In ASEAN banks, robust ESG strategies raise risk‑adjusted lending returns outright, whereas AI pays off only when it tightens credit‑risk discipline; stricter supervisors magnify both effects.

Main Finding

ESG strategic intensity and AI capability both improve risk-adjusted lending performance in ASEAN-5 banks, but they operate differently: ESG has a direct positive effect on performance and also works partially through improved credit‑risk discipline (CRD); AI capability has no direct effect on performance and improves outcomes only indirectly via CRD (full mediation). Regulatory pressure strengthens the effects of both ESG and AI on CRD.

Key Points

• Theoretical framing: Resource-Based View (RBV) and Institutional Theory — internal capabilities (ESG, AI) interact with external regulatory forces to shape bank outcomes.
• Causal pathway: ESG strategic intensity → CRD → risk-adjusted lending performance (RALP); AI capability → CRD → RALP.
  • ESG: both direct effect on RALP and indirect effect via CRD (partial mediation).
  • AI: effect on RALP is entirely via CRD (full mediation).
• Credit‑risk discipline (screening rigor, early‑warning systems, covenant enforcement) is the proximate operational mechanism that converts strategic capabilities into better loan outcomes (lower NPLs, more stable provisioning and ROA/ROE).
• Regulatory pressure (variation across ASEAN jurisdictions — e.g., ISSB alignment, MAS FEAT/Veritas uptake, sustainable finance taxonomies) amplifies the ESG→CRD and AI→CRD relationships.
• Practical governance implications highlighted: need for governed data pipelines, explainable/oversightable risk models, effective early‑warning mechanisms, aligned managerial incentives, and supervisory expectations.
• Societal implications: disciplined, ESG‑ and AI‑enabled lending can reduce information frictions, improve equitable credit access for credible SMEs and households, stabilize credit cycles, and mitigate adverse selection — conditional on privacy, transparency, and bias controls.

Data & Methods

• Sample: Cross-sectional employer‑side survey of 486 banking professionals across 62 listed commercial banks in five ASEAN countries (Indonesia, Malaysia, Philippines, Singapore, Thailand).
• Constructs and measurement:
  • ESG strategic intensity: employer/manager perceptions of ESG integration into credit policy, product design, incentives.
  • AI capability: modeled as a formative higher‑order construct (data infrastructure, analytics talent, governance frameworks, lifecycle management, model tools).
  • Credit‑risk discipline (CRD): behavioral/procedural measures — screening rigor, EWS usage, covenant enforcement.
  • Risk‑adjusted lending performance (RALP): institutional outcomes — NPL ratio, provisioning stability, volatility in ROA/ROE.
  • Regulatory pressure: jurisdictional differences in disclosure and supervisory regimes (ISSB alignment, FEAT/Veritas, taxonomies).
• Empirical approach: Partial Least Squares Structural Equation Modeling (PLS‑SEM) with:
  • Second‑order constructs (AI capability formative measurement).
  • Robustness checks for multicollinearity, reliability, validity, and predictive relevance.
  • Mediation assessed via bootstrapped indirect effects; moderation via interaction terms.
• Key empirical results: significant positive paths from ESG→CRD and AI→CRD; CRD→RALP positive and strong; ESG→RALP direct path significant; AI→RALP direct path insignificant; regulatory pressure positively moderates ESG→CRD and AI→CRD.

Implications for AI Economics

• Economic value of AI in banking hinges on organizational integration and governance, not just model accuracy:
  • AI capability alone (analytics, models) does not directly raise risk‑adjusted performance; benefits accrue when AI is institutionalized into disciplined credit processes (screening, monitoring, covenant enforcement).
  • Investments in data pipelines, model explainability, lifecycle governance, and skilled analytics staff are necessary complements to capture AI’s economic returns.
• Regulation amplifies returns to AI capability:
  • Stricter supervisory regimes (disclosure requirements, model governance frameworks) increase the translation of AI capability into disciplined practices and thus into performance gains. Policy design therefore materially affects private returns to AI investments.
• Distributional and systemic effects:
  • Properly governed AI can expand accurate credit assessment (better screening, earlier warnings), helping credible SMEs and households access credit and reducing adverse selection — with positive macroeconomic stabilizing effects.
  • Conversely, weak governance or poorly specified incentives may create model‑driven biases or amplify correlated errors across institutions, introducing systemic risk; privacy, transparency, and bias controls are essential.
• Measurement and research design takeaways for AI economics:
  • Employer‑side, capability‑level measures (vs archival model‑level performance metrics) reveal how organizational routines mediate technology value — future studies should combine perceptual capability measures with longitudinal, transaction‑level outcomes to estimate causal effects and effect sizes.
  • The finding of full mediation for AI suggests evaluation of AI policy should consider intermediate governance variables (CRD) as targets for intervention (e.g., mandated model‑risk management, EWS adoption).
• Policy recommendations:
  • Regulators aiming to harvest welfare gains from AI should focus on setting governance standards (data quality, explainability, audit trails, model validation) and incentives that encourage embedding AI into disciplined credit processes.
  • Supervisory frameworks that increase transparency and require operational controls will both reduce downside risk and raise the private productivity of AI investments.

(Notes: study is cross‑sectional and based on employer perceptions—limits on causal inference; longitudinal and matched institutional performance data would strengthen external validity and quantification of economic magnitudes.)