What is customer churn rate?

Churn rate is the share of customers who stop buying from you over some period.

That sentence is agreed on. Almost nothing after it is.

Three separate choices sit inside the number, and each one moves it. Who counts as a customer at the start. What counts as leaving. How long the period is.

Change any of the three and the figure changes without a single customer behaving differently.

Customer churn rate, as published on 2026-08-26. One formula adds newly acquired customers back into the calculation and names the error of not doing so. A second publishes the same arithmetic without that term. A third leaves new customers out of the population entirely. Outside subscriptions there is no cancellation to count, so the window that defines “churned” has to be chosen — measured from the store’s own orders in one guide, read off the product type in another. And the same figure is published as a monthly rate and an annual one, with a conversion between them that is not multiplication by twelve.

What is customer churn rate?

Churn rate is the percentage of your customers who stop buying over a chosen period.

The period is a choice you make, not one the metric hands you. Yotpo’s guide puts it plainly — “you need to decide on the time period for the calculation. This is usually a month, quarter, or year.” It recommends monthly or quarterly for most ecommerce businesses, so “daily changes” do not skew the result.

Hold on to that. Choices like it sit behind the disagreements below, and they are not always printed beside the number.

Why can’t an online store just count who cancelled?

Because outside subscriptions there is no cancellation to count.

In a subscription business the customer performs the loss event. Shopify’s guide says the calculation is straightforward “for any given calendar window (often a month or year), using a customer’s time of cancellation”. There is a date, a record, a row in a table.

A store selling skincare or apparel has no such row. The same guide is direct about it: businesses without active cancellations “can’t be measured this way”, and — its words — “In reality, you can’t say with 100% certainty.”

So the number a non-subscription store reports is not an observation. It is the output of a definition somebody chose.

Does the method change with the business model?

Yes, and this is the one thing the sources agree on.

Shopify: “There are multiple ways to calculate churn rate based on the type of business.” Gorgias, independently: “Your calculation method depends on your business model. Subscription businesses track when customers cancel, while non-subscription stores use cohort analysis to identify when customers stop returning.”

Two unrelated publishers, one position. They diverge immediately after.

What is the subscription churn formula?

The version with the correction term takes the customer count at the start of the period, subtracts the count at the end, adds the customers acquired during it, and divides that by the start-of-period count.

(Customers at the start of the period − Customers at the end of the period + New Customers Acquired in the Period) / Customers at the start of the period

Three inputs, not two. The third — the correction term — is the one that does the work.

Why are new customers added back in?

Because without the correction term, the people you just acquired paper over the people you just lost.

The reasoning is published alongside the formula: review only the total change over the period and “it’s possible that your online store’s newly acquired customers would” “hide” “the churn from your existing customers.” The version carrying the correction term “avoids falling into this analysis trap.”

Any store whose acquisitions roughly offset its losses looks almost flat on net change while having churned everyone it churned. The uncorrected version reports the offset, not the loss.

What does the formula look like with real numbers?

The worked example: 100 customers on 1 October, 105 on 31 October, 10 acquired during the month, which returns 5% churn.

(100 − 105 + 10) / 100 = 5/100, or 5% churn rate

Note what kind of month that is. The store finished with more customers than it started with. It grew, and it still churned 5%. That is exactly the case the correction term exists for.

What happens if the correction term is left out?

Gorgias’s guide publishes the same operation without the correction term. It takes the start-of-period count minus the end-of-period count, divided by the start-of-period count, times 100.

[(customers at the beginning of the time period − customers at the end of the time period) / customers at the beginning of the time period] x 100 = customer churn rate (%)

Its own worked example: 5,000 subscribers at the start of the month, 4,800 at the end, giving “4% monthly churn rate”.

That page does not say how many customers it acquired that month. Its instructions, though, ask you to “find how many customers canceled during that same period” — a quantity the formula underneath then never uses. The steps gather cancellations; the arithmetic takes a net change. Those are the same number only when nobody joined.

Can a churn rate come out negative?

It can, and one published example makes it happen. Run the growing store from earlier — 100 at the start, 105 at the end, 10 acquired in the month — through the uncorrected formula.

[(100 − 105) / 100] × 100 = −5%

Same store, same month: 5% churn with the correction term, minus five without it.

A negative churn rate is not a small discrepancy to reconcile. It is the metric reporting that customers arrived from somewhere and calling it retention. Any store whose customer count grew over the period gets a number below zero from that formula.

Is there a third way to treat new customers?

There is, and it is neither of the above. Instead of correcting for newly acquired customers, you can leave them out.

The glossary version states it directly: “you usually focus on customers who were already around at the beginning of your chosen time period. You don’t typically include brand new customers who just joined in that same month” — they have not yet had the chance.

That is a different population from either of the other two. Three published treatments of the same people: corrected for, silently absorbed, excluded.

What counts as a customer who left?

Here the third approach runs into the problem from earlier. Its formula divides the number of customers who left by the number of customers at the start, then multiplies by 100.

Churn Rate = (Number of Customers Who Left / Number of Customers at the Start) x 100%

The numerator needs a count of who left — the quantity a store without cancellations cannot observe.

The resolution is to define it. Count, per the guide’s own step, “how many of those initial customers didn’t make a purchase or canceled their service by the end of the period”. Not purchasing becomes the same input as cancelling.

Does a quiet month make a customer churned?

Under the no-purchase definition, yes — and that is worth sitting with before you adopt it.

