Adopting AI-driven agile project management measurably improves Chinese firms' sustainability performance, driven by efficiency gains, faster innovation and stronger stakeholder engagement; benefits are largest for big, tech-intensive and privately owned firms.

Main Finding

AI-driven agile project management causally improves corporate sustainability performance among Chinese listed firms. Using a staggered difference‑in‑differences design on 12,290 firm‑year observations (2019–2024), the preferred specification estimates that adopting integrated AI + agile project management raises a firm’s composite ESG score by about 4.892 points (~10.7% of the sample mean). Effects operate through operational efficiency, stakeholder engagement, innovation acceleration, and improved risk management, and are larger for large firms and technology‑intensive industries.

Key Points

• Sample & scope: 12,290 firm‑year observations for Chinese listed firms from 2019–2024; ~34% of observations involve AI-driven agile project management (AIAP = 1).
• Estimated effect: DiD + controls (firm & year fixed effects, clustered SEs) → AIAP coefficient = +4.892 (statistically significant).
• Primary mechanisms (hypothesized and supported):
  • Operational efficiency: predictive analytics, automated task allocation, predictive maintenance → lower resource use and emissions.
  • Stakeholder engagement: NLP and analytics + agile feedback cycles → faster, tailored responses to stakeholder concerns.
  • Innovation acceleration: AI insights + agile prototyping → faster development/deployment of sustainable products/processes.
  • Risk management: AI early‑warning + agile response → better anticipation and mitigation of ESG risks.
• Heterogeneity: stronger positive effects for larger firms, tech‑intensive industries, and varying by ownership structure.
• Robustness: results hold under alternative CSP measures, one‑year lagged treatment, PSM‑DiD, and placebo tests. VIF diagnostics indicate low multicollinearity (mean VIF ≈ 1.82).

Data & Methods

• Data sources: CNRDS, CSMAR, corporate sustainability reports, ESG rating providers, technology disclosures, provincial GDP data.
• Outcome (dependent) variable: Corporate Sustainability Performance (CSP) — composite ESG index (0–100); plus component measures (environmental, social, governance).
• Treatment: AI-Driven Agile Project Management (AIAP) — binary indicator from firm disclosures; auxiliary continuous measures include AI adoption intensity (1–5) and agile implementation score (0–10).
• Controls: firm size (ln assets), ROA, leverage, market‑to‑book, board independence, CEO duality, HHI, regional GDP per capita, industry and firm fixed effects.
• Identification strategy:
  • Staggered difference‑in‑differences comparing treated vs. non‑treated firms over time, with firm and year fixed effects and clustered SEs.
  • Additional checks: lagged treatment to address reverse causality, propensity score matching + DiD to address selection, placebo random treatment timing.
• Sample cleaning: excluded ST/PT firms and certain industry outliers; final N = 12,290 observations.

Implications for AI Economics

• Complementarities matter: The paper shows that AI’s economic returns depend on organizational complements (agile practices). Economists studying AI adoption should model complementarities between technology and management practices when estimating productivity or welfare effects.
• Diffusion and absorptive capacity: Larger and tech‑intensive firms gain more, implying adoption externalities and heterogeneous returns. Policy or firm‑level interventions to raise absorptive capacity (skills, complementary investments) can widen AI’s sustainability payoffs.
• Valuation and investment signals: A measurable ESG improvement from AI+agile adoption suggests channels through which digital investments could affect firm value, cost of capital, and investor preferences—important for asset pricing and corporate finance models incorporating ESG.
• Labor and task reallocation: Faster innovation cycles and automated decision‑making imply shifts in required skills (toward analytics, cross‑functional agile teams). Economic models should account for task reallocation and potential short‑term displacement vs. long‑term re‑skilling benefits.
• Policy design: Regulators aiming to promote sustainable corporate behavior may achieve greater impact by incentivizing integrated digital‑organizational upgrades (not only AI hardware/software subsidies), e.g., training, governance standards, disclosure norms that make AI+agile implementation more transparent.
• Measurement caution for empirical work: Reliance on disclosure‑based treatment indicators highlights measurement challenges; future AI economics research should triangulate self‑reports with procurement, software usage logs, or third‑party audits to reduce measurement error.

Limitations to bear in mind (for follow‑on research): observational design despite DiD/PSM cannot fully rule out time‑varying unobservables; treatment defined from disclosures may misclassify true intensity; results are from Chinese listed firms (generalizability to other contexts requires testing).