT—especially transparency and human agency—is a key mediator of trust that influences adoption, effectiveness, and the broader economic consequences of AI-supported evaluation systems.