Across several Central and Eastern European economies, AI adoption rarely boosts output unless firms already have standardized processes, management systems and skilled workers; without these complements, AI gets filtered out in a 'productivity funnel'. Policy and investment should prioritize organizational readiness and skills alongside AI tools to realize productivity gains.

Main Finding

AI diffusion by itself does not reliably raise measured firm-level productivity in the sampled Central and Eastern European economies. Productivity gains appear only when AI diffusion operates together with firm-level organizational readiness (formalized processes/quality systems). Firms lacking these organizational complements become stuck in a “complementarity trap”: they may adopt or operate in AI-rich environments but fail to translate that diffusion into measurable value capture.

Key Points

• Productivity funnel framework: AI’s theoretical potential is progressively filtered through five stages — (1) access & infrastructure, (2) organizational readiness, (3) human capital & behavioral adaptation, (4) institutional environment, and (5) measurement & value capture. Most potential is lost at organizational readiness (stage 2) in early-adoption contexts.
• Complementarity trap: the return to organizational readiness rises with AI diffusion; firms without standardized processes and management systems adopt AI but do not convert it into productivity — generating the contemporary “AI productivity paradox.”
• Hypotheses tested:
  • H1: Productivity gains from AI diffusion materialize primarily in firms with high organizational readiness (formalized processes / quality certification).
  • H2: Workforce skill investments strengthen the AI–productivity association.
• Measurement caveats: many AI-enabled gains are intangible (quality, decision speed, risk reduction) and may be invisible to standard productivity metrics, contributing to the observed disconnect.
• Heterogeneity: AI adoption rates and the size of productivity effects vary by country and firm size, and advanced technologies tend to amplify existing organizational heterogeneity rather than equalize performance.

Data & Methods

• Data sources:
  • Firm-level data: World Bank Enterprise Surveys (WBES) for four CEE economies — Serbia (417 firms, 2024), Croatia (456 firms, 2023), Czechia (248 firms, 2024), Romania (919 firms, 2023). Sample: formal firms with ≥5 employees, nationally representative by sector and size.
  • AI diffusion: Eurostat measure of the share of firms using at least one AI technology, computed at the country × firm size-class level — treated as an environment/contextual variable (not firm-level AI use).
• Key variables:
  • Outcome: log(value added per worker) — standard labor productivity measure constructed from firm sales, intermediates, and employment.
  • Organizational readiness (Z): proxied by quality certification (indicator of formalized processes/management systems).
  • Skills: proxied by formal employee training (firm-level).
  • Digital Infrastructure Index (control): sum of website presence, online sales, and internet access constraints (range 0–3).
  • Controls: process innovation, firm age, sector, country, and size-class fixed effects.
• Empirical specification:
  • ln(VA/L)i,c,s,k = β1 Zi + β2 AIc,s + β3(Zi × AIc,s) + γ Xi + δc + δs + δk + εi,c,s,k
  • Interaction term tests complementarity between organizational readiness and AI diffusion (central to H1).
  • Standard errors: heteroskedasticity-robust.
• Identification & limitations:
  • Identification leverages cross-sectional variation in AI diffusion across country–size environments and within-environment firm heterogeneity in readiness and skills.
  • Key limitations: cross-sectional design (no causal dynamics/panel), AI diffusion is environment-level (not firm-level adoption), and organizational readiness proxied by certification (an imperfect but observable indicator). Measurement issues for intangible value remain.

Implications for AI Economics

• Explains the modern productivity paradox: weak aggregate productivity responses to AI can reflect organizational and institutional bottlenecks, not solely technological failure. Economics of AI must therefore emphasize complements, not just diffusion.
• Policy directions for early-stage/adopting economies and SMEs:
  • Prioritize investments in organizational capabilities (process standardization, management practices, quality systems) alongside digital infrastructure.
  • Support bundled interventions: training that is tightly linked to process redesign and implementation support (e.g., management consulting, change management subsidies).
  • Subsidize or accelerate quality management and certification programs that build absorptive capacity for AI.
  • Improve measurement of AI impacts by capturing intangible gains (quality, decision speed, risk reduction) and promote collection of firm-level AI adoption data (task-level usage, tools, deployment depth).
  • Strengthen institutional complements (labor market flexibility, IP regimes, competitive pressures) to allow productive firms to scale and laggards to restructure.
• Firm strategy:
  • Firms should sequence investments: secure basic digital readiness and then invest in organizational reforms and targeted skill development before expecting large productivity returns from AI.
  • Emphasize process redesign, governance, and validation/oversight routines to realize and sustain productivity gains.
• Research agenda for AI economics:
  • Use panel/quasi-experimental designs and firm-level measures of AI use to establish causal pathways.
  • Assess the timing (J-curve dynamics) and persistence of productivity returns, and quantify intangible value capture.
  • Study how AI amplifies heterogeneity across firms, sectors, and countries and the distributional consequences for wages and employment.
• Broader implication: AI is a general‑purpose technology whose macroeconomic payoff depends heavily on complementary organizational capital. Effective policy and managerial responses that build those complements are central to turning AI diffusion into broad-based productivity growth.