AI Adoption Productivity and Employment | BIS Research

AI Adoption and the European Productivity Challenge

Europe faces a defining economic paradox: despite hosting world-class research institutions and a highly educated workforce, the continent consistently lags behind the United States in translating technological innovation into firm-level productivity growth. A landmark working paper from the Bank for International Settlements (BIS) by Aldasoro and colleagues directly confronts this challenge, providing the most rigorous causal evidence to date on how AI adoption affects productivity and employment across European firms.

The study, leveraging data from over 12,000 non-financial EU firms surveyed through the European Investment Bank Investment Survey (EIBIS) from 2019 to 2024, combined with financial data from Moody’s ORBIS, offers a comprehensive picture of AI’s economic impact. The findings carry profound implications for policymakers, business leaders, and workers navigating the rapidly evolving AI landscape in Europe.

What makes this research particularly valuable is its methodological rigor. Rather than relying on simple correlations — which can be misleading when more productive firms naturally adopt more technology — the BIS team developed an innovative instrumental variable approach that isolates the causal effect of AI adoption on firm performance. The results challenge both techno-optimists and techno-pessimists, painting a nuanced picture of measured gains with important distributional implications.

Measuring AI Adoption Across European Firms

The BIS research uses the EIBIS survey question about firms’ use of “big data analytics and AI” to measure adoption, capturing both partial implementation (AI used in specific business areas) and full organizational integration. This measure, while necessarily broad, provides consistent cross-country and cross-sector comparisons that reveal striking patterns in how AI diffusion is proceeding across Europe.

The headline adoption figures reveal substantial variation. When weighted by firm value-added, approximately 45% of large European firms (250+ employees) report AI adoption, compared to just 24% of small firms (10-49 employees). This size-based gap reflects the reality that AI implementation requires upfront investment in technology infrastructure, data systems, and skilled personnel that larger organizations can more readily absorb.

Country-level differences are equally dramatic. Sweden leads European AI adoption at approximately 52%, reflecting the country’s strong digital infrastructure, highly educated workforce, and supportive innovation ecosystem. At the other end, Romania registers only around 22% adoption, illustrating how economic development, financial market depth, and institutional quality create vastly different conditions for technology diffusion.

The US comparison provides important context. With an average reported AI adoption rate of approximately 34%, the United States serves as both a benchmark and, in the BIS methodology, a source of exogenous variation for estimating causal effects. Notably, the most financially developed EU countries achieve adoption rates comparable to the US average, suggesting that the “European gap” is not uniform but concentrated in countries with less developed financial and innovation ecosystems.

The 4% Productivity Gain: Causal Evidence from BIS Research

The central finding of the BIS research is that AI adoption increases firm-level labor productivity by approximately 4%. This estimate emerges from an instrumental variable approach that addresses the fundamental identification challenge: more productive firms may simply be more likely to adopt AI, making naive correlations unreliable guides to AI’s actual impact.

The identification strategy draws inspiration from the influential Rajan and Zingales (1998) approach. Each European firm is matched to similar US firms based on sector, size, and observable characteristics that predict AI adoption — including investment intensity per employee, capital market access, innovation activity, and use of strategic business monitoring systems. The AI adoption rate of these matched US peers then serves as an instrument for the European firm’s adoption, capturing variation in AI exposure that is plausibly exogenous to individual firm characteristics.

The difference between naive and causal estimates is instructive. Simple OLS correlations suggest AI adoption is associated with productivity increases as large as 16% — a figure that almost certainly reflects selection bias, as high-performing firms gravitate toward technology adoption. Once the instrumental variable approach strips away this selection effect, the estimated productivity gain settles at a more modest but still economically meaningful 4%.

The European Investment Bank’s survey data, spanning from 2019 through 2024, captures the critical period during which AI adoption accelerated significantly. The regression models include firm-level controls for lagged investment, profitability, leverage, and assets, alongside fixed effects for firm size, age, and country-sector-year combinations, providing confidence that the 4% estimate reflects a genuine causal relationship rather than confounding factors.

Capital Deepening: How AI Drives Productivity Growth

Perhaps the most important mechanistic finding is that AI-driven productivity gains operate through capital deepening rather than labor displacement. When firms adopt AI, they increase their investment per worker — particularly in intangible assets such as software, data systems, and process automation — creating more productive working environments without necessarily reducing headcount.

This mechanism aligns with a growing body of evidence suggesting that AI functions primarily as a complement to human labor rather than a substitute. By automating routine analytical tasks, AI frees workers to focus on higher-value activities requiring judgment, creativity, and interpersonal skills. The capital deepening channel implies that firms are investing in the tools and systems that make existing workers more productive rather than simply replacing them with machines.

