Text-mined evidence from Chinese A-share firms (2011–2023) shows AI adoption materially lowers warehousing and logistics costs; gains operate through increased task specialization, supplier diversification and better inventory forecasting and are strongest in eastern regions and tech-intensive industries.

Main Finding

AI adoption by Chinese A‑share listed firms significantly reduces warehousing and logistics costs. This effect is robust to alternative measures, exclusion of pandemic-year data, and instrumental-variable estimation. Mediation analysis indicates the cost reduction operates via (1) enhanced specialized division of labor, (2) greater supply‑chain diversification, and (3) improved inventory management (real‑time demand forecasting and predictive analytics). The effect is stronger for firms in eastern China and for technology‑intensive industries.

Key Points

• Contribution: Provides firm‑level micro evidence from China linking AI application intensity to lower logistics/warehousing costs, complementing prior macro or case studies.
• Measurement innovation: Constructs an enterprise AI application intensity index by text‑mining annual report disclosures (keyword/frequency analysis).
• Causal strategy: Main estimates use two‑way fixed effects (firm and year); robustness checks include alternative variable definitions, excluding 2020, and an instrumental‑variable approach to address endogeneity.
• Mechanisms identified: AI reduces logistics costs by (a) enabling finer specialization/division of labor, (b) supporting supply‑chain diversification while retaining efficiency, and (c) optimizing inventory through forecasting and dynamic replenishment.
• Heterogeneity: Larger cost reductions for firms in eastern China and for technology‑intensive sectors.
• Practical implication noted by authors: supports policy push for AI integration in the real economy and guides firms toward intelligent supply‑chain management.

Data & Methods

• Sample: Chinese A‑share listed companies. Abstract reports the period 2011–2023 (panel data). The Methods section contains an internal inconsistency stating the study period spanned 2024–2025; the 2011–2023 range in the abstract and empirical framing appears to be the intended panel period—this discrepancy is noted.
• AI measure: Text‑mined AI application index derived from firms’ annual reports via systematic keyword frequency analysis (documented as a novel index).
• Outcome: Firm logistics/warehousing costs (as reported in financial statements; paper frames results around warehousing and logistics cost control).
• Econometrics:
  • Baseline: two‑way fixed effects regressions (firm and year).
  • Endogeneity: instrumental variable estimation (details of instrument not reproduced here).
  • Mechanism tests: mediation/mediating effect models for division of labor, supply‑chain diversification, and inventory management efficiency.
  • Heterogeneity checks: by geographic region and industry technology intensity.
  • Robustness: alternative variable constructions and exclusion of special years (e.g., 2020 COVID shock).
• Limitations in methods noted by authors (and apparent from text): potential measurement error from text‑mined AI index; sample limited to listed firms; an internal inconsistency on stated study period.

Implications for AI Economics

• For research:
  • Micro‑level evidence: Strengthens the case that firm AI deployment affects operational costs, not just productivity aggregates—useful for models linking technology adoption to firm cost structure and organization.
  • Measurement approach: The text‑mining AI intensity index is a replicable method for capturing technology adoption intensity from firm disclosures; future work should compare it to investment or expenditure measures and validate it externally.
  • Mechanisms: Empirically connects AI to transaction‑cost reduction, altered specialization/division of labor, and inventory dynamics—these channels can be incorporated into structural models of firm organization and supply‑chain design.
  • Open questions: generalizability to non‑listed and smaller firms, long‑run dynamic effects (capex vs. opex), labor reallocation and distributional consequences, and precise causal pathways (e.g., validating instruments).
• For policy and firms:
  • Policy: Results support policies that promote AI adoption and digital infrastructure (especially outside eastern China) to realize logistics cost savings and enhance supply‑chain resilience.
  • Firms: Investing in AI for forecasting, routing/optimization, and process automation can lower logistics and inventory costs; complementary organizational changes (data sharing, cross‑department integration) are important to realize benefits.
• Caveats:
  • Measurement and sample limitations mean findings apply directly to listed Chinese firms that disclose AI usage; effects may differ for SMEs or informal sectors.
  • The paper contains an internal reporting inconsistency about the study period; readers should check the published appendix/data for exact sample years and instrument details before using the index or estimates for policy simulation or meta‑analysis.