What is a normal churn rate for a DTC subscription?

What is a normal monthly churn rate for a DTC subscription?

Between 5 and 10% of subscribers lost per month, for a typical physical-product subscription on Shopify. But that blended range hides the number that actually matters, because subscription model predicts churn better than anything else about your brand:

Subscription model	Great	Average	Investigate if above	Source basis
Replenishment, monthly billed (supplements, coffee, pet food, personal care)	around 3%	4-8%	8%	Subbly merchant data, 2024 (best quartile 3.1%, median 6.31%, worst quartile 13.8%); Eightx 2026 synthesis
Access and membership programmes	under 5%	5-8%	8%	Eightx 2026 synthesis
Curation boxes (beauty, apparel, discovery)	under 5%	7-15%	15%	Subbly 2024 (best quartile 4.5%, median 7.1%, worst 12.8%); Churnkey via Swell (10-15%); Eightx 2026
Meal kits	under 8%	8-15% for matured cohorts; early cohorts run far higher	15%	Eightx 2026; McKinsey survey; Second Measure card-transaction panel (see tenure section)

All figures are monthly subscriber churn. The tier cut-offs are derived from the published data rather than invented: “great” is the best-quartile figure where a source reports one (Subbly’s best-performing replenishment quartile sits at 3.1% monthly and its worst above 13.8%), “average” is the consensus band across sources, and the investigate threshold is the band’s ceiling, past which the published data says something specific is usually wrong rather than everything being slightly hard.

A note on annual billing, because you will see figures like “0.5 to 1.5% monthly churn for annually-billed subscriptions” quoted around the web: we went looking for the first-hand source of that number and could not find one; the trail ends at synthesis articles citing each other. We have left it out of the table for that reason. The direction is not in doubt (annual billing removes eleven cancel decisions and eleven payment-failure opportunities a year, and self-selects committed buyers), but the specific figure is, so treat it as folklore until someone publishes the data.

Two sanity anchors from opposite ends of the quality spectrum: Subbly’s merchant dataset (2024 data, monthly customer churn across thousands of merchants) puts the overall median at 7.44% with the best replenishment quartile at 3.1%, and Churnkey’s Stripe-partnership data (200 million subscriptions) found subscription businesses lose 39% of the customers they acquire within a year. Both are consistent with the table: where you sit inside your model’s band is the game.

Why the model gap exists is worth internalising, because it tells you which levers you hold. Replenishment subscriptions ride a consumption habit: the product runs out and the subscription refills it, so churn is mostly a delivery-cadence and payment problem. Curation boxes ride novelty: every box has to re-earn its place, fatigue shows up in the data after roughly four to six boxes, and no dunning setup fixes boredom. If you run a replenishment product with box-level churn, you do not have a category problem, you have a programme problem, and that is fixable.

What does your churn rate mean in practice?

Monthly churn rates look small and compound viciously. This table is the fastest way we know to make a churn number mean something; it is arithmetic, not benchmark data, so it holds regardless of which source you trust.

Monthly churn	Cohort left after 12 months	Median subscriber lifetime
3%	69%	~23 months
5%	54%	~13.5 months
7%	42%	~10 months
10%	28%	~7 months
15%	14%	~4 months
20%	7%	~3 months

Read it with your own base. At 7% monthly churn, the middle of the DTC pack, a January cohort of 1,000 subscribers is down to about 420 the following January, and a 2,000-strong subscriber base loses 140 subscribers every month, each of whom must be replaced by paid acquisition before the programme grows at all. The difference between 7% and 5% churn sounds cosmetic; over a year it is the difference between keeping 42% and 54% of every cohort you pay to acquire, which on most DTC unit economics is the difference between a subscription programme that compounds and one that quietly refills a leaking bucket.

This is also the honest answer to the question the benchmark pages never address, which is what a scary number like 20% monthly actually means: a median subscriber lifetime of three months, 93% of every cohort gone within a year, and a business model that has to rebuy its entire subscriber base five times a year to stand still.

When do subscribers cancel?

