Community member churn: everyone leaves eventually

Ten people are holding your community up right now. You know exactly who they are. They answer questions before you do, they defend the project when it wobbles, they keep the room warm on a slow Tuesday.

None of them will be there in a year.

That's not a prediction about your specific community. It's the default state of every community, including the ones I run. The mistake isn't losing those ten people. The mistake is building as though you won't.

What community member churn actually is

Community member churn is the rate at which members stop participating. Most advice treats it as a leak to be plugged, something you fix with exit surveys and win-back campaigns. That framing works for a paid membership where churn means a cancelled subscription.

It's the wrong frame for the people who actually run your room.

Leaving is the normal case, and there's a number on it. In 2025, a field experiment published in MIS Quarterly measured what happens to newcomers in an online community and found that the probability of a newcomer returning within 12 months is 38%. Roughly three in five people who post once never post again. The researchers then tried to fix it with a warmth intervention and moved that number by 3.7 percentage points, which is about what being nicer buys you.

Your core members aren't subscribers. They're free people doing unpaid work because they like you, they like the project, or they like the vibes. That's a different animal from the crowd an incentive brings in, which empties out the week the rewards stop. Each one has a runtime in your community: a length of stay that the events and roles and reasons-to-return you build can extend, but never make infinite. Churn isn't the leak. Churn is the clock.

The dependency only runs one way

Once a few members get active and stay active, it's natural to fold them into the operation. They start answering the questions you'd have answered. They calm the chat down before you've seen the message. Quietly, they become part of your team.

The moment they do, you start depending on them.

Here's the part that gets forgotten. The dependency runs one way. You lean on them for activity, for answers, for social proof, for the feeling that the room is alive. They lean on you for nothing. You aren't paying them. They can walk tomorrow and lose nothing at all.

The scale of that unpaid dependency has been measured. In 2022, researchers at Northwestern and collaborators estimated that Reddit's roughly 21,500 volunteer moderators were putting in 466 hours of work a day, worth at least 3.4 million dollars a year, which came to about 2.8% of Reddit's 2019 revenue. The same team found something more uncomfortable in the moderation logs: the visible work is a median of just 23% of the total. Three quarters of what your volunteers do for you, you never see, which means you also have no idea what you'd be replacing.

Most community managers privately think the audience only wants something from the project. The rewards, the money, the time. That's true, and it's only half of it. You need them just as badly, for the messages that make the server look busy and the answers that save you the eighteen-hour days the job otherwise costs. It's a mutual arrangement between free parties, and neither side owes the other permanence.

If you never internalize that, every departure feels like betrayal. It isn't. Nobody signed anything. You built a dependency on people who never agreed to be depended on. If you're earlier than that and still deciding whether to take this work at all, start with the five things to know before you become a community manager.

What churn looked like in a community I run

I run a creator program called Inner Circle, a four-week cohort under the Syndicate brand. Two seasons are finished, and the platform behind it logs every submission, so I can tell you what churn looks like with real numbers instead of a feeling.

Season one took 657 applications and accepted 79 people. Of those 79, 57 ever submitted anything. Twenty-seven closed eight or more core tasks. So the room that looked like 79 members was, in operating terms, 27 people.

Then watch what those 27 did to the weekly numbers. Active members by week ran 56, then 47, then 40, then 34. A 39% decline in people over four weeks, which on a dashboard looks like a community falling apart.

Output went the other way. Submissions by week ran 893, 799, 856, and then 1,033 in the final week, the highest of the season, at a 95.1% approval rate. The room lost 22 of its active people and produced its best week.

The people who left were mostly people who'd never really started. At the end of the season I removed the role from 13 members who had zero submissions and had been silent for two weeks or more. Nobody load-bearing left mid-season. The tail fell off and the center held, which is what a healthy version of this looks like.

Your core is smaller than the leaderboard suggests

The most quoted number in community work is Jakob Nielsen's 90-9-1 rule, published by Nielsen Norman Group in 2006: in most online communities, 90% of people lurk, 9% contribute occasionally, and 1% produce almost everything. For blogs he put it closer to 95-5-0.1.

It holds up badly for anyone assuming their own room is the exception. The same 2022 study of Reddit moderation logs measured how unevenly the work sits. Across 36 subreddits the Gini coefficient on moderator actions ran from 0.47 to 0.90, with a median of 0.74, and in one community the single most active moderator performed 72% of all the moderation work. That subreddit was one person away from having no moderation at all.

