Digital tools help Portuguese SMEs sell more abroad but curb a key learning channel; exports enabled by digitalization raise sales yet deliver smaller productivity improvements, implying a trade-off between market access and organizational learning.

Main Finding

Investments in digital technologies have a dual effect for SMEs: they increase export activity (higher export sales) but simultaneously reduce the productivity gains that firms derive from export-related learning. In other words, digitalization boosts the quantity and ease of international engagement but dampens the rich, human-mediated learning processes (learning-by-exporting) that historically translated export exposure into productivity improvements.

Key Points

• Dual effects framework: “bright” side (export gains) and “dark” side (learning losses).
• Bright side mechanisms:
  • Digital tools (e‑commerce platforms, cloud services, AI-driven translation/chatbots, automation) lower search and transaction costs, enable continuous market monitoring, and make it easier for SMEs to identify and connect with foreign partners.
  • Digitalization can convey competence-based trust (standardization, verifiable digital records), facilitating more operationally effective exporting relationships.
• Dark side mechanisms:
  • Digitalization standardizes routines, automates decisions, and reduces the diversity of observable partner behavior, weakening opportunities for observational/vicarious learning.
  • Algorithms and platform processes are often opaque (“black box”), reducing visibility into partners’ causal reasoning and contextual decisions, which undermines knowledge transfer.
  • Reduced human-to-human interaction can lower the richness of exchanges necessary for translating export relationships into domestic productivity improvements.
• Empirical result: Export activity that is induced or mediated by digitalization yields smaller productivity gains than export activity not facilitated by these digital technologies.
• Additional qualitative case studies and robustness checks in the paper corroborate the mechanism and suggest hybrid (human + digital) approaches as a possible mitigation.

Data & Methods

• Data: Firm-level panel combining management and accounting measures for 20,133 Portuguese firms over 2010–2019.
• Outcome variables: export activity (export sales performance) and productivity gains attributable to exporting (learning-by-exporting).
• Empirical strategy: hypotheses on (1) digitalization → export activity and (2) digitalization-mediated export activity → learning-by-exporting were tested using the firm-level panel data. The authors report supportive results and supplement quantitative work with additional analyses and case studies to unpack mechanisms.
• Note: The article states the use of “new analytical techniques” to increase rigor but does not specify details in the provided excerpt; results are reported as robust across supplemental analyses.

Implications for AI Economics

• Returns to digitalization are heterogeneous and time‑dependent:
  • Short-run gains: digital/AI adoption can increase market access and sales for SMEs, implying positive demand-side effects in trade models.
  • Long-run costs: diminished learning-by-exporting implies weaker productivity spillovers from trade, altering the conventional trade→productivity channel in growth and international economics models.
• Human–AI complementarity vs substitution:
  • Evidence suggests partial substitution of boundary-spanning and interpretive human activities by digital systems can erode knowledge accumulation—critical for models that assume complementarities between trade exposure and firm learning.
  • Policies and firm strategies that preserve human roles in interpretive, boundary-spanning tasks (i.e., hybrid approaches) could sustain long-run productivity gains while retaining digital-enabled market access.
• Measurement and modeling issues:
  • Standard productivity measures may overstate the net welfare benefits of export expansion if they ignore reduced learning spillovers caused by digital mediation.
  • Models of technology diffusion and firm heterogeneity should incorporate the possibility that digital adoption changes not only costs of trading but also the endogenous formation of firm capabilities and learning externalities.
• Distributional and policy considerations:
  • SMEs in digitally mediated export channels may face slower capability accumulation compared to firms that engage in richer human-mediated partnerships—this has implications for inequality across firms and regions.
  • Regulatory and policy levers worth exploring: improving transparency of platform/algorithmic decision-making, subsidizing training and boundary‑spanning roles, supporting hybrid human–digital workflows, and incentivizing modes of international engagement that preserve learning channels (e.g., partnership programs, exchange visits).
• Research directions for AI economics:
  • Disaggregate digital technologies (marketplace platforms vs AI decision systems vs standardization tools) to identify which technologies most strongly drive the trade-off.
  • Quantify dynamic effects: do initial export gains from digitalization eventually lead to capability accumulation via other channels, or are learning losses persistent?
  • Cross-country comparisons to assess whether institutional environments, firm capabilities, or sectoral composition moderate the dual effects.
  • Micro-level studies linking types of human–AI interaction (assistive vs autonomous) to learning outcomes and productivity trajectories.

In sum, this paper cautions that AI and digitalization bring clear trade and market-access benefits for SMEs but can weaken the learning channels that historically amplified the productivity gains from exporting—an important consideration for economic models, firm strategy, and policy in the AI era.