Early. This is the most consistent finding across every source we reviewed, and the four independent datasets agree on the shape:

• 60 to 70% of the subscribers a brand loses are gone between order one and order three (Eightx, across most categories).
• 44% of subscription-box cancellations happen in the first 90 days (Swell).
• More than a third of subscribers cancel within three months of starting any given subscription, and over half within six (McKinsey’s consumer survey; meal kits worst at 60-70%+ within six months).
• Churn by tenure falls off a cliff: Churnkey’s payments data has subscribers in their first three months churning at roughly 12% monthly, falling to about 3.2% monthly once they pass a year. Recharge’s platform data shows the same survivorship shape from the other direction: roughly 45% of subscribers still active at six months and about a third at twelve (15,000+ merchants, 2022 data).
• The extreme case is instructive: independent card-transaction data on the meal-kit category (Second Measure’s panel, and Daniel McCarthy’s analysis of the IPO prospectuses) found HelloFresh retaining just 17% of customers at six months and Blue Apron about 28%, with twelve-month retention across nine meal-kit brands ranging from 19% down to 7%. Early-cohort churn in that category runs far above any published “average” band, which is why our meal-kit row above only holds for matured cohorts.

The practical translation: your blended churn rate is mostly a weighted average of a very leaky early curve and a very solid late one, which means it moves when your acquisition mix changes even if nothing about your programme has. It also means the fix is concentrated in a narrow window. A subscriber’s second and third orders are where the routine either forms or does not, and the offers, timing and messaging around those two orders carry more churn leverage than everything you do afterwards combined.

We see the same shape in client data. When we took over the subscription programme of a raw dog food brand, cancellations clustered in month two, before the routine had formed: the acquisition discount expired after order one and the price jump landed exactly when the habit was weakest. Restructuring the offer to hold the discount through order two, and rewarding continuity through the danger window, cut subscriber churn 13% and upcoming-order churn 61.7% while the subscriber base more than tripled, and while new-customer volume fell 30%. Nothing about the platform changed. The month-two cliff was an offer-design problem wearing a churn costume.

What does our own cancellation data show?

Most benchmark content recycles the same three datasets, so here is a fourth. For this article we pulled a snapshot from our own side of the fence: 39,766 cancellation events from January to July 2026, sampled from nine brands across Everboost’s client and audit base and chosen for spread rather than size, across the UK, US and Nordics and spanning supplements, sports nutrition, women’s health, skincare, oral care, hydration, household refills, pet food and juice cleanses. All nine are replenishment-model consumables brands, so these findings describe that model, not curation boxes. Cancellation logs carry no active-subscriber denominators, so everything below is about the composition and timing of churn rather than churn rates, which is also why you can trust it: these are counts, not modelled estimates.

Finding (our panel, Jan-Jul 2026)	Number	Published comparison
Cancellations at or before the 3rd completed order	78% weighted (range 57-95% by brand)	60-70% (Eightx synthesis)
Cancellations at or before the 2nd completed order	67%	no published equivalent
Cancellations within 90 days of subscription start	50% weighted (range 36-73%)	44% for boxes (Swell)
Payment-failure share of cancellation events	4-11% by brand	20-40% (aggregators); under 10% (Churnkey 2025)
Overstock as share of reasoned cancellations	23-46% at eight of nine brands, the top reason almost everywhere	27% of cancellations (Ordergroove)
Cancel-flow save rate	median 14% (range 10-27%)	10-15% typical, 35-40% best (published)
Saves involving no discount	42-62% of saves	no published equivalent

Three of these deserve a sentence. First, the early cliff in our data is worse than the published consensus: nearly four in five cancellations happen by the third order, and two in three by the second. If anything, the industry line understates how early this game is decided. Second, our payment-failure share sits at single digits, near Churnkey’s figure and far below the 20-40% that aggregator content repeats. The honest reading is selection: these are programmes with actively managed payment recovery, which suggests the 20-40% band describes what unmanaged dunning looks like, not what involuntary churn has to be. Third, the overstock finding: at eight of the nine brands, some version of “I have too much product” is the largest stated cancellation reason, a quarter to nearly half of all reasoned cancellations. Subscribers mostly do not leave because they dislike the product; they leave because it piles up faster than they use it, which is why a skip button saves customers a discount cannot. Our smaller amnesia numbers agree with the UK consumer research: “I subscribed by accident” and “I forgot I had this subscription” together account for roughly 5 to 7% of reasoned cancellations where the option is offered.

How much churn is involuntary?

