AI adoption in Malaysian SMEs correlates with better employee performance, largely because it strengthens workers' intrinsic motivation rather than solely improving operational efficiency.

Main Finding

AI adoption in Malaysian SMEs positively affects employee performance, and this effect is largely transmitted indirectly through increases in employees’ work motivation. Mediation analysis shows work motivation partially mediates the AI → performance link, with the indirect effect (β = 0.322, 95% CI [0.245, 0.405]) exceeding the direct effect (β = 0.215, p < 0.001).

Key Points

• Theoretical framing: integrates Self-Determination Theory (SDT) and Task–Technology Fit (TTF). AI can satisfy psychological needs (competence, autonomy) and better fit tasks, producing motivational and performance gains.
• Hypotheses tested:
  • H1: AI Adoption → Employee Performance (supported; β = 0.215, t = 3.842, p < 0.001; small effect f2 = 0.065).
  • H2: AI Adoption → Work Motivation (supported; β = 0.595, t = 12.651, p < 0.001; large effect f2 = 0.548).
  • H3: Work Motivation → Employee Performance (supported; β = 0.542, t = 9.874, p < 0.001; large effect f2 = 0.402).
  • H4: Work Motivation mediates AI Adoption → Employee Performance (supported; indirect β = 0.322).
• Explained variance: R2 = 0.354 for Work Motivation, R2 = 0.528 for Employee Performance (moderate–strong).
• Measurement quality: scales adapted from established instruments (Venkatesh et al., WEIMS, IWPQ). Reliability and convergent validity reported (Cronbach’s α and CR > 0.70; AVE > 0.50). Discriminant validity via HTMT < 0.85.
• Sample/context: employees of Malaysian SMEs in services and manufacturing; N = 326 respondents (majority report daily or weekly AI use). Study situated in Malaysia’s MyDIGITAL policy push for SME digitalization.

Data & Methods

• Design: quantitative, cross-sectional survey; positivist/explanatory approach.
• Sampling: purposive non-probability sample targeting employees who interact with AI tools (e.g., AI analytics, automated CRM, RPA).
• Sample size and profile: 326 valid responses; sectors — services 56.4%, manufacturing 43.6; usage frequency: daily (41.7%), weekly (27.6%).
• Instruments:
  • AI Adoption: 6 items (adapted; emphasis on Task–Technology Fit).
  • Work Motivation: 12 items (WEIMS, SDT-based).
  • Employee Performance: 8 items (IWPQ; task & adaptive performance).
  • 5-point Likert responses.
• Analysis:
  • PLS-SEM via SmartPLS 4; two-stage evaluation (measurement model, structural model).
  • Bootstrapping with 5,000 resamples for path significance and mediation (Preacher & Hayes approach).
  • Reported metrics: path coefficients, t-values, p-values, f2 effect sizes, R2, Q2 predictive relevance.
• Limitations noted by the study (and implicit):
  • Cross-sectional and self-reported data → limits causal inference and may incur common-method bias.
  • Purposive sampling → limited generalizability and potential selection bias.
  • No objective performance metrics (reliance on subjective measures).

Implications for AI Economics

• Mechanism matters: economic returns to AI adoption depend substantially on human/psychological channels (motivation), not just on technical deployment. Models of AI-driven productivity should incorporate mediators like motivation, skill acquisition, and job redesign.
• Investment priorities: for SMEs and policymakers, pure technology procurement is insufficient. Complementary investments in training, participatory implementation, and job enrichment (to enhance competence and autonomy) will likely yield higher productivity per dollar spent.
• Cost–benefit analyses: include costs for change management, upskilling, and measures to maintain employee autonomy/relatedness. Failing to account for these may overestimate net gains from AI.
• Heterogeneity & policy targeting: results from Malaysian SMEs (high proportion of small firms, cultural factors) suggest targeted support (e.g., subsidized training, advisory services) in developing economies will be important to realize AI’s aggregate productivity benefits.
• Measurement for macro/meso studies: when aggregating firm-level AI adoption effects, incorporate behavioral indicators (motivation, perceived fit) alongside adoption intensity to better explain variation in performance outcomes.
• Future research directions relevant to AI economics:
  • Longitudinal or experimental designs to identify causal pathways and persistence of effects.
  • Inclusion of objective productivity metrics (output, sales, error rates) and cost-side outcomes (labor substitution, wage effects).
  • Exploration of heterogeneity by firm size, sector, and worker skill level; and modeling potential displacement vs. augmentation trade-offs.
  • Quantifying the returns to complementary investments (training, redesign) to inform subsidy and policy design.

Concise takeaway: AI raises employee performance in Malaysian SMEs, but most of that gain flows through increased work motivation. Economic assessments and policy for AI should therefore budget for human-centered complements (training, autonomy-supportive implementation) to capture full productivity benefits.