An Experiment on AI Adoption
How to cite
Faia, Ester; Perez-Truglia, Ricardo; Cullen, Zoe; Guglielminetti, Elisa; Rondinelli, Concetta. 2025. "An Experiment on AI Adoption." AEA RCT Registry. AEARCTR-0016177. https://doi.org/10.1257/rct.16177-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
In partnership with Banca d'Italia, we conducted an information-provision experiment with a sample of up to 5,000 Italian firms. We added a new module to a long-standing survey of firms to elicit prior beliefs about the share of competitors that have already adopted advanced technologies (defined as predictive AI, generative AI, or robotics). Half of the firms were then randomly assigned to receive accurate information about the actual share of competitors who had adopted these technologies. After the information-provision stage, we elicited posterior beliefs about the share of competitors expected to adopt advanced technologies in the future, as well as the firm’s own intentions to adopt these technologies in the near term. The hypothesis is that beliefs about competitors’ adoption may influence a firm’s own adoption decisions.
We have already collected and analyzed the baseline survey responses. This pre-registration pertains to the next step: merging and analyzing two additional sources of post-treatment data: (i) a follow-up survey to be collected approximately 12 months after the experiment, and (ii) outcome variables from administrative records, such as firm-level employment from employer-employee datasets.
Intervention
The treatment consists of information about the technological adoption of competitors.
Experimental design
The main hypothesis is that beliefs about competitors’ adoption may influence a firm’s own adoption decisions. An additional hypothesis is that a firm’s adoption of advanced technologies may affect firm-level outcomes such as employment.
A copy of the survey instrument is attached to this form. The baseline survey begins by eliciting prior beliefs about the share of competitors that have already adopted advanced technologies (defined as predictive AI, generative AI, or robotics). Half of the firms were then randomly assigned to receive accurate information about the actual share of competitors who had adopted these technologies. After the information-provision stage, we elicited posterior beliefs about the share of competitors expected to adopt advanced technologies in the future. We also elicited the firm’s own intentions to adopt these technologies in the future.
Following the standard design of information-provision experiments, we leverage treatment heterogeneity in prior beliefs. Specifically, individuals who underestimated the share of adopters are expected to revise their beliefs upward; those with accurate priors should exhibit little or no belief updating; and those who overestimated the share are expected to revise their beliefs downward.
Due to the panel nature of the survey, some of the firms will be surveyed again a year later, in which they will be asked again about their own AI adoption. With those responses we will be able to measure whether the information they received in the baseline survey affected their adoption choices 12 months later. Moreover, we may be able to measure the effects on other outcomes measured from administrative records, such as firm-level employment from employer-employee datasets.
Design details
Cited by
No citation data available for this registration yet.