Firms that adopt AI, cloud services, robotics and smart devices are more likely to export, while big data and blockchain show no clear export effect; AI boosts exports broadly except to China, and robotics predicts exports to all destinations examined.
How Digitalization Shapes Export Potential: Firm-Level Insights From The EU and Beyond
Using probit and multivariate probit models on firm-level data, the study finds that adoption of AI, cloud computing, robotics, and smart devices is positively associated with the probability of exporting (while big data and blockchain are not), and that AI and robotics are linked to exports across most destination regions.
The primary objective of this study is to examine the relationship between the adoption of advanced digital technologies, such as artificial intelligence (AI), and the probability of exporting, utilizing a probit model. The results indicate that adopting AI, cloud computing, robotics, and smart devices significantly increases the likelihood of exporting. Big data analytics and blockchain technologies do not have a statistically significant effect on firms’ likelihood of exporting.The study investigates the relationship between specific digital technologies and export destinations as a secondary objective by applying a multivariate probit model, which accounts for potential correlations among the error terms across destination-specific export decisions. The findings suggest that AI plays a role in exports to all regions except China. Cloud computing determines exports to countries outside the European Union and China. Big data analytics and blockchain technologies show no significant correlations with export destinations. Robotics is found to be a key factor for exports to all destinations, while smart devices are particularly influential for exports to China and other countries.
Summary
Main Finding
Firms that adopt artificial intelligence (AI), cloud computing, and robotics are significantly more likely to export. Effects vary by destination: AI and smart devices are linked to exports to EU markets; cloud computing is associated with exports to all regions except China; big data analytics is associated with non‑EU exports; blockchain shows no significant relationship. Robotics shows a consistent positive association with export participation across all destination regions.
Key Points
- Data: Flash Eurobarometer 486 (SMEs, Start‑ups, Scale‑ups, and Entrepreneurship), 2020. Sample size = 9,242 firms across EU countries and selected non‑EU countries (including UK, US, Turkey, Japan, Brazil, Canada, Norway, etc.). Sectors: manufacturing, retail, services, industry.
- Export outcomes: 34.96% of firms report exporting; 31.14% export to EU, 4.90% to China, 12.62% to other non‑EU/non‑China countries.
- Digital adoption prevalences: cloud computing 53.3%, smart devices 30.4%, big data analytics 16.2%, robotics 9.94%, AI 9.06%, blockchain 3.72%.
- Estimation:
- Primary: binary probit for overall export participation.
- Destination analysis: multivariate probit for simultaneous export decisions (EU, China, other), allowing correlated error terms across destinations.
- Controls: standardized sales per employee, firm age and age^2, log(employees), border proximity dummies, global value chain participation, group ownership, patent indicator, innovation indicator, human capital measure; country and sector fixed effects.
- Main econometric results:
- Positive and statistically significant coefficients for AI, cloud computing, and robotics in the overall export probit.
- No statistically significant overall export association for big data analytics, smart devices, or blockchain.
- Destination heterogeneity: AI and smart devices → EU; cloud → all regions except China; big data analytics → non‑EU destinations; robotics → positive across regions; blockchain → not significant.
Data & Methods
- Source: Flash Eurobarometer 486 (European Commission), 2020 survey of firms across EU and several non‑EU countries.
- Sample: 9,242 observations; descriptive statistics reported unweighted (authors use unweighted multivariate probit).
- Dependent variables:
- EXPORT (binary): firm exports vs operates only domestically.
- EEU, ECHN, EOTHER: binary indicators for exporting to EU, China, and other non‑EU/non‑China countries.
- Key independent variables: binary indicators for firm adoption of AI, cloud computing, big data analytics, robotics, smart devices, blockchain.
- Controls: firm productivity proxy (sales per employee standardized), size, age, age^2, employees (log), geographic proximity, global value chain membership, group ownership, patent applications, recent innovation, staff skill availability; country and sector fixed effects included.
- Estimators:
- Probit model for overall export decision.
- Multivariate probit for destination choices to model correlated unobservables across export destinations.
- Limitations noted by authors: cross‑sectional design (potential endogeneity/reverse causality), self‑reported adoption indicators, low prevalence of some technologies (blockchain), and unweighted estimation.
Implications for AI Economics
- AI adoption increases the probability a firm participates in export markets, supporting theories that treat knowledge diffusion and firm‑level technology adoption as drivers of internationalization. AI should be modeled as a firm‑level productivity/knowledge stock driver in trade models.
- Destination heterogeneity matters: AI and smart device adoption appear particularly valuable for EU export markets (possibly reflecting regulatory, standards, or demand complementarities), while cloud computing supports broader market access by lowering ICT fixed costs.
- Robotics' consistent positive association suggests automation raises competitiveness across destinations by improving productivity and product quality — relevant for models linking capital‑embodied technology to trade margins.