I checked our own distribution against it, and ours is much flatter. Season two's final leaderboard had 75 ranked members and 11,039 points between them. The top member accounted for 2.1% of all points. The top 10 accounted for 20.9%. The top half accounted for 74.2%. Median score was 182 against a top score of 236.

That flatness is designed, not lucky. Contribution in the program is capped by task, so no single person can run away with the room, and there's a floor of structured work anyone can complete. Structure compresses participation inequality.

It also means the leaderboard hides the real risk. A flat leaderboard doesn't tell you which four people would take the atmosphere with them if they left, because atmosphere doesn't submit tasks. In season two, the members I'd have missed most were not the top scorers. They were the ones answering newcomers in the chat at midnight, and no dashboard I own counts that.

Why members leave, and why it's rarely you

Sometimes there's no reason you'll ever find. They got busy. They rotated to another chain. They found a room they like more.

The offline volunteering world has been tracking this for longer than we have. A 2007 research brief from the Corporation for National and Community Service found a year-to-year volunteer retention rate of 69%, meaning about 31% did not come back, and that organizations replaced only 83.2% of the ones they lost. It studied Baby Boomers specifically, though turnover for other age groups came out similar at 29%.

The useful part is the dose response in the same report. Retention was 79% among people volunteering 12 or more weeks a year, against 53% for those giving two weeks or less. Light involvement is where churn lives. That's the tail, not the core, and it matches what my own weekly numbers do.

Sometimes it's a chain reaction you didn't start and can't stop. Two members get into it. One takes a break from the chat. A third who liked that person cools off and drifts. Three people gone over a conflict that had nothing to do with you.

And sometimes they burn bright and vanish. In season two, four of the sharpest new members went quiet inside two weeks. One had logged 21 submissions before disappearing, another 12. They weren't unhappy as far as I could tell. They spent everything they had in week one and had nothing left for week two.

The clearest version of this in my own data is what happened between seasons. Nineteen veterans of season one produced literally nothing in season two, one month later. Same Discord, same people, same program, zero activity. Of 123 tracked accounts, 25 had already dropped the role themselves before I got around to reviewing anyone.

That's the honest shape of community member churn. Not a dramatic exit. People quietly stop.

How someone leaves decides what they do next

This is the actual skill, and it's emotional rather than tactical.

The community manager who forgot these are free people gets bitter when a regular goes quiet. Takes it personally. Guilt-trips them, cools toward them, makes the exit ugly. That's the worst available move, because how someone leaves decides what they do next.

Handle a departure with grace and a core member who moved on stays a friend of the project. They come back for the big moments. They refer people. They defend you from the outside, and an unprompted defence from someone with no stake left is worth more than anything they'd have posted while still inside.

I looked for research proving that a graceful exit produces advocacy later, and it doesn't exist for communities. What does exist is the employment version. A 2021 study in the Journal of Management tracked 30,714 people and found that among employees who came back to a former employer, 64% of those who had first left for a negative reason left for a negative reason the second time too. Rehires also turned over at 36.6%, against 20.9% for internal moves. Why someone left the first time keeps predicting their behavior years later, which is the closest hard evidence I can point at for something community managers already know in their gut.

This is why, when I audited the quiet members in season two, the recommendation I wrote for myself was one DM asking what happened, not a role removal. And for the season-one veterans sitting inactive, the call was to strip the current-cohort role and leave the veteran role in place. They keep the status they earned, the roles stay honest, and nobody gets told they've been demoted for having a life.

Poison the exit and you lose all of that, plus you look small doing it.

Always be forming new ties

Because everyone leaves, this job never closes. You keep forming new ties with new members while your current core is still active, even when a new bond slightly cools the one you had with an original member. That trade feels disloyal. It isn't. Rotating the room's center of gravity is the job.

The goal is a community where no ten people are load-bearing. If the whole thing collapses when three specific members leave, you don't have a community. You have a group chat with a single point of failure.

Open source has the cleanest data on this, because every commit is public. A December 2024 study of more than 36,000 open-source projects found that 89% lost their entire core team at least once, and 70% of those losses happened in the first three years. The authors argue this is normal rather than fatal, which is the point I'm making too. The number that should worry you is the recovery rate: only 27% of abandoned projects ever attracted a new core developer. Losing the core is survivable, and failing to replace it usually isn't.

Earlier work on the truck factor, meaning how many people would have to disappear to sink a project, found that 34% of 133 popular GitHub applications had a truck factor of one, and another 30% had a truck factor of two. Two thirds of well-known open-source software could be stalled by losing two people.

