Effective Referrals

How to cite

Bapna, Sofia; Bapna, Ravi; Burtch, Gordon. 2025. "Effective Referrals." AEA RCT Registry. AEARCTR-0017075. https://doi.org/10.1257/rct.17075-1.0

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

Sofia Bapna* , Ravi Bapna , Gordon Burtch

* Primary investigator

Registry ID
AEARCTR-0017075
DOI
https://doi.org/10.1257/rct.17075-1.0
First registered
Last updated
Status
In development
Trial dates
2025-10-23 – 2028-04-22
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/17075

Abstract

After using a product or service or making a donation, individuals are often given the option of making referrals. Peer-referral programs leverage existing customers' social networks to reach potential new customers and are commonly used in practice.

Such programs have been shown to be very effective in many settings [Bapna and Umyarov (2015); Jung, Bapna, Golden and Sun (2020)]. However, the underlying mechanisms by which peer referrals operate remain elusive. In particular, research has yet to address whether and when peer referrals are effective because referring individuals bring to bear private information about the ideal candidates for recruitment, whether referring individuals exert influence over their social connections, or some combination of the two.

The objective of this study is to explore the mechanisms behind referral efficacy.

By providing an understanding of the underlying mechanism for peer referral program efficacy, this study will have immediate implications for organizations that use such programs. Specifically, organizations will be able to maximize conversion rates in peer referral programs.

Intervention

Individuals (contributors) employ an Android app (referred to as The App) to contribute short recordings of their voices, which will be part of a public (open-source) speech dataset for Indian languages.

After they have contributed their voice recordings, users of The App will be asked if they would like to refer any of their social connections (i.e., if the contributor would like to identify other individuals who they believe may also be interested in contributing). To solicit referrals, the contributor will be shown the text, “Share our app with your friends. Click the link below to help recruit more contributors. Help us create AI tools (like ChatGPT) that can converse in Indian languages!” If the user agrees to provide referrals (i.e. to help recruit contributors), they will be randomly assigned to one of three conditions:

T1, Direct Solicitation (Targeting and Influence) - The user will contact their peers directly, to refer them for participation. This group’s performance will reflect a combination of private information that an existing member has about the ideal candidates for referral, as well as the influence or persuasion they exert over their connections.
The text shown to the referee (person receiving the referral) is:
"[First and last name, phone number] is inviting you to help [organization name] build better speech recognition for Indian languages!
If you have expertise in Robotics, Mathematics, Machine Learning, Business, or related fields, you can help—just record your voice using the [app name] to contribute to a public dataset of speech recordings.
Sign up here: [Signup Link]
Your voice can power public speech technology for India!"

T2, Indirect Solicitation (Targeting only) - The user will be asked to identify peers but will not directly solicit their participation. Instead, the referrals will be contacted and solicited directly by the organization collecting the data (the referrer will not be mentioned). This group’s performance will solely reflect the private information that an existing member has about ideal candidates for referral. As solicitation will be undertaken by the organization collecting the data, peer influence will play no role.
The text shown to the referee is:
“Help [organization name] build better speech recognition for Indian languages!
If you have expertise in Robotics, Mathematics, Machine Learning, Business, or related fields, you can help—just record your voice using the [app name] to contribute to a public dataset of speech recordings.
Sign up here: [Signup Link]
Your voice can power public speech technology for India!"

C, No solicitation (Control) - The user will be asked to identify peers, but will not directly solicit their participation. The referrals will be contacted and solicited directly by the organization collecting the data after a delay of one week with the name of the original referrer as the person sending the referral. This group’s performance in the first week will reflect whether the individuals referred would have participated anyway (i.e. without the referral).
The text shown to the referee is the same message as T1, after a delay of 1 week.

On clicking the signup link, referees have the option to fill out an application form (Google form) to contribute. If their application is approved, they will be able to contribute voice recordings.

Experimental design

After they have used an app to provide data for an open-source repository, users of the app (contributors) will be asked if they would like to refer any of their social connections (i.e., identify other individuals who the contributor believes may also be interested in contributing). If the individual agrees to provide referrals, they will be randomly assigned to one of three conditions:

T1, Direct Solicitation (Targeting and Influence) - The user will contact their peers directly, to refer them for participation. This group’s performance will reflect a combination of private information that an existing member has about the ideal candidates for referral, as well as the influence or persuasion they exert over their connections.

T2, Indirect Solicitation (Targeting only) - The user will be asked to identify peers, but will not directly solicit their participation. Instead, the referrals will be contacted and solicited directly by the organization collecting the data (the referrer will not be mentioned). This group’s performance will solely reflect the private information that an existing member has about ideal candidates for referral. As solicitation will be undertaken by the organization collecting the data, peer influence will play no role.

C, No solicitation (Control) - The user will be asked to identify peers, but will not directly solicit their participation. The referrals will be contacted and solicited directly by the organization collecting the data (same text as T2) after a delay of one week with the name of the original referrer as the person sending the referral. This group’s performance in the first week will reflect whether the individuals referred would have participated anyway (i.e. without the referral).

Referees contacted have the option to fill out an application form (Google form) to contribute. If their application is approved, they will be able to contribute voice recordings.

The main dependent variables (DVs) include whether referees submit the application form, and whether the application is approved. Additional DVs include: whether the referees created an account on the app, whether they contributed, and the quality of their contribution.

Design details

Randomization method
By a computer. App users (contributors) providing referrals in the first 20 secs of a minute are assigned to T1, next 20 secs to T2, last 20 secs to C.
Randomization unit
Randomization into one of the three experimental conditions, is at the level of the person making the referral.
Clusters
No clusters
Observations
It is difficult to predict the number of people who will use the app and the number of users who will refer others. We will keep the experiment live for 18 months.
Arms
We expect that each arm will have a similar number of individuals (see the randomization method described above).
Primary outcomes
The outcomes (dependent variables) are: i) Whether the referred social connection decides to fill up the application form to contribute (binary 0/1) ii) Whether the referred social connection’s application is approved (by the app’s creators) or not (binary 0/1) H1A: A direct solicitation by the referrer (i.e., the solicitation is sent by the referrer and thus mentions the name of the referrer) is more likely to result in an application, relative to not having a referral. H1B: A direct solicitation by the referrer is more likely to result in an application that is approved, than not having a referral. H2A: An indirect solicitation by the referrer (i.e., the solicitation is sent by the company and does not mention the name of the referrer) is more likely to result in an application, relative to not having a referral. H2B: An indirect solicitation by the referrer is more likely to result in result in an application that is approved, relative to not having a referral. H3A: A direct solicitation by the referrer is more likely to result in an application, relative to an indirect solicitation. H3B: A direct solicitation by the referrer is more likely to result in an application that is approved, relative to an indirect solicitation. Note: not having a referral is the same as control condition (no solicitation) in the week before the solicitation is sent to the referees.
Secondary outcomes
i) Whether the referred social connection created an account on the app (binary - 0/1) ii) Whether the referred social connection contributed via the app (binary - 0/1) iii) The quality of contributions supplied by the referred social connection

Cited by

No citation data available for this registration yet.

Keywords