- The absence of a general export effect for big data analytics and blockchain may reflect measurement/adoption intensity issues, early diffusion stages, or sectoral heterogeneity; researchers should be cautious inferring null effects from cross‑sectional binary adoption measures.
- Policy implications:
- Support programs that lower adoption costs and build absorptive capacity for AI, cloud, and robotics (training, subsidies, standards), especially for SMEs.
- Trade and digital policies should be coordinated: digital infrastructure, data governance, and interoperability can amplify the export impact of AI and cloud.
- Destination‑specific trade facilitation (e.g., standards alignment with EU markets) may magnify returns to certain technologies.
- Directions for future research:
- Causal identification (panel data, instrumental variables, policy experiments) to disentangle adoption → export from export → adoption.
- Explore intensity and types of AI (narrow vs. advanced ML), complementarities among technologies, sectoral heterogeneity, and interactions with firm size and foreign ownership.
- Examine export margins (extensive vs intensive), export performance (sales shares, quality), and dynamic effects over time.
Assessment
Paper Typecorrelational
Evidence Strengthlow — Findings are based on observational associations from probit models without a clear causal identification strategy; adoption is endogenous to firm characteristics (productivity, managerial quality, market access) and reverse causality or omitted variable bias could drive results.
Methods Rigormedium — The study uses appropriate discrete-choice models (probit and multivariate probit) and accounts for correlated errors across destination-specific decisions, which is methodologically sound for the questions posed; however, lack of strategies to address endogeneity (IVs, natural experiments, panel variation) and limited information on robustness checks reduce overall rigor.
SampleFirm-level cross-sectional data on firms' adoption of multiple digital technologies and their export status and export destinations (regions/countries); exact sample size, country/sector coverage, time frame, and sampling design are not specified in the provided summary.
Themesadoption innovation
IdentificationCross-sectional regression analysis using probit and multivariate probit models to estimate associations between firm-level adoption of digital technologies (AI, cloud, robotics, smart devices, big data, blockchain) and the probability of exporting and export-destination choices; controls likely included but no exogenous variation, instruments, panel fixed effects, or quasi-experimental source of identification reported.
GeneralizabilityResults are conditional on the (unspecified) country, sector and time coverage of the sample and may not generalize across countries or industries., Cross-sectional design limits inference about dynamics — adoption may follow exporting rather than cause it., Measurement of 'AI' and other technologies may vary across firms and be subject to reporting bias., Aggregate export destination categories (e.g., "China", "EU", "other") may mask heterogeneity in market access and trade costs., Potential sample selection (e.g., survival/response bias) could affect external validity.
Claims (11)
| Claim | Direction | Confidence | Outcome | Details |
|---|---|---|---|---|
| Adopting artificial intelligence (AI) significantly increases the likelihood that a firm exports (probit model result). Adoption Rate | positive | high | likelihood/probability of exporting (firm-level) |
0.3
|
| Adopting cloud computing significantly increases the likelihood that a firm exports (probit model result). Adoption Rate | positive | high | likelihood/probability of exporting (firm-level) |
0.3
|
| Adopting robotics significantly increases the likelihood that a firm exports (probit model result). Adoption Rate | positive | high | likelihood/probability of exporting (firm-level) |
0.3
|
| Adopting smart devices significantly increases the likelihood that a firm exports (probit model result). Adoption Rate | positive | high | likelihood/probability of exporting (firm-level) |
0.3
|
| Adopting big data analytics does not have a statistically significant effect on a firm's likelihood of exporting (probit model result). Adoption Rate | null_result | high | likelihood/probability of exporting (firm-level) |
0.3
|
| Adopting blockchain technologies does not have a statistically significant effect on a firm's likelihood of exporting (probit model result). Adoption Rate | null_result | high | likelihood/probability of exporting (firm-level) |
0.3
|
| AI adoption is positively associated with exports to all destination regions examined except China (multivariate probit model that accounts for correlated errors across destination-specific export decisions). Adoption Rate | mixed | high | exporting to specific destination regions (binary/region-specific firm export decisions) |
0.3
|
| Cloud computing adoption is significantly associated with exports to countries outside the European Union and China (multivariate probit model result). Adoption Rate | positive | high | exporting to specific destination regions (binary/region-specific firm export decisions) |
0.3
|
| Big data analytics and blockchain technologies show no significant correlations with exports to specific destinations (multivariate probit result). Adoption Rate | null_result | high | exporting to specific destination regions (binary/region-specific firm export decisions) |
0.3
|
| Robotics adoption is a key factor (positively associated) for exports to all destination regions examined (multivariate probit result). Adoption Rate | positive | high | exporting to specific destination regions (binary/region-specific firm export decisions) |
0.3
|
| Smart devices adoption is particularly influential (positively associated) for exports to China and to other countries (multivariate probit result). Adoption Rate | positive | high | exporting to specific destination regions (binary/region-specific firm export decisions) |
0.3
|