The use of AI in accounting tasks

How to cite

LOURENCO, SOFIA; Monteiro, Januario; Motoki, Fabio; Gomez-Ruiz, Laura; Guven-Uslu, Pinar; Malagueno, Ricardo. 2025. "The use of AI in accounting tasks." AEA RCT Registry. AEARCTR-0016215. https://doi.org/10.1257/rct.16215-1.1

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

SOFIA LOURENCO* , Januario Monteiro , Fabio Motoki , Laura Gomez-Ruiz , Pinar Guven-Uslu , Ricardo Malagueno

* Primary investigator

Registry ID
AEARCTR-0016215
DOI
https://doi.org/10.1257/rct.16215-1.1
First registered
Last updated
Status
In development
Trial dates
2025-06-27 – 2026-05-31
Countries
Private
JEL codes
M00, M41, M49, M53, O39
Registry record
http://www.socialscienceregistry.org/trials/16215

Abstract

Recent research shows that the use of AI can benefit performance for some tasks and be detrimental for other tasks. Additionally, research also suggests that AI can reduce or conversely increase the performance gap between heterogeneous participants.
This study explores whether and how the use of AI can enhance performance in accounting tasks. To this end we design a framed field experiment with accounting professionals where we vary the absence/presence of AI and the type of task. Our study also explores whether and how the use of AI varies with the characteristics of the participants, e.g. experience.

Intervention

The participants of the experiment will perform accounting tasks online and answer a survey at the end. Randomly assigned participants will be provided with access to an AI platform to support their work in an accounting task, whereas the control group will not.
The control group will only have access to the AI platform after completing the post-experiment survey.

Experimental design

The experiment is a 2x2 where we vary AI (absent vs present) and task type (less vs more complex).
After completing the task, participants answer a survey about their characteristics.
The control group will only have access to the AI platform after completing the post-experiment survey.

Design details

Randomization method
Randomization done in office by a computer
Randomization unit
Individuals
Clusters
There is no clustering so the number of clusters is equal to the number of individuals.
Observations
240 individuals.
Arms
60 individuals.
Primary outcomes
Accuracy – the number of correct answers per question/task Reasoning – the length of the text and the quality of the arguments presented to support their opinion Speed – the number of minutes/seconds spent in each question/task
Secondary outcomes
Use of AI – decision to use or not AI Number of prompts in AI platform Quality of prompts in AI platform Time spent in the AI platform Revision of answers by the control group once they have access to the AI platform

Cited by

No citation data available for this registration yet.

Keywords