Memberships and boxes retain differently, not just worse
Memberships and subscription boxes do not only churn at different rates. They churn in different shapes. Recurly's network data puts median churn at 4.25% for ecommerce and 4.14% for digital media, close enough to look like the same business. Underneath, a box loses subscribers on a shipment clock and a membership loses them on a renewal clock, and Churnkey measures month-one churn at 28% to 47% for low-priced consumer subscriptions. Same headline number, different curve, different offer.
The benchmarks post answers "what is normal for my store". This one answers a different question: why two stores with identical monthly churn need opposite save flows. The answer is that each vertical runs on its own clock, and the clock decides where in the curve your subscribers leave.
The levels look similar. The curves do not.
Start with the levels, because they are the part people quote and the least useful part.
Recurly's churn benchmark research (July 2026 network data, published as median annual churn) splits total churn into its voluntary and involuntary components by industry.
| Industry | Total | Voluntary | Involuntary |
|---|---|---|---|
| SaaS | 3.22% | 2.16% | 1.06% |
| Business and professional services | 3.44% | 2.27% | 1.18% |
| Travel, hospitality and entertainment | 3.91% | 2.63% | 1.28% |
| Digital media and entertainment | 4.14% | 2.55% | 1.59% |
| Ecommerce | 4.25% | 2.87% | 1.38% |
| Education | 4.99% | 3.30% | 1.69% |
| All industries | 3.60% | 2.34% | 1.25% |
The spread between the leakiest and tightest industry is under two points. If churn level were the whole story, a membership site and a coffee box would be running nearly the same playbook.
They are not, and the reason is visible in a second dataset. Churnkey's voluntary churn benchmarks, built on Stripe transaction data plus roughly 25 million subscriptions on its own platform, report monthly churn by subscriber tenure: 12.0% under three months, 7.40% at three to six months, 4.90% at nine to twelve months, and 3.20% past twelve months. Early-tenure churn runs about 3.75 times the mature rate. A single monthly average is a blend of those bands, and the blend depends entirely on how your cohorts are distributed. Two stores can post 5% monthly churn with completely different tenure mixes and completely different fixes.
One methodological warning before going further. Recurly's figures are labelled median annual on its own page while Churnkey publishes explicitly annual figures roughly seven to nine times larger. The two are not comparable and should never be put in the same chart. Below, each source is quoted with its own label and used for its own purpose: Recurly for the voluntary/involuntary split, Churnkey for shape.
Four clocks
The transferable idea is that a subscription's retention shape is set by what event forces the customer to make a decision.
| Vertical | The clock | Where churn concentrates | Dominant reason | Offer that fits |
|---|---|---|---|---|
| Content membership | Renewal charge | Spread, slow gradient | Not using it | Downgrade or pause |
| Community | Social activity | Follows the room, not the file | Nobody is here | Neither; fix the room |
| Media / streaming | Release calendar | Sawtooth, high return rate | Nothing to watch | Cheap re-entry |
| Curated box | Shipment number | Cliff at shipment 1 to 3 | Bored, piled up | Skip or cadence change |
| Replenishment | Consumption rate | Cliff at first cadence mismatch | Too much supply | Skip or interval change |
Everything below is that table, with the evidence.
Content memberships: consumption is invisible until the charge
A membership is always available and never delivered. Nothing arrives to remind the subscriber it exists. So under-use accumulates silently and gets converted into a decision only when a charge appears on a statement.
The reason data supports this. Churnkey's analysis of two million cancellation survey responses puts budget limitations at 32.97% of stated cancel reasons and infrequent usage at 30.60%, with expectations not met a distant third at 8.63%. Recurly's 2026 State of Subscriptions reports that 52% of consumers cancelled at least one subscription in the past year due to lack of use.
Two consequences follow, and they are specific to the shape rather than the rate.
