Research Community on WhatsApp: The Migration Tax
A research community on WhatsApp keeps people where they already are. What migrating an MROC costs, what the field literature found, and when a platform wins...
A qualitative team recruits twenty people for a three week study. Recruitment runs on WhatsApp, because that is where a screener reply comes back inside the hour. Then, on day one, the same twenty people are told to install a platform, open an account and learn a new screen. A research community on WhatsApp closes that gap by leaving the study where the participants already are.
Key Takeaways
- A market research online community is a recruited group that answers moderator activities over days or weeks, at one or two activities per week.
- WhatsApp passed 3 billion monthly users in 2025. It is the one app a participant does not have to install to take part.
- Peer reviewed work ran 15 focus groups inside WhatsApp in Malawi. It named the limits plainly: power, network and who the participant is.
- Fewer than 10 percent of members post without being asked. Reminding and chasing decide whether the study has material at all.
What a market research online community actually is
A market research online community is a recruited group of people who answer activities posted by a moderator over days or weeks. The same people can be probed over time. The field shortens the name to MROC.
A community manager is the person who keeps that group alive: posting activities on schedule, replying to what comes in, spotting who has stopped answering, and bringing in new members.
The cadence is modest on purpose. Recollective's guide to online research communities puts a long running community at one or two activities per week. It is candid about the curve. Fewer than 10 percent of members create anything on their own. Everything else is prompted. That one figure explains most of what a community platform gets paid to do.
How a research community is run today
A community runs in six steps.
- Screen and recruit through the channel where people reply fastest.
- Move them onto the community platform: an install, an account, a tour of the interface.
- Post the first activity and see who shows up.
- Chase the ones who did not, one by one, often from a different app than the study lives in.
- Refresh the membership as engagement falls, swapping in new people to hold the quota.
- Export the material, code it, and write it into the report.
Steps four and five are where the hours go, and they are the two nobody puts on the proposal. Recollective is explicit that a long community has to be refreshed. Attrition and fatigue are the normal case, so step five never ends. Step six is [the problem of coding without losing the evidence chain](/blog/qualitative-coding-evidence-chain).
Most teams still track steps three to five in a spreadsheet beside the platform. The platform reports who logged in. The researcher needs to know who answered.
The migration tax nobody puts on the timeline
Step two is paid for in engagement. Engagement is the raw material of the method.
Every install, every account and every new screen is a point where a recruited person quietly stops. Nobody books those losses, because they happen before the study opens and look like ordinary recruitment shrinkage.
Meta confirmed more than 3 billion monthly users on WhatsApp on its first quarter 2025 earnings call. In most consumer markets the app is already on the phone, notifications on, the habit of answering formed. It is the only app that asks nothing of the person before the study begins.
Two paths to the first activity
Both paths start from the same point, a participant recruited inside WhatsApp. The upper path leads to a dedicated community platform and passes through three gates before any research happens: install the app, create an account, learn the interface. Each of those three is marked as a drop-off point, where a recruited participant can quietly stop. Only after all three does the participant reach the first activity. The lower path goes straight from recruitment to the first activity, with no install, no account and no new screen to learn, because the app is already on the phone. The difference between the two paths is the migration tax, and it is paid in engagement rather than in hours on the timeline.
The three gates on the upper path are paid for in engagement. Engagement is the raw material of the method.
What the field literature says about running groups inside a messaging app
This has been run, written up and peer reviewed.
A team in Malawi ran fifteen focus groups with teenagers entirely inside WhatsApp. They published the research method in full: recruitment, moderation and the ethics review. The write up in Gates Open Research is worth reading for its limitations section alone.
A mobile instant messaging interview is an interview held inside a messaging app, typed rather than spoken, spread over hours or days instead of one sitting. Kaufmann and Peil named it MIMI in 2020. The wider map of internet interview methods sits in this chapter from Springer. A WhatsApp focus group is the group version of the same tool. A survey sent through the same app is a different thing. It gets [tested for path coverage before field](/blog/survey-testing-before-fieldwork) like any programmed questionnaire.
Where the method still costs you something
UNIQUE INSIGHT Three limits are real, and better software solves none of them.
The first is infrastructure. An activity does not happen without a charged phone and a working network. Neither is a given on every day of a 21 day study. Plan the cadence around that, or the data comes back with holes shaped like power cuts.
The second is fit. Typed chat favors people who type well and check their phone often. That covers most consumer segments, and not all of them. Consider a study among low literacy readers, or people who use a phone as a phone. That needs a different room.
The third is the material. A community produces thousands of messages, voice notes, photos and reactions, and none of it arrives sorted. Export and sorting are a work stream of their own. A team that treats them as a footnote finds that out in the last week.
Who answers when engagement drops
PERSONAL EXPERIENCE The pattern repeats on the community studies we have run.
It is Thursday. The Tuesday activity has 14 replies out of 20. Nobody has noticed yet. Noticing means opening the platform, exporting the participation list and comparing it against the recruitment sheet. That happens when the researcher has an hour, usually on Friday afternoon.