The BIS data shows that investment per employee becomes significantly more positive after instrumenting AI adoption, confirming that the productivity pathway runs through enhanced capital intensity. Firms that adopt AI invest more per worker, and this additional investment — in software platforms, data infrastructure, AI-specific hardware, and supporting systems — translates directly into measurable productivity improvements.

The wage data reinforces this interpretation. Workers in AI-adopting firms earn higher wages on average, consistent with the complementarity hypothesis. If AI were primarily displacing workers, downward wage pressure would be expected. Instead, the wage premium suggests that AI adoption increases the marginal productivity of remaining (and additional) workers, enabling firms to justify higher compensation.

AI Adoption and Employment: No Evidence of Job Displacement

The employment findings may be the research’s most policy-relevant contribution. Despite widespread public anxiety about AI-driven job losses, the BIS study finds no statistically significant evidence of short-run employment displacement at the firm level. The instrumental variable estimate produces a small negative coefficient for employment that fails to reach conventional significance thresholds.

This null finding is substantively important and requires careful interpretation. Simple OLS regressions suggest a positive association between AI adoption and employment — approximately 8% higher employment for adopters in controlled specifications. However, this likely reflects selection: growing, expanding firms both hire more workers and adopt more technology. The causal IV estimate, by stripping away this selection, reveals that AI adoption itself does not drive employment changes in either direction during the sample period.

Several important caveats apply. The EIBIS survey covers 2019-2024, meaning results capture relatively early-stage AI deployment when firms may be augmenting rather than automating tasks. Longer-run effects, as AI systems mature and firms reorganize production more fundamentally, could differ. Additionally, the study examines firm-level employment, not occupational composition — it is possible that AI adoption shifts the mix of occupations within firms even if total headcount remains stable.

The implications for workforce planning are significant. Rather than defensive strategies focused on preventing job losses, the evidence supports proactive approaches that help workers develop complementary skills to work alongside AI systems effectively.

The AI Adoption Gap: Large Firms vs Small Businesses

The distributional findings present a challenge for policymakers. AI adoption and its productivity benefits are heavily concentrated among medium and large firms, raising concerns about widening firm-level inequality across Europe. Large firms with 250+ employees adopt AI at nearly twice the rate of small firms, and they extract proportionally larger productivity gains from implementation.

Several factors explain this size-based adoption gap. Large firms can more easily absorb the fixed costs of AI infrastructure — data platforms, computing resources, specialized personnel, and process redesign. They possess larger and more diverse datasets that improve AI model performance. They have dedicated IT and innovation teams capable of identifying, implementing, and scaling AI solutions. And they enjoy better access to the financial resources needed for upfront technology investments.

The productivity concentration compounds the adoption gap. Not only do large firms adopt more, but they derive greater benefit per unit of adoption. This suggests that the complementary capabilities required to extract value from AI — managerial expertise, organizational flexibility, data governance, and skilled human capital — are themselves concentrated in larger organizations. The result is a self-reinforcing dynamic where AI amplifies rather than reduces pre-existing differences in firm performance.

For small and medium enterprises (SMEs) that form the backbone of European economies, this pattern is concerning. Without targeted policy intervention, AI could become a force for market concentration rather than competitive dynamism, as large adopters pull further ahead of smaller competitors unable to access or effectively deploy the technology.

Financial Development and AI Adoption Across Europe

One of the study’s most striking findings connects AI adoption rates to financial market development. EU countries with deeper, more liquid, and more diversified financial markets exhibit significantly higher AI adoption rates — patterns that closely mirror adoption rates in the United States. Conversely, countries with less developed financial systems lag substantially behind.

The mechanism is intuitive. AI adoption requires upfront investment — in technology, data infrastructure, training, and organizational change — before productivity benefits materialize. Firms in countries with well-developed equity markets, venture capital ecosystems, and accessible debt financing can more readily fund these investments. In countries where SMEs face credit constraints and equity markets are thin, the barrier to entry for AI adoption is correspondingly higher.

The temporal pattern in less-developed EU countries reveals both progress and limitations. AI adoption in these economies rose from approximately 17% in 2021 to 27% in 2022, suggesting a significant catch-up dynamic — potentially driven by pandemic-era digitization pressures and EU recovery fund investments. However, adoption then flattened in 2023-2024, indicating that the initial surge may have captured “low-hanging fruit” while structural barriers to further diffusion remain.

This finding has direct policy implications. Programs that improve SME access to finance — whether through public guarantee schemes, venture capital incentives, or subsidized innovation lending — are not merely financial sector development tools. They are, in effect, AI diffusion policies that can help close the adoption gap between more and less developed European economies.

