Gym member churn prediction: the signals your software already logs and nobody checks

A member cancellation is almost never a surprise. It's the end of a 2-3 month process that your management software records step by step: they stop showing up, stop booking, miss a payment. The data is already there; what's missing is someone actually looking at it. A member who drops from 8 visits a month to 2 will cancel within 60-90 days with very high probability, and you can still do something in that window. Once the cancellation email arrives, you can't.

This article covers the signals ranked by reliability, how to set up the alert this week with your existing software or a spreadsheet, what to do with the at-risk list (and what not to do, which matters almost as much), and where AI genuinely helps versus where it's just noise.

Why churn is a marketing problem, not just an operations problem

Start with the calculation that connects cancellations to your ad budget, because it's what makes the rest worth reading.

Say you acquire 20 members a month with €1,000 in spend. CAC of €50, reasonable. But that same month 15 members leave. Your net growth is 5 members, so the real cost of growing by one member isn't €50, it's €200. Three quarters of your marketing isn't growing the business; it's filling the hole in the bucket.

Same gym, same CAC, and the one on the right grows at the same rate while spending 40% less. Every percentage point of churn you cut makes your entire marketing cheaper without touching a single campaign. That's why this topic lives in the same cluster as the real ROI of marketing: retention is the hidden variable that decides whether your campaigns are profitable. And benchmarks matter: if you don't know whether your churn is normal, here are the industry retention statistics to compare against.

There's a second, subtler connection: churn by campaign. Members acquired through discount hook offers leave earlier than those who clicked an ad with a visible price. If you track cancellations by acquisition source, you discover which campaigns attract real members and which ones rent people for three months. That requires calculating LTV by segment, but the simple version covered here doesn't need it.

The signals, ranked by reliability

Your management software already logs all of these. Listed from most to least predictive.

1. Drop in attendance: the top signal

Nothing predicts cancellation like stopping coming. The typical pattern: a stable member at 8-10 visits/month drops to 4, then to 2, then to 0, with 60-90 days between the first drop and the cancellation email. What you need to watch isn't total absence (by the time someone's been gone a month, it's already too late), it's the change in trend: this month's visits compared to their own average over the previous three months.

The nuance that avoids false alarms: compare the member against themselves, not against the gym average. Someone who always came 3 times a month and still comes 3 isn't at risk; someone who came 10 times and now comes 4 is, even though 4 is more than 3.

2. Stopping class bookings

In gyms with booking systems (boxes, boutiques, studios), this is an earlier variant of the above: the member stops booking before they stop coming entirely, and sometimes books and doesn't show, which is a double signal. If your model is open access with no bookings, skip this one; turnstile attendance does the job.

3. Failed payment

One returned direct debit is often just a banking issue. Two in a row, almost never. A failed payment also has a useful property: it's the only signal your software actively notifies you about, so use it as a trigger for a personal human follow-up, not just an automatic retry.

4. Not renewing a pass or letting a plan lapse

In pass-based models (8 sessions, 10 classes), a member who uses the pass down to the last session and takes two weeks to buy the next one is deciding whether to come back. That gap between passes is one of the few risk windows with an exact visible date in the software.

5. Zero app interaction

If your gym has an app (bookings, progress tracking, community), a member who stops opening it is mentally disconnecting before they physically disconnect. It's the weakest of the five signals, with many false positives (plenty of loyal members who simply hate apps), but as a confirmation on top of another signal it adds weight.

The general rule: one signal in isolation is noise, two signals at once are a warning, and a drop in attendance plus any other signal means a member with one foot out the door.

How to set up the alert without AI, this week

You don't need new software. You need one rule and a 20-minute weekly routine.

The rule, in its minimal form: visits this month under 4, and average for the previous 3 months above 8 → at-risk list. Adjust the thresholds to your model (in a box averaging 3 classes per week, it would be "fewer than 6 versus more than 10"), but don't overcomplicate it at the start.

