Telling firms how many rivals use advanced machines nudges adoption: an information treatment in a Bank of Italy survey raised firms' intended uptake of robotics but left AI intentions unchanged, highlighting information frictions as a brake on technology diffusion.

Main Finding

Providing firms with accurate information about competitors' current adoption rates causally increases firms' intended future adoption of robotics but has no significant effect on intended AI adoption. Firms substantially underestimated competitors' current adoption; receiving information led them to update expectations about future adoption, demonstrating that information frictions inhibit coordination in technology diffusion.

Key Points

• First large-scale field experiment testing strategic complementarities in firm technology adoption (embedded in a Bank of Italy survey of ≈3,000 firms).
• Firms were asked to report beliefs about competitors' adoption of two technologies: Artificial Intelligence (AI) and robotics.
• Half the sample was randomly shown accurate information about actual adoption rates among peers.
• Baseline: most firms substantially underestimated competitors' current adoption of both technologies.
• Treatment effect: information caused firms to update beliefs about future competitor adoption.
• Outcome effects: treated firms increased their intended future adoption of robotics (statistically significant); no significant treatment effect observed for intended AI adoption.
• Provides causal evidence that information frictions impede coordination in innovation and can slow diffusion of at least some advanced technologies.

Data & Methods

• Sample: ~3,000 firms surveyed by the Bank of Italy. (Cross-industry Italian firms; survey-embedded field experiment.)
• Elicitation: firms reported beliefs about competitors' current and expected future adoption of AI and robotics.
• Experiment: randomized information provision — half of firms received accurate peer-adoption statistics prior to reporting future adoption intentions.
• Outcomes: (a) belief updating about competitors' future adoption; (b) firms' stated intended future adoption of AI and robotics.
• Identification: randomized assignment of information establishes causal inference on the effect of information on beliefs and intended adoption.
• Analysis: comparison of means and likely regression adjustments to estimate treatment effects and test statistical significance (heterogeneity explored across firms but detailed results not provided here).
• Limitations: measured outcome is intended adoption (stated plans), not observed realized adoption; context is Italy and survey respondents, so external validity may be limited; differential responses across technologies suggest heterogeneous mechanisms.

Implications for AI Economics

• Information frictions matter: firms underestimating peer adoption implies that lack of accurate information can prevent mutually beneficial coordination on technology choices.
• Strategic complementarities are empirically relevant: when firms see higher peer adoption they may increase their own investment, consistent with positive externalities or network effects in some technologies (here, robotics).
• Technology heterogeneity: the response differs by technology — robotics responds to information nudges whereas AI does not — implying that complementarities, observability, adoption costs, or perceived returns vary across technologies. Policies should be tailored by technology.
• Policy levers: low-cost informational interventions (disseminating peer-adoption statistics) can accelerate diffusion for certain technologies and help overcome coordination failures. For AI, additional policies (subsidies, technical assistance, training, standards, risk mitigation) may be needed if information alone is insufficient.
• Research agenda: track realized adoption post-intervention to measure long-run effects; unpack mechanisms behind the AI vs robotics difference (cost, irreversibility, skill constraints, regulatory uncertainty, observability); test in other institutional contexts and consider interactions with complementary policies (finance, workforce training).