Can AI Help Remote Workers? Experimental Evidence from a Chat Support Tool

How to cite

Ozguzel, Cem; Aksoy, Cevat Giray. 2025. "Can AI Help Remote Workers? Experimental Evidence from a Chat Support Tool." AEA RCT Registry. AEARCTR-0016421. https://doi.org/10.1257/rct.16421-1.0

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

Cem Ozguzel* , Cevat Giray Aksoy

* Primary investigator

Registry ID
AEARCTR-0016421
DOI
https://doi.org/10.1257/rct.16421-1.0
First registered
Last updated
Status
In development
Trial dates
2025-07-31 – 2026-05-01
Countries
Private
JEL codes
M12, M54, J24, L84
Registry record
http://www.socialscienceregistry.org/trials/16421

Abstract

This randomised controlled trial evaluates the impact of an AI-driven chat assistance tool on the performance of remote call center agents. The tool provides real-time answers to job-related questions, helping agents address queries more effectively. We randomly assign 500 agents into two groups: 250 receive access to the chat tool, and 250 continue without it. The trial will run for three months and assess differences in key performance indicators and employee perceptions between groups.

Intervention

Chat Assistance Tool Access: Agents in the treatment group will receive access to an AI-powered chat assistance tool embedded in their workflow. The tool provides instant, task-relevant information to support query resolution. Agents in the control group will work under existing conditions without access to the tool. Performance will be tracked through administrative data, and surveys will assess perceptions and satisfaction.

Experimental design

A randomised controlled trial with individual-level randomisation among 500 call center agents. Half will be assigned to the treatment group with access to the chat tool, the other half to the control group.

Design details

Randomization method
Based on the odd/even number of agents’ employee ID numbers.
Randomization unit
Individual agent
Clusters
500
Observations
500
Arms
250 controls, 250 treatment
Primary outcomes
Production Outcomes - Number of calls handled per hour - Talk time per call (in seconds) - Call duration per call (in seconds) Service Quality Outcomes - Customer rating (from post-call surveys) - Random audit rating (based on internal quality assessments)
Secondary outcomes
Survey-based Measures (from baseline and endline surveys administered to all agents): - Self-reported stress levels and mental health - Overall job satisfaction and happiness - Frequency and perceived usefulness of AI tool usage at work - Use of AI tools outside work (e.g. for personal tasks, information, or support)

Cited by

No citation data available for this registration yet.

Keywords