How to run it, depending on what you have:

• Your software includes it. Many systems have "inactive members" or "at-risk" reports that nobody has ever opened. Check there first; if you're comparing management software options, retention alert capability is a criterion almost nobody weighs and should.
• Your software doesn't have it, but exports. Export attendance for the last four months to a spreadsheet, pivot table with visits by member and month, one column with the rule formula, filter. The first time takes an hour; after that, 15 minutes every Monday.
• Pass-based model. Even easier: list of passes expiring or running out in the next 15 days, crossed with who hasn't bought the next one.

The routine produces a short list, 5 to 15 names per week in a gym of 300 members. And here's the part that decides whether any of this is useful.

What to do with the list (and what not to)

What doesn't work, plainly: the generic "we miss you" email with a 10% discount. Negligible response rate, and the discount trains your members that stopping coming is rewarded. Newsletter blasts and push notifications from the app don't work either: the member who's disengaging has already stopped opening your mass messages.

What works is a personal message from someone the member actually knows, ideally their coach or the instructor of their regular class. WhatsApp or a call, the tone of a real person not a company, without ever mentioning cancellation. A script that works:

"Hey Marta! It's Carlos from the box. We've missed you at the 7pm classes, everything okay? This week we're doing [something specific: clean technique, mobility session]. If that slot doesn't work for you anymore, let me know and I'll find you another. Take care!"

Three ingredients and none are negotiable: sent by a person with a name the member recognises, asks before proposing anything, and offers something concrete and easy (a different class, a different time), not a discount. Half the time the response reveals the real reason (injury, shift change at work, moving house), and many of those reasons have a solution if you know about them in time. Freezing a membership for a month is worth more than losing the member, and switching to a different time slot costs nothing.

Does it always work? No. Some names on the list will leave anyway, because some cancellations are unavoidable. The realistic goal is recovering between 20% and 35% of at-risk members contacted in time. In the 300-member gym from the example, that's 3-5 avoided cancellations per month. At €45 per membership and an average extra retention of 8-10 months, each monthly batch of avoided cancellations is worth €1,200-2,000 in future revenue. For 20 minutes of spreadsheet work and four WhatsApps from a coach.

Compare that to acquisition: replacing those 3-5 members via ads would cost €150-400 in ad spend plus sales time, and the new member still has the highest-risk cancellation period ahead of them (the first 90 days), while the recovered member already got through it. Retention costs 5 to 10 times less than acquisition, and it connects directly to your revenue data, which is where real marketing decisions happen.

Where AI actually helps and where it's smoke

The spreadsheet rule has real limitations. It looks at one signal at a time, uses fixed thresholds the same for everyone, and runs once a week. Continuous multi-signal scoring does the same thing better: it combines attendance, bookings, payments, and tenure into a probability per member, learns each person's normal pattern (the stable 3-visits-a-month member doesn't trigger; the one who drops from 10 to 5 does), and recalculates daily. That's AI applied with purpose, and the practical difference is catching the at-risk member 2-4 weeks earlier with fewer false positives than the fixed rule.

The smoke is also plentiful. Three red flags to spot it: they sell you "churn prediction" without access to your attendance data (without that signal there's no valid prediction), they promise accuracy rates without telling you what they were measured against, or the product ends in a pretty dashboard that nobody looks at rather than a list of names with assigned actions. Prediction without the coach's WhatsApp behind it is just another report. The bottleneck in anti-churn is almost never the algorithm, it's someone actually making the contact.

Our view at Pilotium, where we work on the other half of this equation (making sure the member who came in through a campaign is the kind who stays), is that a simple rule well executed captures 70% of the value, and AI scoring makes sense once you already have the contact routine working and want to refine it. Habit first, algorithm second.

Where to start on Monday

Export attendance for the last 4 months. Apply the rule (fewer than 4 visits this month, more than 8 on average before). Pull the list, share it among coaches with the script above, and note every response. In 4 weeks you'll know your real recovery rate and have the number that turns this into a habit: how many euros of future membership revenue you saved that month.

And once you have two or three months of data, do the second read: cross the cancellations against the acquisition campaign for each member. If members from one particular campaign consistently leave earlier, a coach with WhatsApps won't fix that, changing the offer will. Churn is giving you free marketing information. Almost nobody collects it.