Every community team has a dashboard, and almost every one of them reports the same things: members, active members, messages, engagement rate. And almost every one of them gets the same reaction from the person who controls the budget: a polite nod and no follow-up questions. That is not because leadership does not care about community. It is because those metrics are activity counts computed over numbers nobody trusts, and finance has learned not to argue with a chart it cannot check. This is a measurement model that fixes both problems.
The denominator problem
Start with the thing every community metric silently depends on: the size of the community. Engagement rate is engagement divided by members. Retention is members who stayed divided by members who joined. Reach is followers. All of it sits on top of one number, and that number is usually wrong.
In the first production scan CommunityOS ran on a Web3 project's X audience, 78,181 followers went in. 90.96 percent failed authenticity checks: bots, farms, dormant shells, follow-for-follow rings. 5,806 real accounts came out. Any engagement rate that project had ever reported was computed over a denominator more than thirteen times larger than the real one. The chart was not slightly off. It was fiction.
So the first health metric is not a rate at all. It is the real-member base: how many of your members survive a deterministic authenticity filter. Compute everything else over that number. If you cannot compute it, every other metric on the dashboard is an estimate of noise, and leadership is right to ignore it.
Metric one: real-member base
The filter has to be rules-based and reproducible, because the number will be challenged. The signals that hold up: an activity floor, follower-to-following imbalance, posting-burst patterns that reveal one operator running many accounts, and linguistic uniformity that catches templated text. Run it on every scan and track the real base over time. A community that is growing in raw followers but flat in real members is not growing; it is accumulating noise.
Metric two: reciprocity rate
This is the metric most dashboards miss, and it is the strongest predictor of retention we know of. Do not measure how much a member posts. Measure whether they have had a genuine two-way exchange: someone replied to them, and they replied back.
The pattern is consistent. A member who posted twice and was engaged both times tends to stay far longer than a member who posted twenty times into silence. Volume without response is a churn signal dressed up as engagement. And the corollary is actionable in a way most metrics are not: unanswered first posts are the churn you can actually prevent. Find them and respond. The reciprocity rate is also hard to fake, which is why it survives as a health signal when raw activity does not.
Metric three: archetype mix
A healthy community is not one big number. It is a mix of different kinds of contributors, and the mix matters more than the total. Four archetypes cover it: Champions, who speak about you with depth and conviction; Amplifiers, who carry your message into rooms you cannot enter; Builders, who ship things; and Early Adopters, who show up and validate before it is obvious.
Track the count of each over the real base. A community with reach and no conviction is a broadcast list. One with conviction and no reach is an echo chamber. One with no Builders will never produce the assets that keep people coming back. The mix tells you what kind of community you actually have, and what to ask of it.
Metric four: verified actions
Here is the metric that changes the conversation with finance. Stop reporting impressions. An impression is an estimate, computed over an audience that is mostly bots, of how many people might have glanced at something. Nobody can check it. Report verified actions instead: a specific real member did a specific thing, and the proof is attached.
It is a much smaller number. That is the point. It is checkable, and checkable numbers survive a CFO. In CommunityOS every completed outreach mission passes Proof Review before it counts, so the report at quarter end is a ledger of things that actually happened rather than a story about reach. In the scan above, 298 real accounts surfaced as worth acting on that week; the report tracks what happened with each of them, not how many people theoretically saw a post.
Putting it on one page
The health report that works has four lines, each over the real base, each compared to your own prior period:
- Real-member base, and its trend.
- Reciprocity rate, and the count of unanswered first posts you closed.
- Archetype mix, and how it shifted.
- Verified actions, with the ledger available on request.
No engagement rate. No impressions. Nothing computed over a number you cannot defend. It is a less flattering report than the one most teams send, and it is the first one leadership will actually read to the end.
Quick answers
What are the most important community health metrics?
Four, in order. First, the real-member base: how many of your members survive an authenticity filter, because every other metric is computed over this denominator. Second, reciprocity rate: the share of members who have had a genuine two-way exchange, which predicts retention far better than post volume. Third, archetype mix: how many Champions, Amplifiers, Builders and Early Adopters you have, since a healthy community needs all four. Fourth, verified actions: the count of real outcomes you can prove, which is the only metric finance will accept.
Why do engagement metrics fail with leadership and finance?
Because they are unverifiable and computed over contaminated numbers. An engagement rate is likes and replies divided by followers, and if a large share of those followers are bots or dormant accounts, the rate is fiction. In one real production scan, 90.96 percent of 78,181 followers failed authenticity checks, so any engagement rate reported over that list was off by an order of magnitude. Finance cannot check a like; it can check a verified action. Report the smaller, checkable number.
How do you measure community engagement without vanity metrics?
Change the unit from volume to reciprocity. Instead of counting how much a member posts, track whether they have had a real back-and-forth: someone replied to them and they replied back. A member who posted twice and got engaged both times tends to stay far longer than one who posted twenty times into silence, and unanswered first posts are the churn you can actually prevent. Reciprocity is also hard to fake, which is why it holds up as a health signal when raw activity does not.
What is a good community retention rate?
The honest answer is that any benchmark quoted without the denominator is meaningless, because retention over a list that is mostly bots measures nothing. Compute retention over the real-member base only, then track it against your own baseline month to month. The useful question is not whether you beat an industry number but whether real-member retention is rising after each change you make, and whether members with reciprocal exchanges retain better than those without. In practice that gap is the single clearest signal of whether your community is healthy.
How do you prove community ROI?
Stop reporting impressions and start reporting verified actions by real people. Impressions are estimates computed over bot-inflated audiences; a verified action is a specific real member doing a specific thing, with the proof attached. It is a smaller number and it is checkable, which is why it survives a CFO. CommunityOS builds this in: every completed outreach mission passes Proof Review before it counts, so the ROI report is a ledger of verified outcomes rather than a story about reach.