Heterogeneous top teams hurt short-term firm value-creation efficiency but prompt stronger AI strategy disclosure that yields a tiny indirect value gain; robust regional digital infrastructure magnifies the positive link from AI strategy disclosure to value creation.

Main Finding

TMT (top management team) heterogeneity is associated with firm value creation efficiency (VCE) via two opposing pathways: a negative direct association with VCE and a positive but economically small indirect association operating through greater disclosed AI strategic orientation (AIO). Regional digital infrastructure strengthens the link from disclosed AIO to VCE. These results are conditional associations (not causal), and disclosed AIO reflects textual disclosure rather than verified AI implementation.

Key Points

• Sample: Chinese A‑share listed firms, unbalanced panel 2015–2024.
  • Main mediation models: 34,823 firm‑year observations.
  • Regional moderation model: 34,649 firm‑year observations.
• TMT heterogeneity:
  • Negatively associated with firm value creation efficiency (VCE).
  • Positively associated with disclosed AI strategic orientation (AIO).
• Mediation/indirect effect:
  • The positive indirect effect of TMT heterogeneity on VCE via disclosed AIO is statistically significant but economically modest.
  • Firm‑cluster bootstrap estimate: indirect effect = 0.000084; 95% percentile CI = [0.000019, 0.000151].
• Moderation by regional digital infrastructure:
  • Stronger digital infrastructure amplifies the positive AIO → VCE association.
• Modeling approach:
  • Industry and year fixed effects; firm‑clustered standard errors.
  • Results represent conditional associations; disclosed AIO is a disclosure measure, not a validated implementation measure.

Data & Methods

• Data:
  • Unbalanced panel of publicly listed Chinese firms (A‑share), 2015–2024.
  • Large sample sizes noted above for mediation and moderation analyses.
• Measurement of disclosed AI strategic orientation (AIO):
  • Computed as ln(1 + ai_index).
  • ai_index = cosine similarity between sentence‑transformer embeddings of a 73‑term AI dictionary and AI‑related sentences in the Management Discussion & Analysis (MD&A) sections of annual reports.
• Empirical specification:
  • Fixed‑effects models controlling for industry and year.
  • Standard errors clustered at the firm level.
  • Mediation (dual‑pathway) analysis estimating direct and indirect effects; bootstrap for inference on indirect effect (firm cluster bootstrap).
• Limitations noted by authors:
  • Observational associations, not causal estimates.
  • AIO is disclosure‑based; it should not be interpreted as confirmed AI adoption or successful implementation.

Implications for AI Economics

• Theory on TMT diversity:
  • Supports a dual‑pathway view: heterogeneity brings both cognitive/resource benefits (encouraging AI strategy disclosure) and coordination/implementation costs (negative direct association with VCE). Net effects depend on magnitudes of these pathways.
• Measurement and interpretation of AI activity:
  • Textual disclosure measures (embedding + dictionary approach) can detect strategic orientation but are limited as proxies for implementation or operational AI capability. Empirical work should complement disclosure measures with operational metrics (patents, product features, AI hiring, expenditures, deployment outcomes).
• Role of regional infrastructure:
  • Digital infrastructure is a key contextual moderator — investments in regional digital capacity can increase the returns (in VCE) to firms’ AI strategic orientation. This has implications for regional policy and for firm location/expansion strategies.
• Managerial implications:
  • Heterogeneous TMTs may be more likely to signal AI strategies, but firms should watch for coordination/implementation frictions that can depress immediate value creation. Building organizational capabilities and local infrastructure can help convert AI strategy signals into measurable value gains.
• Directions for future research:
  • Causal identification of the pathways (e.g., quasi‑experimental designs, instrumental variables).
  • Broader/alternative measures of AI implementation and performance effects.
  • Disaggregation of TMT heterogeneity dimensions (functional background, education, tenure, gender) to see which drive disclosure versus implementation outcomes.
  • Cross‑country comparisons to assess generalizability beyond Chinese listed firms.
  • Dynamic/longer‑horizon outcomes of AI strategy disclosure and actual adoption.