Retention is a product problem
Membership businesses generally manage retention commercially: pricing, contract length, win-back offers. Those matter, and they act late, on people who have already decided.
The decisions that determine whether someone stays are made much earlier, in the product, and they are unusually tractable. This paper sets out where they sit.
1. Cancellation is a lagging indicator
By the time someone cancels, the behaviour changed weeks or months before. Nothing went wrong — no complaint, no failed payment. Attendance thinned and the direct debit ran on out of inertia.
Which means retention cannot be managed by exception. There is no exception to catch. It has to be managed by watching for the absence of something.
2. The window is early and short
The pattern is consistent enough to plan around: a new member either establishes a routine within the first few weeks or does not. If they do, they stay for a long time. If they do not, the membership becomes a subscription to a building.
Practical implication: effort spent in week two is worth several times the same effort spent at cancellation. A discount offered to someone halfway out the door is expensive and rarely works.
The number worth managing is not overall churn, which moves too slowly to act on. It is: of the members who joined last month, how many attended at least four times in their first six weeks? If nobody can answer that, it is the first thing to fix.
3. The signals that arrive first
Well before cancellation, the record shows:
- A broken rhythm. Not a missed session — a gap in a pattern that had formed.
- Narrowing. Someone who tried three things now does one, or none.
- Booking without attending. Stronger than not booking at all, because it shows intent that did not survive the day.
- Never started. The member who joined and never came is the easiest to lose and the easiest to help.
None of this requires a model. It requires the booking record and the attendance record to be the same record — which is where most operators are actually stuck, because those live in different systems.
4. Capacity is where booking product decisions bite
Booking systems are built as diaries. The thing that needs managing is the gap between booked and attended:
- A no-show costs twice — the empty place and the waitlisted member turned away.
- Waitlists should release automatically on cancellation, offer a real acceptance window, and stop offering once a class has started.
- Cancellation windows matter less than the reminder that lands before the window closes. That single message converts no-shows into released places.
- Penalties, if any, must be visible and knowable in advance. Rules enforced invisibly feel arbitrary and cost goodwill for little gain.
Measure attendance against capacity by class, instructor and time slot — not bookings. The two rankings are rarely the same, and the difference is where the timetable should change.
5. Attendance capture has to be nearly free
All of the above depends on knowing who actually turned up. In most operations that is an instructor’s job during the busiest thirty seconds of their day.
If capture costs more than a moment, it will not happen reliably, and every retention signal downstream becomes guesswork. Design for the door: a single tap, a scan, an automatic check-in — not a form completed afterwards.
6. Make progress visible
The most underused retention mechanism is showing people what they have already done. A member who can see eleven sessions this quarter has a streak to protect. One who cannot has nothing to lose by stopping.
Rules that matter:
- Honest. Inflated numbers are noticed and resented.
- Compared to themselves, never to other members. Comparison to strangers drives away exactly the people you are trying to keep.
- Unsought. In the app they already open, not in a monthly email.
- Recoverable. A broken streak should not read as failure, or the first missed week becomes the last week.
7. Intervene like a person
Once the signals exist, the temptation is to automate the response. A generic “we miss you” is worse than nothing: it announces that you noticed and did not care enough to be specific.
What works is a short list of members worth a word this week, in front of the right person, with something specific to say. The software’s job is to produce the list and get out of the way.
8. The app is the relationship
For a member attending twice a week, the app is your most frequent contact — more than reception, more than any newsletter. If the icon carries a platform’s brand rather than yours, the habit being reinforced belongs to the platform, and when they change supplier your members go with it.
This only stacks up commercially if branding is per-tenant configuration rather than a bespoke build. Where it is, even a small operator can have its own app; where it is not, only large ones can, and the argument collapses.
9. What to instrument
A short list, most of which operators do not currently have:
| Signal | Why |
|---|---|
| Joined-to-first-visit time | The single best early predictor |
| Visits in first six weeks | The habit window |
| Booking-to-attendance ratio, per member | Drift, before it becomes absence |
| Variety of activities used | Narrowing predicts leaving |
| Attendance vs capacity, per class | What to change in the timetable |
| Waitlist conversion | Whether recovered capacity is real |
| Days since last visit, by cohort | The list worth acting on |
A sequence that works
- Join bookings and attendance onto one record. Nothing else works without it.
- Make attendance capture trivial at the door.
- Instrument the first six weeks, and report on the joining cohort weekly.
- Fix the waitlist loop, which recovers capacity without adding classes.
- Show members their own progress.
- Produce the human list — and give someone the time to use it.
Only then is there anything worth modelling. Most operators find they never need to.
Working on retention? Get in touch, or read about our health and fitness work.