Testing an LLM-Based Intervention Tool in Uganda

Last registered on June 15, 2026
View Trial History

Pre-Trial

Trial Information

General Information

Title
Testing an LLM-Based Intervention Tool in Uganda
RCT ID
AEARCTR-0018800
Initial registration date
June 03, 2026

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
June 15, 2026, 5:43 AM EDT

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

Locations

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

Request Information

Primary Investigator

Name
Frank Germann
Affiliation
University of Notre Dame
Contact Primary Investigator

Other Primary Investigator(s)

PI Name
Stephen Anderson
PI Affiliation
Contact Investigator
PI Name
Arthur Ssembajja
PI Affiliation
PI Name
Dominik Papies
PI Affiliation

Additional Trial Information

Status
On going
Start date
2025-10-01
End date
2027-12-31
Keywords
Firms & Productivity
Additional Keywords
JEL code(s)
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.
Abstract
This research examines whether large language models (LLMs) can provide scalable, AI-driven mentorship to entrepreneurs in emerging markets, with a focus on Uganda. Entrepreneurship is critical for economic development and poverty alleviation, yet many entrepreneurs in emerging markets lack access to high-quality mentorship, training, and market-oriented guidance. We therefore investigate whether LLMs can serve as effective surrogate mentors that help entrepreneurs improve their market understanding and stimulate demand.

We study this question through a randomized controlled trial with entrepreneurs in Kampala, Uganda. Following a baseline survey, participants were randomly assigned to two equal-sized groups: a treatment group and a control group. Entrepreneurs in the treatment group receive access to a customized LLM designed to support market perception, customer understanding, and demand generation. We measure changes in business practices and performance through multiple post-intervention survey rounds.
External Link(s)

Registration Citation

Citation
Anderson, Stephen et al. 2026. "Testing an LLM-Based Intervention Tool in Uganda." AEA RCT Registry. June 15. https://doi.org/10.1257/rct.18800-1.0
Sponsors & Partners
Experimental Details

Interventions

Intervention(s)
Intervention Start Date
2025-11-15
Intervention End Date
2026-11-15

Primary Outcomes

Primary Outcomes (end points)
The key outcome variables are changes in entrepreneurs' business practices and firm performance. Business-practice outcomes include (e.g.) finance and accounting, operations and human resources, and marketing and branding practices. Firm-performance outcomes include (e.g.) sales and profits. We will measure these variables through multiple post-intervention survey rounds to assess whether access to the customized LLM changes how entrepreneurs manage their businesses and whether these changes translate into improved performance.
Primary Outcomes (explanation)
We will also examine standardized indices for key outcome domains.

Secondary Outcomes

Secondary Outcomes (end points)
For mechanism evidence, we will measure outcomes related to improving market perceptions and stimulating demand (e.g., communication materials, product-service design, and brand strength).
Secondary Outcomes (explanation)
For additional process evidence, we will also examine administrative data from the LLM platform (e.g., LLM usage intensity and heterogeneity).

Experimental Design

Experimental Design
We are conducting a randomized controlled trial with entrepreneurs in Kampala, Uganda. After completing a baseline survey, participants were randomly assigned to a treatment group or a control group of equal size. The treatment group receives access to a customized LLM intended to support market-oriented business decision-making. Outcomes will be measured through follow-up survey rounds and compared across the treatment and control groups.
Experimental Design Details
Not available
Randomization Method
Randomization was conducted by computer at the individual entrepreneur level, stratified by entrepreneur gender, business sector, and a pre-specified baseline index.
Randomization Unit
The unit of randomization is the individual entrepreneur. Entrepreneurs were randomly assigned to either the treatment or control group at the individual level. Randomization was stratified by entrepreneur gender, business sector, and a pre-specified baseline index, but there was no cluster-level randomization.
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
There are no treatment-assignment clusters. The planned number of clusters is therefore 0. Randomization was conducted at the individual entrepreneur level, stratified by entrepreneur gender, business sector, and a pre-specified index constructed from baseline variables such as prior AI usage and previous exposure to mentorship programs.
Sample size: planned number of observations
332 entrepreneurs running small businesses in greater Kampala, Uganda. All included entrepreneurs completed the baseline survey and were randomly assigned to the treatment or control group at the individual level.
Sample size (or number of clusters) by treatment arms
Treatment group: 166 entrepreneurs. Control group: 166 entrepreneurs. Total sample: 332 entrepreneurs.
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Supporting Documents and Materials
IRB

Institutional Review Boards (IRBs)

IRB Name
University of Notre Dame Institutional Review Board
IRB Approval Date
2025-09-11
IRB Approval Number
25-08-9524
Analysis Plan