Almost nobody leaves a community. They stop arriving, which is a different event and produces no record of itself. There is no cancellation, no goodbye post, no moment you could point to — just a person who used to be there.
This is why most advice about exit surveys does not transfer. It assumes a leaving event, and in a community there usually is not one.
The four ways people actually go
They look the same in a member count and need completely different responses.
| How they left | What you can see | What it tells you |
|---|---|---|
| Never activated | Joined, looked once, never returned | The promise and the first visit do not match |
| Faded | A regular whose posts thinned, then stopped | Something changed — often something you changed |
| Left after an incident | Active, then abruptly absent | Usually a conflict you may not have seen |
| Cancelled | An actual billing event | The only group you can reliably ask |
The second row is where the information is, and it is the one owners investigate least, because nothing happens on the day a regular fades. The first row is the largest by volume and the least informative — most people who join anything never come back to it.
The reason people give is rarely the reason
Ask directly and you will mostly get "no time", which is true and useless.
Nobody has time for anything; they make it for things that repay it. "I got busy" almost always means the community stopped being worth the twenty minutes it cost, and the interesting question is when that changed rather than whether they were busy.
So ask about the last time it was good rather than about why they left. "What were you getting from it when you were most active" produces specifics — a person, a thread, a weekly thing that stopped — where "why did you leave" produces politeness.
Asking someone who faded
This is the highest-value conversation available to you and almost nobody has it.
Send one message, personally, four to six weeks after they went quiet — early enough that they remember, late enough that it does not read as pressure. Reference something specific they posted. Ask one question, not a survey.
Make it genuinely easy to not reply, and easier to be honest. Say you are not trying to win them back, and then do not try to win them back in the same message, because the moment it becomes a retention pitch you get a polite non-answer and lose the information — the mechanics of what to do with it are in reducing community churn.
The people who never started
The largest group of leavers never really arrived, and they are worth a different question.
Someone who joined and never returned cannot tell you much about your community, because they never saw it. What they can tell you is what they expected — and the gap between what your landing page promised and what the first screen delivered is the single most fixable thing on this list.
You will not get replies from many of them. You do not need many: five answers describing the same mismatch is a finding, and the fix usually belongs to onboarding rather than to anything deeper.
When it was an incident
Somebody who was active and then vanished within a day did not get busy.
Check what happened in the space around that date. Often it is a disagreement you did not read, a joke that landed badly, or a moderation decision that felt arbitrary to someone watching. The person who leaves after an incident is rarely the person who was in it — bystanders leave more often than participants, and they never mention it.
These are worth asking about directly and gently, because the answer usually names something still happening — see handling difficult members.
The cancellation is your one clean signal
Paid members produce an actual event, which makes them the only group you can survey reliably.
Keep it to one question with a few concrete options and a text box, asked at the moment of cancelling rather than by email afterwards. Options that describe situations — "I stopped using it", "it got too expensive for what I use", "I got what I came for" — beat abstract satisfaction scales, because people recognise themselves in situations.
"I got what I came for" is a good outcome wearing the costume of a bad one. A community someone joined to solve a specific problem has succeeded when they leave solved, and counting that as failure pushes you toward manufacturing dependence — the pricing angle is in how to price a community membership.
What not to do
Three responses are common and all of them cost more than they return.
The win-back discount. It buys back the people least attached to the thing and teaches everyone that leaving is a negotiating position.
The guilt message. "We miss you" asks someone to manage your feelings about their departure, and they will simply not reply.
The long survey. Anyone willing to complete twelve questions about a community they left is unrepresentative by definition, and you will build a plan on their answers.
Turning answers into something
The findings are usually unglamorous, and that is a sign they are real.
People leave because the weekly thing stopped, because the one person they talked to left, because the space got too busy to follow, or because they never found anyone like them. Each has an obvious fix, and the reason it went unfixed is that nobody had said it out loud.
Fix one thing visibly and tell the community you did, naming the reason. That does more than the fix itself, because it demonstrates that leaving quietly is not the only option — and if the answers keep pointing at the same month, check whether it was a season rather than a decline, per why communities go quiet.
The bottom line
Accept that most leaving is silent, and stop waiting for an event that will not come.
Message a few people who faded, ask what it was like when it was good, survey only at the cancellation moment, and treat "I got what I came for" as a success. Then watch whether the same reason appears three times, because one leaver is an anecdote and three is a fault in the room — which is the difference between listening and collecting, and the measurement side is in the metrics that actually matter.