Internal HR quality—not external pressures—drives perceived HR effectiveness in Kazakhstan's civil service, and managers are about 61% more likely to be active AI users while each higher tenure category reduces AI use (odds ratio 0.846); access to modern digital tools further boosts uptake.

Main Finding

In a nationally representative survey of 12,562 Kazakhstani civil servants (June 2025), AI use in HRM is still rare—only 28.3% report weekly use and 7.1% report daily use—despite generally positive perceptions of HR quality. Internal (individual-level) HR quality matters more for perceived HR effectiveness than external (organisational) factors, and active AI adoption is most strongly associated with managerial role, shorter tenure, male gender, and access to modern digital tools. Path analysis confirms stronger standardized effects of internal vs external HR quality on effectiveness (≈β=0.463 vs β=0.227).

Key Points

• Sample and prevalence
  • N = 12,562 public servants across all regions; survey via official civil service system (9–11 June 2025).
  • AI adoption: 44% “no use, no plans”, 44% “interested but not yet using”, 28.3% weekly or more; only 7.1% daily.
• Composite measures (Likert 1–5)
  • Internal HR quality index (6 items; e.g., competencies, education, CPD): mean 4.30; Cronbach’s α = 0.924.
  • External HR quality index (8 items; e.g., leadership, working conditions, digital tools): mean 4.16; Cronbach’s α = 0.959.
• OLS (dependent = HR effectiveness composite)
  • R² = 0.446. Internal index coefficient = 0.560 (p<0.001); external index = 0.239 (p<0.001).
  • Education negatively associated with perceived HR effectiveness (β = −0.130, p<0.001); tenure small positive (β = 0.021).
• Logistic regression (dependent = weekly+ AI use; McFadden R² = 0.032; accuracy 71.9%)
  • Managerial position: OR = 1.609 (p<0.001).
  • Tenure: OR = 0.846 per tenure category (≈15% lower odds per step; p<0.001).
  • Age: OR = 0.917 per age category (p<0.001).
  • Female: OR = 0.802 (≈20% lower odds than men; p<0.001).
  • Access to modern digital tools: OR = 1.104 (p<0.001).
  • Higher perceived HR effectiveness weakly associated with lower AI adoption: OR = 0.922 (p<0.001) — interpreted as a saturation/less-urgency effect.
• Path analysis (standardised)
  • Confirms internal HR quality exerts a larger effect on perceived HR effectiveness than external factors (standardised βs ≈ 0.463 vs 0.227); effectiveness and digital access then predict AI readiness/use.

Data & Methods

• Data: 12,562 responses from civil servants across 17 regions + 3 metro areas; women = 61.6% of sample; most in local executive bodies; 79% non-managerial.
• Survey blocks: socio-demographics; perceived HR quality (6 internal + 8 external items on 5-pt Likert); AI tool use (0–3 ordinal + binary weekly-use indicator); attitudes to E-Kyzmet platform.
• Variable coding:
  • Composite internal/external HR indices = item means; HR effectiveness = mean of two items (system and own org).
  • Demographics encoded as ordinal categories; dummies for female, managerial, central body.
  • Binary AI adoption = weekly+ use vs otherwise (28.3% = 1).
• Econometric approach:
  • Model 1: OLS with HC1 robust SEs predicting HR effectiveness (controls: education, age, tenure, manager, female, central).
  • Model 2: Binary logistic predicting weekly+ AI use (adds HR effectiveness composite and single-item digital-tools access).
  • Model 3: Path analysis in standardized metrics linking HR quality → perceived effectiveness → AI readiness (uses full ordinal 0–3 AI scale).
• Reliability: Cronbach’s α: internal 0.924; external 0.959 (excellent).

Implications for AI Economics

• Diffusion drivers and heterogeneity
  • Adoption is concentrated among managers and recently hired staff, implying early diffusion is tied to positional authority and weaker path-dependent routines. Economics of technology diffusion in the public sector must account for organisational hierarchies and tenure-driven inertia.
  • The gender gap (women ~20% less likely to be weekly users) may amplify inequality in digital skill accumulation and future career returns in the civil service; policy interventions could alter human-capital accumulation trajectories.
• Role of facilitating conditions vs perceived need
  • Access to modern digital tools raises adoption odds (small positive effect), but perceived HR systems that are “working well” reduce urgency to adopt AI. Cost–benefit calculations for public AI investments should therefore consider both infrastructure and demand-side incentives: making utility salient and targeted use-cases visible may be as important as providing hardware/software.
• Human capital vs organisational investment
  • Internal HR quality (skills, CPD, competencies) has larger effects on perceived effectiveness and readiness than organisational factors, suggesting returns to investing in employee skills (training, upskilling) can have multiplier effects on AI uptake and effective use. For economists modeling returns to AI in the public sector, incorporate complementarity between employee human capital and AI-tool productivity.
• Labor-market and policy consequences
  • Concentrated adoption among newer cohorts suggests potential changes in task allocation and career progression: adopters may gain productivity advantages, possibly altering internal labor markets, promotion paths, and external mobility. Economists should model potential wage/rent shifts, retraining needs, and retention incentives.
• Measurement and modelling cautions
  • Low McFadden R² for the logistic model (0.032) indicates many unobserved determinants of individual adoption (e.g., managerial encouragement, informal norms, procurement constraints, legal/regulatory clarity). Evaluations of AI policy interventions should include experimental or quasi-experimental designs to identify causal impacts.
• Policy implications relevant to AI economics
  • Targeted interventions (training, pilots) focused on mid-career and long-tenure staff could overcome routine inertia more cost-effectively than blanket procurements.
  • Address gender-specific barriers to digital uptake to prevent widening digital-skill divides in public employment.
  • Design procurement and incentive structures that make AI’s productivity benefits visible in HR workflows—this increases perceived usefulness, a key determinant of adoption.
  • Combine investments in digital infrastructure with human-centered design and governance (transparency, fairness audits) to increase acceptability and sustained use.

Limitations to keep in mind: cross-sectional self-reported data prevent causal claims; single-country case (Kazakhstan) may limit external generalisability; some relevant institutional factors likely unmeasured.