Less than most content claims, and more than most brands manage. The honest answer is a spread, and it is worth seeing whose data sits where: Recurly’s 2024-vintage benchmark (1,200+ subscription sites, 2023 data) put involuntary churn at 1.0 points of a 4.1% overall monthly rate, roughly a quarter of total churn. Churnkey’s 2025 report (15 million subscriptions, 2024 data) puts it under 10% of the total. Aggregator articles consistently quote 20 to 40%, and payment-industry sources have claimed up to half. Our own nine-brand panel above lands with Churnkey: 4 to 11% of cancellation events by brand, in programmes where payment recovery is actively worked. Read together, the spread mostly measures how well dunning is managed, not how much involuntary churn naturally exists: treat 20 to 40% as what neglect costs, and single digits as what managed recovery achieves. Either way, your own split is one report away and beats every published number.

Some involuntary churn is structural before a single decline happens: payment cards are typically reissued around every three years, so roughly one subscriber card in 36 expires in any given month, which sets a floor on payment failure that exists regardless of how good your product is.

This is simultaneously the least glamorous and most fixable churn you have, because nobody involved wants the cancellation. Published benchmarks put recovery at 40 to 70% of failed payments for a well-configured setup, and the platform tooling has become genuinely good: smart retry ladders, quiet retry windows before customer-facing messaging, incentivised card updates, in-portal payment fixes. In our own client work, rebuilding one brand’s payment recovery took it from 48% to 87% of failed payments recovered; the full numbers are in the case study.

Two operating rules keep involuntary churn honest in your reporting. First, split it from voluntary churn in every report, because the fixes share nothing: voluntary churn is an offer and experience problem, involuntary churn is a payments operations problem, and a blended number hides which one you have. Second, treat the voluntary/involuntary split itself with suspicion at low AOV: a customer who wanted to leave and lets a card failure do the cancelling shows up in the involuntary column, which is one reason recovered subscribers still need the same habit-building attention as new ones.

Why do published benchmarks disagree so much?

Because the sources measure different things on different samples and cite each other carelessly, and this SERP is worse for it than most. Four specific traps, all of which we hit while assembling this page:

Different bases. The most-cited source in this space, Recurly’s benchmark research, has published figures across several vintages that measure different things. Its 2024-vintage page (archived, 2023 data, 1,200+ subscription sites) reported average monthly churn of 4.1% overall, and 6.5% for its B2C industry group, a figure usually recycled as “consumer goods churn” although it actually averages digital media, consumer goods and retail, and education together. Its current page (July 2026 data) lists ecommerce at 4.25% labelled as a “median annual churn rate”, while the same page’s FAQ notes that 2% monthly churn translates to roughly 22% annual. Those figures cannot describe the same quantity. Recycled versions of “the Recurly number” circulate from every vintage without the basis attached, which is how an operator ends up benchmarking monthly churn against a number that is neither monthly nor comparable. If you cite the 6.5% figure, cite the archived page and say what it averaged.

Miscitation, including probable citogenesis. One figure repeated across several pages ranking for this topic, “7.1% monthly churn, 4.1% voluntary plus 3.0% involuntary, per Recharge’s State of Subscription Commerce”, does not appear in any first-hand source we could locate, on any Recharge property or anywhere else. Here is what we did find: Recurly’s 2024-vintage benchmark page reports overall churn of 4.1%, voluntary churn of 3.0% and involuntary churn of 1.0%. Add the overall rate to the voluntary rate, 4.1 plus 3.0, and you get 7.1, with the two addends surviving as the “split”. The most likely story is that someone summed the wrong columns of Recurly’s data, attributed the result to Recharge, and the number has been recycled by everyone citing everyone else since. We may be wrong about how the error happened; we are confident no first-hand version of the 7.1% figure is currently findable.

Wrong kind of “retention”. Retention reports from adjacent industries share vocabulary but not metrics: a returns platform’s “retention benchmarks” measure refund value converted to exchanges, an email platform’s measure campaign-attributed repeat orders, a B2B survey’s measure annual account renewal. None of them says anything about your subscriber churn, and all of them rank for the searches you will run.

Samples with survivorship built in. Platform-published benchmarks describe brands healthy enough to be on that platform, agency portfolios describe brands that hired an agency, and public-company disclosures describe survivors at scale (Peloton’s famously low 1.6% monthly equipment-linked churn tells a DTC supplement brand nothing). None of this makes the numbers useless; it makes the sample worth reading before the number.

The working rule we apply for clients: before any benchmark enters a board deck, it needs four attributes attached: what was measured, on whose data, from which period, on what basis. If a number arrives without them, it is content, not data.

What actually moves the churn number?

