Improving matching of clients and workers on an online platform

How to cite

McKenzie, David; Piza, Caio. 2025. "Improving matching of clients and workers on an online platform." AEA RCT Registry. AEARCTR-0016380. https://doi.org/10.1257/rct.16380-1.0

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

David McKenzie* , Caio Piza

* Primary investigator

Registry ID
AEARCTR-0016380
DOI
https://doi.org/10.1257/rct.16380-1.0
First registered
Last updated
Status
On going
Trial dates
2025-07-11 – 2026-06-30
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/16380

Abstract

We work with an online jobs platform in Mozambique that aims to match casual/informal workers with clients. The experiment tests whether administering a short survey of preferences, asking clients for their preferences on gender of workers, and using AI recommender systems improve matching compared to the status quo algorithm.

Intervention

Clients coming to the platform to look for workers for odd-jobs are randomly assigned to be matched to workers under different sets of matching conditions.

Experimental design

Clients are assigned to one of the following five groups:
1) Status quo matching
2) The legacy algorithm, incorporating survey preferences
3) The legacy algorithm, incorporating survey preferences + gender preferences
4) AI matching
5) AI matching + gender preferences

Design details

Randomization method
Random assignment by computer at the client level
Randomization unit
Client
Clusters
Sample size uncertain - clients will be randomized as they search for a client, and then all subsequent searches by the client will be under the same matching approach - so search results will be clustered at the client level. We anticipate randomizing clients on a rolling basis for 3-4 months, with approximately 1000-2000 clients per month.
Observations
Uncertain, but several thousand clients
Arms
Uncertain, since budget limits restrict the proportion the partner can allocate to AI.
Primary outcomes
We use platform data to measure four key outcomes: 1) Share of searches that result in client requesting worker information 2) Share of searches that result in client contacting worker 3) Share of searches that result in a work contract occurring 4) Share of work contracts going to women

Cited by

No citation data available for this registration yet.

Keywords