Italian firms that adopt AI see higher productivity and profits and shift hiring toward white-collar roles, while overall employment remains unchanged; adoption is currently concentrated among larger, knowledge‑intensive firms.

Main Finding

AI adoption among Italian firms (2024) is limited but growing (10% current users; ~30% plan adoption within two years). Adoption is concentrated in larger, knowledge-intensive, and higher-labour-cost firms. Using linked survey and administrative data, the authors find that AI adoption raises firm-level labour productivity and profitability, reallocates employment toward higher-skilled (white-collar) occupations while reducing blue-collar employment, and does not change overall employment. Adopting firms also expect smaller own-price increases and lower medium-to-long-run inflation than non-adopters, consistent with anticipated efficiency gains.

Key Points

• Prevalence and plans
  • ~10% of firms report current AI use (2024).
  • ~30% plan to adopt AI within two years.
• Adoption correlates
  • More likely among larger firms, knowledge-intensive sectors, and firms with higher labour costs.
  • Suggests organizational capacity, technological complementarities, and efficiency incentives drive adoption.
• Firm-level effects (difference-in-differences evidence)
  • Positive impacts on labour productivity and profitability after adoption.
  • No detectable change in total employment levels.
  • Statistically significant reallocation across occupations:
    • Expansion of white-collar (higher-skilled) employment.
    • Contraction of blue-collar employment.
• Expectations and pricing
  • Adopters anticipate smaller increases in their own prices and lower medium- to long-term inflation than non-adopters.
  • Implies expected efficiency gains influence pricing behaviour and macro expectations.

Data & Methods

• Data sources
  • Newly collected firm survey on AI adoption and expectations (2024).
  • Linked administrative balance-sheet data and employer–employee records (for outcomes like productivity, profitability, employment by occupation).
• Sample
  • Italian firms; descriptive stats: 10% adopters, 30% planned adopters.
  • Adoption concentrated in larger and knowledge-intensive firms.
• Empirical strategy
  • Difference-in-differences (DiD) design comparing adopting vs non-adopting firms over time.
  • Outcomes: labour productivity, profitability, total employment, employment composition by occupation, firms’ price-change expectations, inflation expectations.
  • Likely includes standard controls and fixed effects (firm and time) and robustness checks (event-study-style dynamics and placebo tests are implied by the DiD framing, though specific diagnostics are not detailed in the summary).
• Identification caveats
  • DiD helps control for time-invariant firm heterogeneity and common shocks, but adoption may still be endogenous (e.g., anticipatory investment by firms on upward trends).
  • Measurement of “AI adoption” is survey-based and may capture heterogeneous technologies and intensities of use.

Implications for AI Economics

• Microeconomic effects and mechanisms
  • Evidence supports productivity and profitability gains from AI at the firm level.
  • Occupational reallocation (upskilling bias) rather than net job destruction: AI appears to substitute for some blue-collar tasks while complementing higher-skilled white-collar work.
  • Adoption clustering in larger, knowledge-intensive firms highlights the role of organizational capacity and complementarities (complementary skills, processes, and data).
• Labor-market policy
  • Emphasizes need for training/retraining and education to ease transitions toward higher-skilled roles and to capture complementarities.
  • Target policies for smaller and less knowledge-intensive firms to reduce adoption barriers (finance, implementation support).
• Competition and distributional concerns
  • Heterogeneous adoption and benefits may increase firm-level dispersion (returns to scale, incumbent advantages), with implications for market structure and inequality.
• Macroeconomic and price-level implications
  • Adopters’ expectation of smaller price increases suggests AI-driven efficiency gains could be disinflationary at the firm level; widespread adoption may affect aggregate inflation dynamics.
  • Results inform macro models that incorporate productivity gains and sectoral reallocation from AI.
• Measurement and research priorities
  • Need for standardized measurement of AI adoption intensity and types to compare effects across firms and countries.
  • Longer-term and cross-country studies to assess persistence of productivity gains, wage effects, and aggregate employment outcomes.
  • Further causal work to isolate mechanisms (task-level changes, complementary investments, managerial practices) and to address potential selection into adoption.

Limitations and open questions to note: external validity beyond Italy, heterogeneity across AI technologies and sectors, possible endogeneity of adoption timing, and long-run effects on wages, career progression, and aggregate labor demand remain to be explored.