The test is whether a purchase landed inside the period. Pick a monthly period, as that guide suggests, and any customer whose normal gap between orders runs longer than a month fails the test in at least one period between two perfectly ordinary purchases.

Somebody who reliably reorders every quarter is counted as lost twice, then returns from the dead on the third month. Nothing about that customer changed. The period did.

How do you measure churn with no cancellation at all?

By cohort. Group customers by when they first bought, then watch that group.

Both publishers describe the same mechanism. One: “you would group customers based on their date of first purchase, and all activity after that”. The other: “This method groups customers by when they made their first purchase and tracks how many return within your expected repurchase window.”

The mechanism is agreed. The window is not.

How long before a customer counts as churned?

Two published answers, and they come from opposite directions.

One derives the window from your own data: measure “the average number of days between their orders” across your repeat customers, then treat a cohort’s churn as “the percentage of customers who didn’t reorder over a timeline of two times the average repeat purchase timeline.”

The other assigns it from the catalogue: “Define your repurchase window based on your product type (e.g., 30 days for consumables, 90 for apparel).”

One window is a property of your orders. The other is a property of your catalogue. Nothing makes them agree, and the churn rate is computed over whichever one you picked.

Do the two window methods give the same answer?

Their worked examples do, and that is the trap.

One: 1,000 customers in January 2022, 300 reordered over the following six months, “that cohort would have a churn rate of 70%”. The other: 1,000 first purchases in January, 300 who “purchased again within 90 days”, “your churn rate is 70% [(1,000 - 300) / 1,000 x 100]”.

Identical headline. Six months against ninety days — windows a factor of two apart, producing the same 70% because each example pairs its own window with its own numbers.

Read those two side by side and you would conclude the field agrees. The arithmetic agrees. The question being answered does not.

Is churn rate monthly or annual?

Both are published, under the same two words, and the gap between them is not small.

Shopify’s guide gives “a 5% monthly churn rate” as an average for subscription businesses. A subscription billing network’s benchmark page reports “median annual churn rates” — with “2% to 4% annual churn” as the range where well-run subscription businesses sit, and 4.25% median annual for ecommerce specifically.

5% a month and 4.25% a year are not two views of one business. They are different businesses.

Does a monthly rate times twelve give the annual rate?

No, and one of these pages says so outright.

A published band puts subscription ecommerce at “3-8% monthly churn (36-96% annually)” — 3 × 12 and 8 × 12.

The benchmark page states the rule against it: “annual churn is not simply 12x monthly churn because of compounding. A 2% monthly churn rate translates to roughly 22% annual churn, not 24%.”

Its own example checks out. Compounding 2% over twelve months gives 21.5%, which is the “roughly 22%” it prints.

What does compounding do to that published range?

Compounded over twelve months rather than multiplied, 3% a month is 30.6% a year instead of 36%, and 8% a month is 63.2% instead of 96%. The top of the range moves a long way.

At the low end the error is about five points. At the high end it is nearly thirty-three. A store told to expect up to 96% annual churn is being handed a number the arithmetic behind it does not produce.

Why does a benchmark stop matching its source?

Because the source can change under a stable link, and the sentence citing it does not know.

The 5% monthly figure is credited, on the live page, to that same benchmark URL. That URL today publishes annual medians. The two earliest archived captures line up with each other — the benchmark page was publishing monthly figures in October 2022, three weeks before the earliest capture showing the citing sentence already in place. What the archive does not establish is when that sentence was written.

What it does establish is that the page moved from monthly to annual and the citation stayed put. Nobody made an error. A reader following that link today is sent for a monthly average and lands on annual medians, and neither page says anything is wrong.

What is a good churn rate?

The honest answer is that the published figures are not on one scale.

For single-purchase brands, one guide puts “about 75% churn per cohort” at average — a sentence carrying no source. Elsewhere, non-subscription stores “experience 60-80% annual churn on average”, credited to an analysis of over 1,000 stores measuring customers who “made at least one purchase but didn’t return within a year”.

A one-year window is the measured-window rule only for a store whose average repeat gap happens to be six months. For anyone else, those two bands are answers to different questions.

Which benchmark can you actually compare against?

The one that publishes its basis, its population and its date together.

The network benchmark page does: “All figures are updated with July 2026 data”, drawn from a named network of subscription businesses, split by vertical and by revenue band.

It also argues against its own headline number. The most useful comparison, it says, is against businesses with a similar customer profile, “not against an industry average that blends premium B2B SaaS with low-ARPC direct-to-consumer products.”

The bands earlier on this page cannot be lined up against each other either: one names its study and its window, the other names neither.

What should you write down beside the number?

Four things, and they take one line:

• The denominator. Start-of-period count, and whether newly acquired customers were corrected for, absorbed, or excluded.
• The definition of churned. A cancellation, or a window with no purchase in it.
• The window. In days, and where the number came from — your order data or a category rule.
• The period and the unit. Monthly or annual, and if you converted between them, that you compounded rather than multiplied.

A churn figure with those four beside it can be compared to last quarter’s. Without them it cannot be compared to anything, including itself.

What does this change about reading a churn figure?

Treat a published churn rate as a number plus four hidden settings, and assume the settings differ from yours until you have read them.

When a dashboard figure and a benchmark disagree, the gap is more likely to be a denominator, a definition, a window or a unit than a fact about your customers. Check all four before you change anything about the business.

And when your own churn rate moves, the first question is whether anything in the calculation moved with it. A number that can come out negative on a growing store is a number that deserves its assumptions written down.