AI improves SME HR efficiency—particularly in recruitment and performance analytics—but adoption in China is driven by state-backed platforms that reduce costs while creating dependency and governance challenges.

Main Finding

AI adoption in SME human resource management (HRM) largely delivers operational efficiency—especially in recruitment, HR workflow automation, and performance analytics—but benefits are conditional on firms’ digital capabilities, cost constraints, and governance requirements. In China, adoption follows a policy- and platform-mediated pathway that lowers entry barriers but increases dependence on external ecosystems and regulatory compliance costs.

Key Points

• Scope and evidence base
  • Systematic review of 21 studies (published 2019–Mar 2026) drawn from 1,562 initial records; majority of studies published after 2023 (post-LLM/ChatGPT surge).
  • Research is dominated by cross-industry reviews and conceptual work; empirical evidence for SMEs is limited and sector-specific empirical studies are scarce.
• Primary AI applications in SME HRM
  • Recruitment and selection: resume parsing, candidate ranking, automated scheduling, AI video-interview screening, speech/NLP tools.
  • Training & development: personalized learning platforms, VR simulations, LLM-supported content generation.
  • Performance management: real-time analytics, sentiment analysis, productivity dashboards.
  • Less evidence on Compensation & Benefits and on fully autonomous HR decision-making.
• Reported benefits
  • Efficiency gains (time and error reductions), reduced administrative burden, faster screening, and potential productivity and innovation spillovers when AI tools are integrated with broader digital transformation.
• Persistent challenges and risks
  • Algorithmic bias, data privacy and protection concerns, opaque decision logic reducing trust, implementation and maintenance costs, limited AI/digital literacy, and employee resistance/fear of job displacement.
• China-specific findings
  • Distinctive adoption pattern: modular AI tools embedded in platform ecosystems driven by state-backed digitalization and provincial pilot zones.
  • Legal landscape materially affects adoption: Cybersecurity Law (CSL), Data Security Law (DSL), Personal Information Protection Law (PIPL), plus algorithm- and generative-AI-specific measures.
  • Incentives: targeted AI/subsidy programs and National Pilot Zones encourage adoption but create geographic heterogeneity and platform dependence.
• Research gaps
  • Very few Chinese empirical studies (three noted: iFlytek case, Jiangsu SME survey n=339, healthcare sector study).
  • Over-reliance on SEM and cross-sectional methods; shortage of causal, sector-specific, longitudinal, and qualitative work.

Data & Methods

• Review protocol
  • Followed Tranfield et al. (2003) framework and PRISMA 2020 reporting guidelines.
  • Databases searched: Google Scholar, Web of Science, Scopus, ScienceDirect, CNKI.
  • Search terms combined AI-related and HRM/SME terms (English + Chinese); publication window Jan 2019–Mar 2026; included peer‑reviewed articles, conference proceedings, high-quality reports.
• Inclusion/exclusion & quality
  • Included studies explicitly addressing AI in SME HRM, empirical/conceptual/review papers with clear methods, English/Chinese, full text available.
  • Excluded studies focused only on large firms, non-HRM AI applications, pre-2019, and non-peer sources.
  • Quality filter: modified MMAT, retained studies scoring ≥4/5.
• Screening and synthesis
  • PRISMA flow: 1,562 initial records → 362 duplicates removed → 1,200 screened → 350 full texts assessed → 21 included.
  • Thematic analysis using Braun & Clarke six-phase approach; intra-coder re-coding of 25% sample yielded 91% agreement.
• Methodological composition of included studies
  • Systematic review/bibliometric: 11 (52%)
  • Empirical (SEM/survey): 5 (24%)
  • Case study/qualitative: 1 (5%)
  • Conceptual/theoretical: 4 (19%)

Implications for AI Economics

• Productivity and firm capabilities
  • AI-HRM tools can raise SME productivity and innovative capacity when bundled with digital transformation—suggesting complementarities between AI adoption and human capital/digital capabilities. Economic models should account for these complementarities and non-linear returns to scale in small firms.
• Adoption drivers and policy levers
  • Platform-based service delivery and targeted subsidies materially lower entry costs for SMEs; hence public policy and ecosystem firms play pivotal roles in diffusion. Economists should model adoption as endogenous to policy incentives, platform market structure, and spatial heterogeneity (provincial pilot zones).
• Market structure and platform dependence
  • Platform-mediated adoption creates dependency and potential lock-in externalities. There are implications for competition (vendor concentration), bargaining power, and long-run costs for SMEs that economists should quantify (e.g., switching costs, rent extraction).
• Labor market and distributional effects
  • Automation of routine HR tasks shifts HR worker tasks toward strategic roles; however, risks of job displacement, skills mismatches, and heterogenous adoption across firms can increase inequality between digitally capable SMEs and laggards. Empirical work should estimate employment composition effects, wage impacts, and re-skilling returns.
• Regulatory compliance and adoption costs
  • Data-protection and algorithmic governance (PIPL, DSL, algorithm measures) impose compliance costs that are proportionally larger for SMEs. These regulatory frictions should be modeled as adoption taxes or fixed costs altering optimal investment thresholds and diffusion speeds.
• Research & measurement priorities for AI economics
  • Need for causal and sectoral micro-evidence: difference-in-differences, RCTs, or instrumental-variable strategies exploiting policy rollouts/subsidies or pilot-zone eligibility to estimate productivity, hiring quality, employment, and wage effects.
  • Cost–benefit and welfare analyses that incorporate compliance burdens, algorithmic harms (bias), and platform dependence.
  • Structural models of adoption that include firm heterogeneity (size, digital capability), ecosystem access (vendors/platforms), and multi-level policy instruments.
• Practical policy recommendations (economic framing)
  • Subsidies and technical assistance targeted to reduce fixed adoption costs and build complementary human capital can accelerate productive diffusion.
  • Encourage interoperable standards and competitive platform markets to limit lock-in and reduce long-term costs.
  • Support measurement and reporting standards for algorithmic impact assessments so SMEs can meet compliance with limited resources.

Suggested next steps for researchers: develop province- and sector-level causal studies in China, estimate the long-run impact of AI-HRM on SME productivity and employment, and model the interaction of platform markets and regulatory regimes in shaping adoption and welfare outcomes.