Your Discord is not a software repository, but the failure mode is identical. A small group carries it, the group is unpaid, and nothing about the arrangement guarantees a replacement.

Practically, that means a few unglamorous habits. Recruit while the room feels healthy, because healthy is when you're still adding ties, not when you stop. Never build a flow only one person can run, and where the flow is repetitive, hand it to a tool instead of a volunteer. Write down what your power users know, so their knowledge outlives their interest. And keep a way for alumni to stay attached at low effort, which for us is a veteran role that costs them nothing to keep.

A room that survives its own core is the only kind that survives you.

What quietly creates the risk	What reduces it
Treating power users as permanent staff	Treating them as volunteers with a runtime
Flows only one person knows how to run	Written procedure, or a bot doing the repetitive half
Recruiting only when the room feels quiet	Recruiting hardest while it feels healthy
Taking a quiet member personally	One message asking what happened
Removing status when someone drifts	An alumni role that costs nothing to keep
Bending the room around your loudest members	Accepting that someone is always unsatisfied

You can't satisfy everyone

There will always be someone unsatisfied. Most of the time they're right to be, and it's still not a problem you can solve, because satisfying everyone was never on the table.

Chasing 100% is how you burn out. Worse, it's how you end up hostage to your loudest, most demanding members, bending the whole room around the few people you're most afraid of losing. The ones you're most afraid to lose are exactly the ones you should be least dependent on.

That's the same discipline as everything above. Know that they're free, build so their leaving doesn't break anything, and let them go well when they go.

I do this work for a living, across more than fifteen crypto projects and two cohorts of my own. If you're looking at a room that runs on four people and you'd like a second opinion on how exposed you are, book a call.

Sources

• Inner Circle season one and season two operating data, Syndicate cohort platform (production database), May to July 2026. Figures cited: applications, accepted members, weekly actives, weekly submissions, approval rates, and the final leaderboard distribution, which I computed for this article.
• Jakob Nielsen, "The 90-9-1 Rule for Participation Inequality in Social Media and Online Communities," Nielsen Norman Group, 8 October 2006, retrieved 2026-08-09. https://www.nngroup.com/articles/participation-inequality/
• Li, H., Hecht, B., et al., "All That's Happening behind the Scenes: Putting the Spotlight on Volunteer Moderator Labor in Reddit," ICWSM 2022 (moderator-log Gini distribution and invisible labor), retrieved 2026-08-09. https://arxiv.org/abs/2205.14529
• Li, H. et al., "Measuring the Monetary Value of Online Volunteer Work," ICWSM 2022, retrieved 2026-08-09. https://arxiv.org/abs/2205.14528
• Northwestern University, "Unpaid social media moderators perform labor worth at least $3.4 million a year on Reddit alone," May 2022, retrieved 2026-08-09. https://news.northwestern.edu/stories/2022/05/unpaid-social-media-moderators
• Pethig, F. et al., field experiment on newcomer retention in an online community, MIS Quarterly 49(2), 2025, retrieved 2026-08-09. https://www.florianpethig.com/preprints/nudge.pdf
• Foster-Bey, J., Grimm, R., Dietz, N., "Keeping Baby Boomers Volunteering: A Research Brief on Volunteer Retention and Turnover," Corporation for National and Community Service, March 2007, retrieved 2026-08-09. https://www.govinfo.gov/content/pkg/GOVPUB-Y3_N21_29-PURL-LPS113705/pdf/GOVPUB-Y3_N21_29-PURL-LPS113705.pdf
• "Myth: The loss of core developers is a critical issue for OSS communities," arXiv preprint, December 2024 (36,000+ projects), retrieved 2026-08-09. https://arxiv.org/abs/2412.00313
• Avelino, G., Passos, L., Hora, A., Valente, M.T., "What is the Truck Factor of popular GitHub applications? A first assessment," PeerJ Preprints, 2015, retrieved 2026-08-09. https://peerj.com/preprints/1233.pdf
• Arnold, J.D., Van Iddekinge, C.H., Campion, M.C., Bauer, T.N., Campion, M.A., "Welcome Back? Job Performance and Turnover of Boomerang Employees Compared to Internal and External Hires," Journal of Management 47(8), November 2021, pp. 2198-2225 (n = 30,714), retrieved 2026-08-09. DOI https://doi.org/10.1177/0149206320936335 (open copy: https://apps.it.purdue.edu/sites/Home/DirectoryApi/Files/baea7601-cbe7-4b10-b3ad-cd93e060e489/Download)