By then the six quiet people have had three days to drift. Two of them will not come back. Each replacement has to be found, screened and briefed. The new member starts on activity four with no memory of the first three. The moderator absorbs all of it, on top of the reading and probing that is the actual job.
Where Cassi.ai comes in
Cassi.ai is a software engineering company specialized in the pains of market research, innovation and insights, working with research teams and agencies. Cassi.ai builds the community inside WhatsApp itself, with a moderator agent that runs the mechanics of the group, asks the scheduled questions, probes when an answer asks for it, and flags who has gone quiet, while the human moderator watches, joins whenever she wants, and keeps every call that needs judgment.
The format does not suit every audience, and it does not make raw material sort itself. What it removes is the migration and the manual attendance check, the two costs nobody writes down. The wider set of problems we automate is in the Cassi.ai portfolio.
What the moderator agent does, and what it never decides
UNIQUE INSIGHT Writing the split down before the study opens is what keeps the method defensible.
A moderator agent is a configured assistant that posts the scheduled activities into the group. It answers procedural questions, asks a shallow follow up when a reply is thin, and reports who has not responded.
- Post each activity at the agreed hour, without a person having to remember.
- Send the reminder to the three who have not answered, at the agreed threshold.
- Ask the shallow probe: what makes you say that, can you show it.
- Report attendance per participant per activity, so the refresh decision lands on Tuesday, not Friday.
What it never decides: which answer becomes a theme, which quote carries the finding, and what the client should do about it. That stays with the moderator. It is the same boundary [an enterprise AI platform has to hold for every other research task](/blog/why-chatgpt-is-not-a-research-platform).
Consent, when the room is an app people already live in
Familiarity raises the bar for consent instead of lowering it.
Someone joining a study group in the same app that holds their family thread has to be told, in plain words, what is collected, who reads it, how long it is kept and how to leave. The ICC/ESOMAR Code of Conduct, in its 2025 revision, puts the weight on accountability, transparency and human oversight.
Two consequences follow. Phone numbers are personal data. A group shows them to every other member, so the consent form has to say so. A moderator agent has to be disclosed as one. A participant deserves to know who wrote the probe.
Dedicated platform compared with a community inside WhatsApp
| Criterion | Dedicated community platform | Community inside WhatsApp |
|---|---|---|
| Install required | Yes, before day one | No, the app is already on the phone |
| Interface to learn | A new screen and new navigation | The one the participant uses daily |
| Where the history lives | On the vendor's servers, exportable | In the group, exported by the operator |
| Who tracks attendance | A person, against a participation report | The agent, per activity, continuously |
| Getting the material out | Built in export, in the vendor's format | Has to be built, and it has to be audited |
| When it fits | Panels of thousands, brand owned communities, end to end audit trails | Studies of 15 to 60 people where reach and reply speed decide |
Cadence, refresh and the participation figures come from Recollective's guide to online research communities. The operating comparison is our own reading.
When a dedicated platform is still the right call
Three cases go to the platform, and none is close.
A community of thousands of members is the first. Threaded discussion, segment level reporting and structured task types, such as card sorts and photo diaries, stop being extras at that size. They become the only way to run it. A brand owned panel meant to last for years is the second. The membership record becomes the asset. The third is any study whose contract demands an auditable environment end to end.
The Greenbook directory of online community suppliers lists 66 firms in that category alone. The open question is whether your study needs one. The same question turns up at the screen where results land, in [building the dashboard from the decision](/blog/market-research-dashboard-decision-first).
FAQ
What is a market research online community?
A recruited group of people who answer activities posted by a moderator over days or weeks. The same people can be probed more than once. Recollective describes a working cadence of one or two activities per week for a long running community.
Can you run an MROC on WhatsApp?
Yes. A team in Malawi ran 15 focus groups entirely inside WhatsApp and published the method in Gates Open Research. The long running version adds a cadence, an attendance check and a refresh rule.
Does a moderator agent replace the human moderator?
No. UNIQUE INSIGHT The agent runs the mechanics: posting on schedule, reminding, shallow probing and attendance. Every judgment call stays human, including what becomes a theme and which quote carries a finding.
How often should a research community post activities?
One or two activities per week is the cadence Recollective reports for long running communities. Push past that and fatigue arrives early. Fall below it and the group goes cold.
How do you handle consent when the study runs inside a personal messaging app?
Explicitly, and in writing, before the first activity. The ICC/ESOMAR Code of Conduct puts its emphasis on transparency and human oversight. That means telling participants what is collected, who reads it, how to leave, that other members see their number, and that a moderator agent is involved.
Everything the method produces depends on twenty recruited people continuing to answer. Anything standing between them and the first activity is a cost, whether or not it shows up on the timeline.
If your last community study lost people between recruitment and activity one, pull two numbers: how many were recruited, and how many posted in week one. That ratio is the migration tax. It is measurable this afternoon.
Published by Cassi.ai. Read the full article at https://www.cassiai.com/blog/research-community-on-whatsapp.