How to predict community churn with a member health score

A member health score is one number, usually 0 to 100, that summarizes how engaged a member is right now. It borrows the idea from software companies, which score customer accounts to spot the ones likely to cancel. In a community the same logic works at the level of a person.

The point is not precision. The point is a weekly list, sorted from lowest score to highest, that tells you who to talk to first. A rough score you check every Monday beats a sophisticated one you never look at.

Five signals cover most communities. Recency: how many days since the member last visited. Contribution: posts, comments and replies in the last 30 days. Learning: progress in courses over the last 30 days. Attendance: live events attended in the last 60 days. Connection: whether other members have replied to them or mentioned them recently.

Connection is the one most creators miss, and it may matter most. A member who posts and gets no reply learns that nobody is listening. A member who has two or three people they talk to regularly has a reason to stay that has nothing to do with your content.

Payment signals belong on the list too, but separately. A failed card is not disengagement; it is an admin problem with a short fuse. Handle it the same day.

Recency, up to 30 points: visited in the last 3 days, 30; last 7 days, 20; last 14 days, 10; longer, 0. Contribution, up to 25 points: 5 or more posts or replies in 30 days, 25; 2 to 4, 15; 1, 5; none, 0. Learning, up to 20 points: course progress in the last 30 days, 20; started but stalled for over 30 days, 5; never started, 0. Attendance, up to 15 points: attended an event in the last 60 days, 15; registered but did not attend, 5; neither, 0. Connection, up to 10 points: received a reply or mention from another member in the last 30 days, 10; otherwise, 0.

Add them up. Then sort into bands: 70 to 100 is healthy, 40 to 69 is drifting, below 40 is at risk. New members in their first 30 days get their own list regardless of score, because early behavior is still forming.

Weights are a starting point, not a law. If your community is mostly a course, raise Learning and lower Contribution. If it is mostly live calls, raise Attendance.

A worked example: Priya visited five days ago (20), replied twice this month (15), has not opened her course in six weeks (5), registered for last week's call but did not attend (5), and nobody has replied to her lately (0). Her score is 45, which puts her in the drifting band. The score also tells you where to aim: the cheapest points to win back are connection and learning, so a reply to her last post and a pointer to the next lesson are better than a generic check-in.

Healthy (70+): nothing special, and that is fine. Occasionally ask one of them to welcome new members or answer a question; healthy members are your best onboarding team.

Drifting (40 to 69): a light, specific nudge. Not "we miss you!" but something useful: "You were halfway through module three, and next week's live call covers exactly that." Mention them in a thread on a topic they have posted about before.

At risk (below 40): a personal message from you or a moderator, written by a human. Ask one question: "Is the community still giving you what you joined for?" Some will tell you what is wrong, which is valuable whether or not they stay. Some will say they are busy, and you can offer a pause instead of a cancellation if your pricing allows it.

The rule to keep: every Monday, the ten lowest scores get a personal message before you do anything else.

After 90 days, look at every member who cancelled. Check what their score was 30 days before they left. If most of them were drifting or at risk a month earlier, the score is doing its job. If many cancellations came from members who scored as healthy, something is missing: maybe they finished what they came for, or their reason was price.

Adjust one weight at a time, and keep a short note of each change. The score gets better by being compared against what actually happened, not by adding more signals.

Some members are quiet and happy. They read every post, take every course, and never comment. A score built heavily on contribution will flag them unfairly, which is another reason the at-risk message should be a real question rather than a guilt trip.

And some churn is success. A member who joined to land a job and landed it may leave satisfied. Track why people leave, in their words, alongside the score. The two together tell you far more than either alone.

HiveLearn's admin includes member health scores, built to help you spot members going quiet without maintaining a spreadsheet. Because courses, community, events and payments are in one place, the activity behind the signals above (posts and replies, course progress, event attendance) is recorded on the same member account, rather than in three tools you would have to stitch together to run the manual model.

The follow-up tools sit next to the score. Member notes and tags record who you contacted and why, segments group members by behavior, the CRM pipeline shows where each relationship stands, and broadcast email reaches a whole segment at once. Members also receive a weekly progress digest email, which is a gentle, automatic nudge for drifting members between your personal messages. See the pricing page for which plan includes health scores.

If your community lives in a chat tool like Discord, you can still run the manual model above. It takes more work to collect the numbers, but the weekly habit of messaging the lowest scores matters more than the software.