The Effect of Automation on Firm-Based Training

Last registered on March 06, 2024


Trial Information

General Information

The Effect of Automation on Firm-Based Training
Initial registration date
February 26, 2024

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
March 06, 2024, 3:28 PM EST

First published corresponds to when the trial was first made public on the Registry after being reviewed.


There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Primary Investigator

University of Bern

Other Primary Investigator(s)

PI Affiliation
University of Bern
PI Affiliation
University of Bern

Additional Trial Information

In development
Start date
End date
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
We investigate the effects of automation on the behavior of firms that provide firm-based training. We survey companies in Switzerland that provide firm-based training – namely apprenticeships – and present them hypothetical scenarios about future work task automation. We analyze how this affects their decision regarding the number of apprenticeships they offer. Specifically, we hypothesize that the shorter time until task automation happens and the higher the proportion of tasks that are automated, the larger the reduction in apprenticeship-positions offered. Crucial to our analysis is the heterogeneity of in the firms' reactions. For this, we divide firms according to profit and investment motif with regards to offering apprenticeships, their exposure to recent innovations in AI versus previous innovations, and their size; and we divide the apprenticeship occupations into cognitive or manual as well as having high language or math requirements.
External Link(s)

Registration Citation

Gschwendet, Christian, Claudio Schilter and Stefan Wolter. 2024. "The Effect of Automation on Firm-Based Training." AEA RCT Registry. March 06.
Experimental Details


Intervention Start Date
Intervention End Date

Primary Outcomes

Primary Outcomes (end points)
The primary outcome is the reduction in percent of apprenticeship positions offered by the firms in the hypothetical scenarios. We have the following primary hypotheses about this outcome:

1) The shorter time until task automation happens and the higher the proportion of tasks that are automated, the larger the reduction in apprenticeship-positions offered.

2) Firms that have a profit motif will react more than firms that have an investment motif (higher elasticity).

3) Firms that offer occupations with a higher exposure to recent innovations in AI technologies, particularly generative AI, will react more (higher elasticity), while firms that offer occupations with a higher exposure to previous innovations in digital technology will react less (lower elasticity).

4) Firms that offer occupations that include primarily cognitive (as opposed to manual) tasks and have high language (as opposed to math) requirements will react more (higher elasticity).
Primary Outcomes (explanation)
Firms with an investment motif are firms that lose money on training apprentices during the apprenticeship itself (but do it for the chance to hire trained workers). Firms with a profit motif are firms that already make a profit during the apprenticeship from offering the apprenticeship. We will use firm characteristics to classify firms. For example, smaller firms are more likely to operate under the investment motif.

The idea of displacement risk for jobs exposed to digital technology is not new and for some occupations, experts have stressed automation prospects for a long time now (see, e.g., Arntz et al, 2016; Frey & Osborne, 2017). However, ChatGTP had a profound impact on how automation risk is perceived (Goller, Gschwent & Wolter, 2023) and evaluations of exposure and potential automation risk post-ChatGTP (e.g., Eloundou et al., 2023; Felten et al., 2023; Hui, Reshef & Zhou, 2023) come to different conclusions than earlier work such as Frey & Osborne (2017). We hypothesize that firms that offer training for newly affected occupations have had less time to adapt to a situation with relevant exposure to new AI technologies and are therefore more responsive to the different scenarios. Moreover, Goller, Gschwendt & Wolter (2023) find on the supply side of apprenticeship-labor that the introduction of ChatGTP makes prospective apprentices search less for apprenticeships that focus on cognitive tasks and have high language requirements. We therefore hypothesize to find a matching heterogeneity on the demand side – again because firms have not had much time to adapt to the type of exposure caused by ChatGPT and similar AI technologies.

Arntz, M., Gregory, T., & Zierahn, U. (2016). The risk of automation for jobs in OECD countries: A comparative analysis.
Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2023). GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models. arXiv:2303.10130v4
Felten, E. W., Raj, M., & Seamans, R. (2023). Occupational heterogeneity in exposure to generative ai. Available at SSRN 4414065.
Frey, C. B., & Osborne, M. A. (2017). The future of employment: How susceptible are jobs to computerisation? Technological Forecasting and Social Change, 114, 254-280.
Goller, D., Gschwendt, C. & Wolter, S. C. (2023). “This Time It’s Different” – Generative Artificial Intelligence and Occupational Choice. IZA Discussion Paper Series, No. 16638
Hui, X., Reshef, O., & Zhou, L. (2023). The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market. CESifo Working Paper, No. 10601.

Secondary Outcomes

Secondary Outcomes (end points)
The outcome always stays the same, but we will further investigate the role of firm and location characteristics to understand the heterogeneity in the treatment effect as good as possible. In addition, we will conduct all primary estimation also by firm size (interactions) as we expect different effects for small and large firms (with small firms being more affected by the fact that the number of hired people has to be an integer – and practically often even 1 or 0).
Secondary Outcomes (explanation)

Experimental Design

Experimental Design
We survey companies that provide firm-based training – namely apprenticeships. We present them hypothetical scenarios about future work task automation and ask how this would affect their current decision about the number of apprentices they will start training this year. Each company is asked about three hypothetical scenarios, in which they receive the result of an hypothetical internal analysis. In each scenario, the analysis concludes that nothing will happen for X years, but then Y% of the tasks of a specialist in the occupation for which the firm provides training will be automated. X can take the values 2,4, and 6, and Y can take the values 20, 40, and 60, resulting in 9 possible scenarios. The three scenarios presented to each firm are randomly selected.
Experimental Design Details
Not available
Randomization Method
Randomization is done by computer using Stata.
Randomization Unit
Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters
3300 firms
Sample size: planned number of observations
Sample size (or number of clusters) by treatment arms
1100 observations for each scenario
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)

Institutional Review Boards (IRBs)

IRB Name
IRB Approval Date
IRB Approval Number