Manufacturers that turn data into reusable assets show higher ‘digital intelligence’: BERT-based analysis of Chinese listed firms (2014–23) finds robust positive links driven by better AI talent, tech innovation and collaboration, with bigger gains in firms that can absorb and govern AI effectively.

Main Finding

Data valorization significantly and robustly drives enterprise digital intelligence. The effect operates by restructuring the AI ecosystem along three channels — AI talent, AI technology, and AI collaboration (resource utilization → assetization → capitalization). The transformational impact is stronger in firms with higher absorptive capacity, more optimal talent structures, and richer governance experience.

Key Points

• Contribution and framing
  • Proposes a systematic theoretical framework: "data valorization → AI ecosystem reconstruction → enterprise digital intelligence."
  • Positions data valorization (the process of turning raw data into measurable/tradable/value-added assets) as the core, persistent driver of firms’ digital-intelligence transformation.
• Mechanisms (AI ecosystem reconstruction)
  • AI talent: Data valorization expands required knowledge breadth, accelerates skill upgrading, and creates specialized roles (e.g., data architects, model compliance officers), improving human-capital foundations for digitalization.
  • AI technology: By improving data quality and circulation and stimulating demand for computing and algorithmic capability, data valorization raises the performance ceiling of AI systems and accelerates innovation and engineering application.
  • AI collaboration: Valorization deepens cooperative arrangements across resource use, assetization, and capitalization stages — fostering external partnerships, data/trading platforms, and commercialization pathways that scale digital capabilities.
• Heterogeneity
  • Positive effects are heterogeneous and amplified in firms with:
    • Strong absorptive capacity (ability to learn and internalize external knowledge),
    • Well-structured AI/digital talent pools,
    • Experience in governance and managing digital/data assets.
• Methodological innovation
  • Dependent variable (enterprise digital intelligence) is measured via text-mining based on a BERT large language model, improving semantic understanding and motivational identification versus traditional lexical-count methods.
• Robustness
  • The main positive relationship holds after multiple robustness checks and mechanism tests (as reported).

Data & Methods

• Data
  • Sample: China’s Shanghai and Shenzhen A-share manufacturing listed companies.
  • Period: 2014–2023.
• Empirical strategy (as described)
  • Constructs an analytical framework grounded in technological economics and innovation theory.
  • Measures the core dependent variable (enterprise digital intelligence) using BERT-based text mining of firm disclosures/documents (improves on lexical approaches).
  • Tests the direct effect of data valorization on enterprise digital intelligence and then investigates mediating mechanisms (AI talent, AI technology, AI collaboration).
  • Conducts heterogeneity analyses across firm absorptive capacity, talent structure, and governance experience.
  • Performs a series of robustness checks (details in article).
• Note on measures
  • The paper emphasizes a multi-stage conception of data valorization (resource utilization → assetization → capitalization) but specific variable operationalizations for "data valorization" and mediators are described in the full text (not fully reproduced here). The use of BERT for the outcome is a highlighted methodological advance.

Implications for AI Economics

• Data as an economic input
  • Reinforces the view of data as a production factor whose valorization (not merely possession) determines economic returns; policy and firm strategy should focus on pipelines that convert data into priced, governable assets.
• Returns to firm-level investment
  • Investments that improve data governance, measurement, and monetization can yield outsized returns by unlocking AI-driven productivity through talent upgrading, technology improvements, and expanded collaborations.
• Market and ecosystem dynamics
  • Data valorization encourages specialization (new AI-related occupations) and the emergence of markets/platforms for data, models, and computing — altering industry structure and competition.
• Heterogeneity and policy targeting
  • Effects vary across firms; interventions (training, standards, financing for data assetization) are likely to be most effective when targeted toward firms with weaker absorptive capacities or immature talent/governance structures.
• Measurement and evaluation
  • Using semantic, LLM-based measures (e.g., BERT) for firm-level digital-intelligence outcomes offers a promising empirical tool for future AI-economics research and for monitoring progress of digital transformation policies.
• Regulatory and governance implications
  • As data moves from raw input to asset/capital, regulatory frameworks for data rights, pricing, liability, and model governance become central to market efficiency and to preventing concentration or misuse of data-derived advantages.

If you want, I can: - Extract and summarize the empirical specification, variables, and robustness-check details from the full paper (if you can provide the methods/results tables). - Convert these findings into policy recommendations for regulators or an executive checklist for firms pursuing data valorization.