Generative AI raises innovation in Canadian SMEs, but the dividend depends on governance: transparency, privacy protection and accountability boost returns, while early-stage fairness controls can slow experimentation. The result implies uneven, capability-dependent gains from GenAI that may widen firm-level disparities unless governance support is provided.

Main Finding

Generative AI (GenAI) adoption increases innovation performance in SMEs, but the size and direction of that payoff depend on firms’ ethical AI governance capabilities. Transparency/explainability (TE), privacy/data protection (PD), and accountability/responsibility (ACCT) strengthen the GenAI → innovation link, while fairness/bias control (FB) unexpectedly weakens it—consistent with a short-run, maturity-stage trade-off in resource-constrained SMEs where rigid or poorly integrated fairness controls can slow early experimentation. Innovation gains then translate into greater adaptability/resilience, operational efficiency, and revenue growth.

Key Points

• Theoretical framing: resource-based view (RBV) + dynamic capabilities; ethical AI governance is treated as a capability bundle (TE, FB, PD, ACCT) that conditions returns to GenAI.
• Core empirical result: GenAI adoption → higher innovation performance.
• Moderation by governance components:
  • TE, PD, ACCT: positive moderators (they amplify GenAI’s innovation payoff).
  • FB: negative moderator (may reduce short-term innovation gains in SMEs).
• Adoption drivers are asymmetric:
  • Positive: green-innovation orientation is the strongest positive driver of GenAI adoption.
  • Negative: carbon-footprint pressure and economic disruption are negatively associated with adoption.
• Downstream outcomes: higher innovation performance is positively associated with adaptability/resilience, operational efficiency, and revenue growth.
• Reframe: ethical AI governance should be seen not only as compliance cost but as a dynamic capability that shapes whether GenAI delivers value.

Data & Methods

• Sample: 218 usable manager responses from Canadian small and medium-sized enterprises (SMEs).
• Analysis:
  • Covariance-based structural equation modeling (CB-SEM) implemented in AMOS 22 to test hypothesized paths.
  • Moderation examined via multigroup analysis with measurement-invariance checks to ensure valid comparisons across groups.
• Key constructs: GenAI adoption, innovation performance, governance subcomponents (TE, FB, PD, ACCT), firm pressures and orientations (green-innovation orientation, carbon-footprint pressure, economic disruption), downstream firm outcomes.
• Limitations of the data/methods: cross-sectional survey design, self-reported measures, single-country SME sample (Canada), potential endogeneity and reverse causality not fully addressed in this design.

Implications for AI Economics

• Heterogeneous returns: Productivity and innovation gains from GenAI are heterogeneous across firms and depend on complementary governance capabilities. Macro- and micro-economic models that assume uniform gains from AI should account for these complementarities.
• Complementarities and dynamic capabilities: TE, PD, and ACCT act as productive complementarities that increase the marginal benefit of GenAI investments; investments in these governance capabilities can raise ROI on AI for SMEs.
• Trade-offs and staging of fairness interventions: The negative short-run effect of FB suggests a potential trade-off—strict fairness controls may impede early-stage experimentation in resource-limited firms. Policy and standards should consider staged or supported implementation of fairness safeguards to avoid suppressing innovation while ensuring long-term responsible scaling.
• Policy levers: Public support (e.g., subsidies, shared governance toolkits, standardized APIs, training) to reduce the fixed costs of transparency, privacy, and accountability capabilities can raise GenAI adoption quality and the aggregate innovation payoff among SMEs.
• Diffusion and inequality: Because governance capability endowments vary, GenAI-driven productivity improvements may be uneven, potentially widening gaps between firms that can build governance bundles and those that cannot. This has implications for firm dynamics, market concentration, and distributional outcomes.
• Research implications: Economic analyses of AI should incorporate governance-capability heterogeneity, consider dynamic maturity effects (short vs long run), and use longitudinal or quasi-experimental designs to establish causal effects of governance investments on AI returns.