Disclosure-based measures of digital strategy and AI adoption correlate with stronger organizational-culture disclosures and higher ROA in a five-firm Indonesian tech panel, but the tiny sample and correlational design limit causal interpretation.

Main Finding

Digital Business Strategy (DBS) and the intensity of Artificial Intelligence Adoption (AIAI) positively reshape disclosure‑based organizational culture (OCDI), and OCDI in turn raises accounting profitability (ROA). Organizational culture partially mediates the effects of DBS and AIAI on ROA: direct effects of DBS/AIAI on ROA are small, while indirect (culture‑mediated) effects are economically meaningful and statistically significant.

Key Points

• Core results (panel regressions / bootstrap mediation):
  • DBS → OCDI: β = 0.412, p < 0.01.
  • AIAI → OCDI: β = 0.385, p < 0.01.
  • OCDI → ROA: β = 0.298, p < 0.01.
  • Direct DBS → ROA: β = 0.078, p < 0.05 (small).
  • Direct AIAI → ROA: β = 0.065, p < 0.05 (small).
  • Indirect (DBS → OCDI → ROA): 0.1228, 95% CI [0.0345, 0.2215] (partial mediation).
  • Indirect (AIAI → OCDI → ROA): 0.1147, 95% CI [0.0310, 0.2082] (partial mediation).
• Interpretation: Investments in digital strategy and AI yield limited direct accounting returns unless they also induce cultural change (data‑driven norms, learning, collaboration) that converts technical capability into better asset utilization.
• Diagnostics: VIFs < 3.5; fixed‑effects model used for OCDI model, pooled OLS for ROA model; mediation tested with firm‑block bootstrap (5,000 resamples).

Data & Methods

• Sample: Balanced panel of technology sector firms listed on the Indonesia Stock Exchange (IDX), 2021–2025. N = 25 firm‑year observations from 5 firms (ATIC, MTDL, PTSN, NFCX, ZYRX).
• Outcome: ROA = Net Income / Total Assets.
• Key independent/mediator measures constructed via quantitative content analysis of annual reports, sustainability and governance disclosures using a 0–3 rubric (0 = no disclosure … 3 = measurable outcomes):
  • DBS Index (DBSI): 6 dimensions (digital scope, scale, speed, value creation, process integration, digital governance).
  • AIAI Index: 8 dimensions (breadth, depth, infrastructure, applications, talent/training, investments/partnerships, governance/ethics, operational impact).
  • OCDI (Organizational Culture Disclosure Index): 4 Denison dimensions (Involvement, Consistency, Adaptability, Mission).
  • Indices normalized to [0,1].
• Controls: Firm size (ln assets), leverage (debt/equity), revenue growth.
• Econometric approach:
  • OCDI model: Fixed Effects panel regression.
  • ROA model: Pooled OLS (chosen for parsimony).
  • Mediation: Firm‑block bootstrap (5,000 reps) to obtain bias‑corrected 95% CIs for indirect effects.

Implications for AI Economics

• Complementarities matter: The paper provides empirical support that returns to AI investment are conditional on organizational complements (culture). Economic models and empirical estimations of AI productivity should explicitly include organizational/capability complementarities rather than treating AI adoption as a standalone input.
• Measurement approach: Disclosure‑based indices (quantitative content analysis of reports) are a feasible way to operationalize intangible constructs (AI intensity, cultural change) when administrative data are unavailable—useful for cross‑firm studies in emerging markets—but disclosure measures can reflect strategic signaling and may be endogenous.
• Policy & managerial design:
  • Cost‑benefit analyses of AI adoption should budget for complementary investments (training, governance, change management) and expect lagged payoffs as culture shifts.
  • Policymakers aiming to raise firm‑level productivity via AI (e.g., subsidies, tax incentives) should pair incentives with support for workforce reskilling and organizational change programs.
• Caveats for inference & next steps in AI economics research:
  • Small sample and limited sectoral scope (5 firms, 25 firm‑years) constrain external validity—estimates may not generalize across industries or countries.
  • Potential endogeneity/reverse causality: higher ROA firms may disclose more or invest more in AI; pooled OLS for ROA and disclosure‑based measures raise identification concerns.
  • Suggested robustness / future designs: larger samples, longer panels, lagged regressors, instrumental variables or natural experiments, difference‑in‑differences around exogenous AI shocks, and alternative outcomes (labor productivity, market value, innovation output).
• Practical modeling implication: When estimating returns to AI at firm or macro levels, include interaction terms or mediation structures capturing organizational capital (culture, skills, managerial practices) and allow for dynamic adjustment paths.