First, the membership curve is a gradient rather than a cliff. There is no shipment to disappoint at month one, so the month-one drop is milder than a box's, but the bleed never really stops because non-use keeps compounding. That produces the frustrating pattern membership operators describe: a curve that looks healthy for the first quarter and is still losing 3% a month at month eighteen.
Second, the save offer that matches is a downgrade rather than a discount. "I am not using it" is a capacity problem, and a smaller tier answers it directly. Churnkey's data on offer durability is the strongest argument here: plan changes added seven to eight months of additional customer life on average and 30% of plan-changers were still subscribed twelve months later, against 5.1 additional months for a discount and 11% still subscribed after a year. Plan changes are accepted by only 7.72% of cancelling customers against 62.49% for discounts, so they save fewer people and keep the ones they save far longer.
Memberships on WooCommerce covers how that maps to WooCommerce Memberships and Subscriptions plan structures.
Communities: the honest gap
Communities are the vertical where we have the least to offer, and it is worth saying so rather than borrowing a number from an adjacent category.
No preferred publisher we could reach publishes a retention curve for paid communities. Recurly folds them into digital media or education; Churnkey's nearest bucket is digital goods at 27% annual voluntary churn, which also contains software and content. Neither is community data.
What can be said from mechanics rather than benchmarks: a community's retention is a function of the room's activity, not the individual's tenure. A member who joins an active community retains on the strength of other members' behaviour. That makes community churn correlated across the file in a way no other vertical's is, which in turn means cohort curves are less predictive there than anywhere else. Treat community retention as an operating metric of the community, not a subscription metric, until somebody publishes real curves.
Media and streaming: a revolving door, not a curve
Media is the only vertical with genuinely good public data, and its shape is unlike the others: churn is high, and so is return.
Antenna's Premium SVOD panel put twelve-month survival for 2024 signups at 36% overall, with standalone services running lower still: Disney+ 33%, HBO Max 31%, Hulu 28%. Those are brutal numbers by any other vertical's standard. But Antenna also measured 30% of 2023 gross additions as resubscribers, and gross churn of 5.3% in September 2024 against net churn of 3.1%. The same customers keep coming back.
The other finding worth stealing is the bundle effect. The Disney+, Hulu and HBO Max bundle led all plans with a 59% twelve-month survival rate, and Antenna notes the survival gap between bundles and standalone services grew by more than 2x from month one to month twelve. Widening, not narrowing. Bundling does not just raise retention; it changes the slope.
For a WooCommerce merchant the analogue is not a media bundle. It is any structure that makes cancellation cost the customer more than one thing at once: a membership that carries a store discount, a box that includes community access. The mechanism is the same and the effect compounds with tenure.
Curated boxes: the novelty clock
A curated box is judged shipment by shipment, and its retention curve is indexed to shipment number rather than to calendar months. Ship monthly and the two coincide; ship quarterly and a "month 3" reading means something completely different.
This is where the public data runs thinnest, and we are not going to paper over it. No source we could reach splits curated boxes from replenishment as separate curves. Recurly lumps both into ecommerce at 4.25% median churn. Churnkey's nearest bucket is merchandise at 30% annual voluntary churn. Recharge's published reports, which are the right source for shipment-indexed data, were unreachable throughout this research. Any table you find online splitting "curated boxes 10-15%, replenishment 5-8%" traces to aggregator sites with no named dataset behind them. We are not repeating those numbers.
What is defensible is the mechanism, stated as reasoning rather than as data. A curated box sells surprise, and surprise decays. The first shipment either matched the expectation set at checkout or it did not, which produces the cliff. Subscribers who clear the third shipment have generally re-bought the concept three times, and the curve flattens. The stated reasons in this category cluster around boredom and accumulation rather than price, which is why the fitting offer is a skip rather than a discount: the subscriber does not want it cheaper, they want it less often. Subscription boxes on WooCommerce goes into the mechanics.
Replenishment: the cadence clock
Replenishment is the vertical where the subscription has an objectively correct interval and the merchant usually guessed it at checkout.
