Employees who use AI report higher job and innovation performance, largely because AI use raises workers' AI self-efficacy and prompts firms to adopt digital HR practices; the result is based on a 750-person cross-sectional survey from four Chinese cities, so causality is not established.

Main Finding

AI adoption improves employees' job performance and innovation performance—primarily by increasing AI self-efficacy (an individual cognitive mechanism) and by being embedded in digital HRM practices (an organizational/HR-system mechanism). Both mediators significantly and positively transmit the benefits of AI use into better performance outcomes.

Key Points

• Sample: survey of 750 valid responses from firms in Beijing, Shenzhen, Xi’an, and Zhengzhou, China.
• Outcomes studied: employee job performance and innovation performance.
• Mediators identified:
  • AI self-efficacy (individual-level cognitive belief in using AI effectively).
  • Digital HRM practices (organization-level digitalization of HR systems that support learning and knowledge use).
• Findings:
  • AI usage → higher AI self-efficacy → improved job and innovation performance.
  • AI usage → stronger digital HRM practices → improved job and innovation performance.
  • The two mediators act as complementary channels; both are significant positive mediators.
• Theoretical contribution: integrates individual (cognitive) and organizational (HR system) perspectives within knowledge management to show how AI-driven learning and capability development are enabled by both human beliefs and HR systems.
• Practical recommendation: firms should embed AI into HR systems and invest in practices that foster learning, knowledge utilization, and continuous innovation (e.g., training to raise AI self-efficacy, digital HR processes that support experimentation and knowledge sharing).

Data & Methods

• Design: Cross-sectional firm-level survey (self-reported measures).
• Sample: 750 respondents from major Chinese cities (Beijing, Shenzhen, Xi’an, Zhengzhou).
• Analysis: Mediation analysis testing whether AI self-efficacy and digital HRM practices transmit the effect of AI adoption onto job and innovation performance; reported effects are statistically significant and positive.
• Limitations of methods:
  • Cross-sectional survey limits causal inference (reverse causality/endogeneity possible).
  • Self-reported performance and mediator measures risk common-method bias.
  • Sample concentrated in Chinese cities — generalizability to other countries/sectors may be limited.

Implications for AI Economics

• Complementarity and returns to AI:
  • The study highlights that returns to AI adoption depend on complementary investments: human capital (AI self-efficacy) and organizational capital (digital HRM). Models of AI-driven productivity should include these complementarities rather than treating AI as a standalone factor.
  • AI self-efficacy functions as a multiplier of AI effectiveness—investments in worker skills and confidence raise realized productivity from AI.
• Firm heterogeneity:
  • Variation in digital HRM and workforce AI self-efficacy helps explain heterogeneous productivity and innovation gains from AI across firms. This implies persistent firm-level performance dispersion even with similar AI technologies.
• Labor demand and wages:
  • When AI is combined with HR practices that augment worker capabilities, AI is more likely to be complementary to labor, supporting upskilling and potentially higher wages for those who acquire AI-related competencies rather than broad automation-driven displacement.
  • Conversely, firms that adopt AI without investing in these complements may see weaker productivity gains and different labor outcomes.
• Policy and investment implications:
  • Public and private policies that subsidize reskilling, digital HR transformation, and diffusion of AI-supportive HR practices can increase the social returns to AI adoption.
  • Measurement of AI’s economic impact should account for organizational and human-capital mediators; simple adoption counts will understate true returns unless complementarity is measured.
• Directions for economic research:
  • Incorporate organizational capital and worker AI self-efficacy into production-function and growth models to better estimate marginal returns to AI.
  • Use longitudinal and quasi-experimental designs to identify causal effects of AI adoption conditional on HR and human-capital complements.
  • Explore heterogeneity by industry, task content, and worker skill level to map where AI + complements yields the largest welfare and productivity gains.

Suggestions for future work: longitudinal/experimental designs to address causality; objective performance/productivity measures; cross-country comparisons; cost–benefit analyses of HR investments that raise AI self-efficacy.