Complementary Investments That Maximize AI Productivity

The BIS research identifies complementary investments as critical amplifiers of AI productivity gains. Firms that combine AI adoption with investments in software, data infrastructure, and workforce training experience significantly larger productivity improvements than those that implement AI in isolation. This finding underscores that AI is not a plug-and-play technology — its value depends on the organizational ecosystem in which it is deployed.

Software and data investments create the digital substrate on which AI operates. Machine learning models require clean, structured, and accessible data; integration with existing business systems demands robust API architectures and data pipelines; and ongoing model refinement needs continuous data flows and monitoring infrastructure. Without these foundations, AI deployments often underperform or fail entirely.

Workforce training represents the human complement to technological investment. The BIS evidence suggests that firms investing in employee upskilling — developing capabilities in data interpretation, AI tool usage, prompt engineering, and human-AI collaboration — extract substantially more value from their AI investments. This resonates with research from the OECD Employment Outlook emphasizing that technology adoption and human capital development are fundamentally intertwined.

The complementarity finding carries a clear message for firms considering AI adoption: technology procurement is necessary but insufficient. Maximizing AI’s productivity impact requires a holistic approach encompassing data infrastructure, process redesign, organizational change management, and sustained investment in workforce capabilities. Firms that approach AI as merely a technology purchase rather than an organizational transformation will likely see disappointing returns.

Policy Recommendations for Inclusive AI Adoption

The BIS research generates a comprehensive set of policy recommendations aimed at broadening AI adoption while maximizing its productivity benefits. The emphasis is on creating conditions for inclusive adoption that prevents AI from becoming primarily a tool for large-firm competitive advantage.

SME-targeted financial support emerges as a priority. This includes dedicated AI integration vouchers, subsidized access to shared AI platforms and labs, and advisory services that help smaller firms identify and implement AI solutions appropriate to their scale and sector. Public-private partnerships can reduce the fixed costs that currently deter small firm adoption.

Workforce reskilling programs should prioritize “fusion skills” that combine domain expertise with AI literacy. Rather than training everyone to be data scientists, the emphasis should be on developing practical capabilities in prompt engineering, data stewardship, human-in-the-loop system design, and AI-augmented decision-making. Modular, lifelong learning formats that allow workers to upskill without leaving employment are essential.

Financial market development serves as an enabling condition for AI diffusion. Policies that deepen equity markets, strengthen venture capital ecosystems, and improve SME access to innovation finance create the conditions for broader technology adoption. The European Central Bank and national regulators can facilitate this through proportionate regulation that encourages risk capital flows to innovative firms.

Data governance frameworks must balance competing objectives. Enabling data sharing to support AI development while protecting privacy, ensuring competition, and maintaining trust requires sophisticated regulatory design. Sectoral data-sharing agreements, standardized data access protocols, and clear rules on permissible AI training data use can facilitate diffusion without sacrificing protection.

Competition policy vigilance is necessary to prevent AI from reinforcing market concentration. As large adopters gain productivity advantages, antitrust authorities should monitor for emergent dominance, review AI-related mergers carefully, and ensure that data portability and interoperability requirements prevent lock-in effects that disadvantage smaller competitors.

Future Outlook for AI Productivity and Workforce Impact

The BIS research captures a specific moment in AI’s integration into the European economy — the early-to-mid deployment phase from 2019 through 2024. As AI systems mature, adoption deepens, and organizations restructure more fundamentally around AI capabilities, the productivity and employment dynamics documented here may evolve significantly.

On the productivity side, the 4% causal estimate likely represents a lower bound of AI’s eventual impact. As firms move from initial AI deployment to deeper organizational integration, and as AI capabilities themselves improve through advances in large language models, multimodal systems, and autonomous agents, the productivity dividend could grow substantially. However, translating firm-level gains into aggregate productivity growth depends on diffusion breadth — if adoption remains concentrated among large firms, the macroeconomic impact will be muted.

The employment outlook requires continued monitoring. The current null finding on job displacement is encouraging but should not breed complacency. As AI systems become more capable and as firms complete the organizational learning required to deploy AI more comprehensively, the balance between task augmentation and task automation may shift. Active labor market policies — including retraining programs, portable benefits, and transition support — should be developed proactively rather than reactively.

The international competitive dimension adds urgency. If European firms fall further behind US counterparts in AI adoption and productivity, the consequences extend beyond individual firm performance to Europe’s overall economic competitiveness and geopolitical influence. The BIS evidence suggests that closing the adoption gap is achievable — financially developed EU countries already match US adoption rates — but requires coordinated policy action on multiple fronts simultaneously.