When members leave, and what to measure before buying more trials

What does leaving even mean here?

You have to decide, because the billing system will not decide for you.

This sounds like pedantry and it is the whole problem. A member on a month-to-month plan who stops attending but keeps paying has not left. A member who cancels at the end of a paid-through period leaves twice, once in intent and once in revenue. A member who freezes for the summer has not left unless they never come back, which you find out months later. A class-pack customer never formally leaves at all; they simply stop buying.

Pick one definition and write it down before you count anything:

None of these is right. They are different questions, and a studio quoting a churn number without saying which one it used has told you nothing. The one thing not to do is average them together, which is what most dashboards silently do.

Why measure from the join date rather than by month?

Because a monthly churn percentage hides the shape of the problem.

A studio that loses the same number of members every month can be losing them all in their first eight weeks or losing them evenly across three years, and those are opposite businesses with opposite fixes. A monthly rate cannot tell them apart.

The alternative is a cohort: take everyone who joined in a given month, and follow that group. How many are still here after one month, two, three, six. Do it for a few consecutive joining months and the shape appears immediately, without any statistics.

That is the measurement worth building, and it is one query against the billing system, not a project. The joining month goes down the side, the months since joining go across the top, and the number of people still active goes in each cell. Read it once and you will know whether your problem is the first month or the twelfth.

Two cautions before you act on the picture. Small cohorts move around for reasons that are not real, and a studio with twenty joins a month is firmly in that territory, which is the argument in small sample sizes and how to avoid fooling yourself. And the joining month itself carries a seasonal signature: a January cohort in a fitness business is not comparable to a September one, for reasons covered in how seasonality distorts month-over-month comparisons.

We are not going to tell you where studios typically lose members. We have not read a source we would stand behind for that figure, and the numbers that circulate come from software vendors describing their own customer bases. Your cohort table will tell you where yours lose them, which is the number that matters anyway.

Can Google Analytics do this?

It has a tool that looks like it can, and one line in the documentation rules it out.

GA4 has a cohort exploration, and its shape is right: an inclusion criterion that puts someone in a cohort, a return criterion that counts them as retained. But the documentation is explicit about what it identifies people by: "Cohorts are based on the user's device data only. User-ID is not considered when creating a cohort."

That is decisive. Even a studio that has done the development work to implement User-ID, the identity space Google itself calls "the most accurate," cannot use it in a cohort. The cohort tool is counting devices. A member with a phone and a laptop is two, and a member who clears cookies is new.

Three more limitations worth knowing before anyone builds a report on it.

The same data produces three different answers. The tool offers Standard, Rolling and Cumulative calculations. Standard counts users who met the return criterion in that individual period. Rolling counts only those who met it in that period and every previous one. Cumulative counts anyone who met it in any period so far. A retention figure quoted without saying which of the three was used is not a figure.

A week is a calendar week. Google states it plainly: weekly granularity runs "from Sunday to Saturday included, not on a rolling 7 days." A cohort defined on a Thursday has a short first week.

And the history is short. User-level data on a standard property is retained for two months or 14, and that limit applies to explorations, which is where cohort analysis lives. A twelve-month retention curve is the most you can build, and only if the property was set to 14 months before the period you want to look at.

There are hard caps on the tool itself as well. A cohort exploration shows at most sixty cohorts, a breakdown dimension shows only its top fifteen values, and demographic breakdowns are thresholded, so small groups are dropped entirely to protect anonymity. For a studio with a few dozen joins a month, that last one quietly removes the segments you were most curious about. And people can appear in more than one cohort at once: Google notes that users "are assigned to all cohorts for which they meet the inclusion criterion," so a member who transacts repeatedly shows up in several rows of the same table.

None of this makes GA4 useless. It makes it the wrong instrument for this particular question, and knowing which instrument answers which question is most of what a useful dashboard is, argued generally in setting a baseline before you change anything.

What about the trial?

Measure it as its own cohort, because it behaves nothing like a membership.

A trial has three possible endings and most studios only count one of them. The person converts. The person does not convert. Or the person never attends the trial they booked. That third group is often the largest and it is invisible in a conversion rate that divides members by trials sold.

So the trial funnel is four numbers, not one: booked, attended, converted, and still active ninety days later. The last one is the one that changes decisions. A promotion that doubles bookings and halves ninety-day survival has made the business worse while improving the number on the wall.

That last number also connects to the acquisition question. If the trials are arriving from a discount offer, they are a different population from the ones arriving from a search for the format, and mixing them into one conversion rate averages two different businesses. Where the trials come from is a search question, and the studio version of it is in getting found for the intro class.

What is the smallest report that answers this?

Five numbers, monthly, on one page.

Joins. Leaves, using the single written definition. Net change. Trial bookings and the share that attended. And one cohort row: the share of the members who joined six months ago who are still here.

That is enough to run a studio. It is not enough to satisfy a consultant, which is a feature. Anything else on the page competes with these for attention, and the failure mode of a studio dashboard is not too few metrics.

Add a reason code on cancellation if the software supports it, with a short fixed list rather than a text box. Five options a front-desk person can pick in two seconds beat forty free-text answers nobody reads.

One more thing worth agreeing before the report exists: who reads it and when. A monthly number with no standing meeting is a number nobody acts on, and the studios we see get value from this are the ones where the same person opens the same page on the same day each month and says out loud what changed.

And keep the definition written at the top of the report. The most common way this measurement dies is that somebody changes what counts as leaving, quietly, and the trend line becomes meaningless without anyone noticing.

What would we do first?

In order, and the first two are free.

Write the definition of leaving down. One sentence, agreed by whoever owns the number. This takes ten minutes and it is the step everyone skips.

Pull one cohort table. Six joining months, six columns. Whatever your billing platform is, this is an export and a spreadsheet, not a data project.

Split the trial funnel into four numbers. Booked, attended, converted, still here at ninety days.

Then decide about acquisition. If the cohort table shows people leaving early, more trials make the leak bigger and cost money doing it. If it shows a stable curve after the first few months, acquisition is a reasonable next spend.

And instrument the second visit, because for most studios that is the earliest signal available and it arrives weeks before a cancellation does. The general case for treating the second visit as the real conversion is in the rebook nobody asks for.

Where these numbers come from

Three Google Analytics help pages, read in full on 7 September 2026. There are no statistics in this article, deliberately.

The statement that cohorts use device data and ignore User-ID, the three calculation types, the calendar-week definition and the tool's limits are from cohort exploration. The identity spaces and the accuracy of User-ID are from reporting identity. The user-level retention limits and the note that they apply to explorations are from data retention.

We have not published a churn rate, a month at which members typically leave, or a trial conversion benchmark for fitness studios. We could not read a primary source for any of them, and every figure we found traces to a software vendor describing its own customers. Your own cohort table is both more accurate and easier to obtain than any of them.