If the customer consumes a bag of coffee in five weeks and you ship every four, supply accumulates. The cancel arrives not when value fails but when the cupboard fills, which is typically two or three shipments in. Nothing about the product was wrong. The interval was.
That is why replenishment carries the highest save-rate ceiling of the common categories: a large share of its cancel intent is answerable with a schedule change that costs the merchant nothing. Skip-next and interval-lengthening both preserve full price. The Churnkey durability figures above apply with unusual force here, because a cadence change is a plan change in everything but name.
The pause data supports the same conclusion from a different angle. Recurly's 2026 report describes top merchants driving a 337% increase in pause usage to save customers who would otherwise have cancelled, and reports three of four subscribers who pause eventually returning; its digital media report puts pause retention of at-risk subscribers at 51.7%. Churnkey measures pause acceptance at 22.32% and pauses adding 5.5 months of additional customer life.
The price axis cuts across all four
The most underused number in Recurly's benchmark set is not the industry table. It is the price table.
| Revenue per customer | Total | Voluntary | Involuntary |
|---|---|---|---|
| $10 to $25 | 4.29% | 2.99% | 1.30% |
| $25 to $50 | 3.84% | 2.73% | 1.11% |
| $50 to $100 | 3.15% | 2.41% | 0.74% |
| $100 to $250 | 2.87% | 2.40% | 0.46% |
| Over $250 | 3.07% | 2.90% | 0.18% |
Voluntary churn is nearly flat across the whole range, moving from 2.99% to 2.90%. Involuntary churn collapses by a factor of seven, from 1.30% to 0.18%. Almost all of the apparent "cheap subscriptions churn more" effect is a payments problem, not a value problem.
That is a cross-vertical finding, and it reorders priorities for every low-priced business in every category above. A $12 membership and a $15 box both have more to gain from card updaters and retry logic than from any offer, and Churnkey's involuntary benchmarks agree: involuntary churn is 29% of all churn for digital goods and 22% for SaaS. Before tuning a save flow at that price point, read the failed payment recovery stack.
Churnkey's scale data adds a wrinkle low-priced merchants should expect: sampling half a million subscribers at or under $20 a month, month-one churn ran 47% for companies under $1M ARR and 28% at $20M-plus, and consumer churn worsened with scale from 12.1% to 20.2% while B2B improved. Growth makes consumer retention harder, not easier.
What to do with the shape
- Index your cohort curve to the right clock. Boxes and replenishment should be plotted by shipment number, not calendar month. If your cadence is not monthly, a calendar chart is actively misleading.
- Read where your mass sits. Cliff-heavy means the first delivery or the first charge is failing. Gradient-heavy means non-use is compounding and no offer will fix it.
- Match the offer to the clock, not to the benchmark. Skip and interval change for shipment clocks. Downgrade for renewal clocks. Cheap re-entry for release clocks. A discount is the default answer only when the stated reason is actually price.
- Check the price band before the flow. Under $25 per month, involuntary churn is roughly a third of your problem and offers cannot touch it.
- Distrust cross-vertical tables, including the ones above. Recurly and Churnkey do not use the same period, the same denominator, or the same buckets. Use each for the comparison it can actually support.
The useful claim in all of this is not that one vertical retains better. It is that the shape tells you which lever exists. A cliff is an expectation problem, a gradient is a usage problem, and a sawtooth is a scheduling problem. The rate alone tells you none of that.
What's next
- WooCommerce churn benchmarks for the category rates themselves and what counts as normal.
- Reading cohort retention curves for how to plot cliff, shoulder and plateau on your own data.
- Replenishment stores for the skip and interval mechanics referenced above.
Sources: Recurly churn rate benchmarks (July 2026 network data) and 2026 State of Subscriptions · Churnkey voluntary and involuntary churn benchmarks (November 2025), discounting at cancellation (August 2025) and retention dynamics on the path to $100M ARR (February 2026) · Antenna, State of Subscriptions: Premium SVOD 2025 Year in Review and Resubscription is on the Rise, accessed August 2026.