Having audited and rebuilt a fair number of these programmes, we would rank the levers in this order, and note that none of them is “switch subscription platform”:

1. The acquisition offer decides the cohort’s churn before the first order ships. Deep-discount and gift-led acquisition recruits deal-takers whose churn curve is visibly worse from month one; the pattern shows up in promo-cohort analysis on essentially every account we have audited. The corrective is not killing offers, it is designing them so the economics survive order two: hold the discount through the habit-formation window rather than cliff-dropping after order one, and sell larger supply sizes to buyers who want them, which lengthens the reorder clock and halves the number of churn decisions per year.

2. The window between order one and order three. Everything in the tenure data says this is where the programme is won: onboarding that gets the product used before the second billing (an unopened pouch is a cancellation with a delay on it), a value-framed rebill reminder rather than a silent charge, and a visible reason to stay through the early orders, whether that is a reward at order three or simply the routine taking hold. Brands that advertise a month-three or month-four benefit during month one are playing this window correctly: they give the wobbling subscriber a concrete reason to let the next order land.

3. Cancel-flow alternatives, honestly built. Published save-rate benchmarks run 10 to 15% for basic flows and 35 to 40% for the best reason-mapped ones; our strongest client result took saves from 9% to 35% of cancel attempts, with eight in ten saves coming from skip, pause and plan changes rather than discounts. That composition is the health signal: a discount save often defers the cancellation, a schedule fix removes its cause. The platform data agrees from both sides: Ordergroove’s merchant analysis found subscribers with skip and pause flexibility retain 135% longer, that product overstock alone drove 27% of cancellations (people drowning in unconsumed product, which a skip button fixes and a discount does not), and 78% of subscribers in Chargebee’s consumer research say they want pause and swap options. Hiding the cancel button does the opposite of all this, and UK subscription regulation is moving explicitly against it.

4. Payment recovery, configured rather than defaulted. The 20 to 40% of churn nobody chose is the fastest pure-ROI fix on this list: benchmarks at 40 to 70% recovered, 87% achievable, and every recovered subscriber is one you already paid to acquire.

5. Measurement, so you fix the right leak. Cohorted churn by subscriber age, split by voluntary and involuntary, split by acquisition offer. One specific trap for migrated programmes: some platform migrations assign every imported subscriber to a single placeholder cohort dated at the migration, which silently corrupts every cohort metric downstream; if your dashboard shows a monster cohort on one date, isolate it before trusting anything blended. Blended churn moved by acquisition-mix changes has sent more than one brand chasing a retention problem they did not have.

For the vantage point behind these rankings: Everboost runs subscription programmes for DTC brands across the major Shopify subscription platforms, including through platform partnerships, and none of the levers above required changing any brand’s platform. The programme moves the number. The platform mostly decides how pleasant the number is to move.

The honest nuance to that rule is that tooling ceilings are real, and a migration done with a programme rebuild can shift the curve where a migration alone would not. When we moved PhycoHealth, a marine-science supplements brand, onto Loop Subscriptions this year as part of a full rebuild of its cancel flow, payment recovery and offer structure, monthly subscriber churn fell by a third from its pre-migration peak, and two months in sits roughly a quarter below its pre-migration average. We attribute most of that to the rebuild and some of it to the tooling that made the rebuild easier to ship; separating the two cleanly is impossible, and any vendor case study that does not admit that is selling something. (Everboost is a Loop Subscriptions partner, for the record, and the same rebuild on another platform would have captured most of the same churn.)

How should you benchmark your own programme?

Use external benchmarks to set the coarse expectation for your model, then benchmark yourself against yourself. The external step takes one row of the model table above: a monthly-billed replenishment brand should expect 4 to 8% monthly churn and treat under 5% as good. Everything finer-grained than that is better answered by your own data, because your acquisition mix, AOV, category and offer design move the number more than any industry average explains.

The internal setup that makes the number diagnostic rather than decorative: monthly customer churn by cohort age (first 90 days versus post-90-days minimum), voluntary and involuntary reported separately with a recovery rate on the involuntary side, save-rate and save-composition on the cancel flow, and churn by acquisition source and offer. Watch trend against your own baseline quarterly. A brand that holds steady at 6% while its early-cohort churn falls and its save composition shifts from discounts to schedule fixes is winning even though its headline number looks static; a brand whose blended rate “improved” because acquisition slowed is not.

And if you want the external comparison done properly against your own numbers, with the cohort maths handled and the voluntary/involuntary split built into your reporting, that is a conversation we have with DTC subscription brands most weeks: book a call.