Subscription Churn Benchmarks by Vertical

Read a vertical benchmark as an observation before setting a target#

An observed churn benchmark describes a particular provider’s measured population. A target is a result your business chooses to pursue. They are different: an industry label alone does not make the population, customer price, paid unit or observation period comparable to yours.

Start with the published metric and its scope. Then compare your own like-for-like result, identify the loss mechanism and choose a target justified by your economics and measurement horizon. Do not relabel a consultant’s suggested “healthy” rate as a measured industry median.

Current published vertical figures and their limits#

As checked on October 3, 2026, Recurly’s industry chart labels the following as median annual churn rates from its network, July 2026. These are the chart’s published observations, not universal targets or independently verified calendar-year cohort survival rates.

The page contains inconsistencies: introductory prose gives software’s median as 3.04%, while the industry chart and FAQ give 3.22%; a high-ARPC prose value also differs from its chart. Keep the chart attribution and annual label intact, and request methodological clarification before using these figures for a financial commitment. Do not average conflicting versions or silently call them monthly. Rounded split figures may not exactly sum to the total.

For subcategories such as replenishment boxes, beauty products or wellness memberships, this broad ecommerce category does not establish a separate observed median. Likewise, its SaaS category is not an isolated enterprise or B2C sample. Use your own segmented cohorts and a genuinely matched dataset rather than inserting an unsupported subvertical range.

Subscription apps require a different comparison#

RevenueCat’s 2026 report draws on apps using its platform that meet activity thresholds. Its main measurement timeframe is 2025, with older cohorts for longer renewal calculations. This is an app-subscription population across supported ecosystems, not a representative survey of every enterprise contract, gym or subscription box.

Those are the report’s year-one retention figures by plan duration. They are not monthly churn rates, nor proof that annual pricing caused better retention. Cadence, acquisition, category and customer selection differ. Compare the relevant category and contract term before making an intervention decision.

RevenueCat’s retention-chart definition starts with paid subscriptions and measures paid periods reached among subscriptions with enough time to reach them. Resubscription creates a new cohort; incomplete periods need separate interpretation. That metric is different from a business’s customer-level cancellation counter or revenue churn.

Define your own unit and loss endpoint#

Choose the timing rule before comparing. A scheduled cancellation can leave access active until the paid term ends. A failed invoice can be in recovery before service ends. Record intent, entitlement, payment and cash separately. For this scorecard, count voluntary or payment-failure churn at the defined end-of-paid-service endpoint, with an unknown-reason category when evidence is missing.

First-month paid-cohort loss is measured over the first month after paid start. First-term nonrenewal is measured among subscriptions whose first renewal became due and has reached the chosen resolution endpoint. For annual plans that endpoint comes much later. State the grace/recovery policy and pending cases; do not compare a thirty-day cancellation intention with a resolved annual renewal loss.

Worked churn, recovery and revenue examples#

In an invented opening group of 1,000 paid customers, 15 end service voluntarily and 5 end after failed-payment recovery is exhausted. Customer churn is 20 / 1,000 = 2%; voluntary loss is 1.5% and involuntary loss 0.5%. Assume no customer has multiple subscriptions here and no other exits occur. These are illustrative counts, not a vertical benchmark.

Suppose 50 unique renewal invoices initially failed and 45 paid within the defined recovery window. Recovery is 45 / 50 = 90%. The remaining five count as involuntary customer loss only if the agreed service-end endpoint actually occurs. A first failure alone would have overstated that final loss tenfold. If some invoices remain pending, report them instead of treating them as resolved failures.

Separately, consider $100,000 opening normalized recurring revenue, $3,000 lost to churn, $1,000 contraction and $6,000 expansion from that opening base. Closing base revenue is $102,000, so NRR is 102%. Exclude $10,000 of new-customer revenue. This separate revenue example does not turn a 2% customer-loss rate into 2% revenue loss; customers can have different values.

Use compounding as a scenario, not measured annual churn#

For a closed starting cohort with constant monthly loss c, no reactivation and the same definition each month, implied twelve-month loss is 1 − (1 − c)^12. At 3% monthly, that is 30.62%; at 5%, 45.96%; at 10%, 71.76%. These are arithmetic scenarios, not observed annual vertical rates.

Do not compound a median across businesses as if it were one cohort’s constant monthly hazard. Do not apply this formula to a mixed population with changing acquisition, annual renewals or net revenue churn without establishing the assumptions. If monthly hazards vary, multiply each month’s survival factors for the same cohort instead.

A calendar-year renewal cohort is also different from “annualized” monthly loss. An annual-plan customer can remain entitled for ninety days without having faced a renewal. Early usage, refund and cancellation-intent data are useful, but cannot establish that a newly acquired annual cohort will renew.

Make the vertical change the diagnosis#

A vertical label helps identify relevant mechanisms; it does not prove one cause. Compare an acquisition promotion with a similar regular cohort before calling early exits a product failure. Separate insufficient use, fulfilment issues, completed learning goals and price concerns from collection failures. Test an intervention on the population it addresses.

Recovery affects the measured loss, not every customer’s intent#

Stripe’s retry documentation describes supported recovery settings and important exceptions. Some failures require a new payment method or action; scheduled attempts after a hard decline can remain unexecuted. Update the payment-method field actually used by the subscription, and respect mandates and cancellation. There is no universal retry schedule for every country and method.

Measure unique initially failed renewal invoices, recovered invoices and final service-ending losses after a fixed window. Track collection method, country and billing channel. Store-managed subscriptions need the store’s recovery process; a direct Stripe policy is not a control over every app-store renewal.

For a timeout with an unknown collection result, look up the original attempt before starting another. A new request key or processor cannot deduplicate an earlier charge. Authenticate and durably retain payment notifications, deduplicate their local effects and preserve actual payment or refund movement even if later local checks fail.

Set targets from comparable evidence and observed exposure#

• Write the paid unit, churn endpoint, observation window and customer segment beside every benchmark.
• Retain the source, chart version and publication/measurement dates; keep annual labels separate from monthly metrics.
• Resolve inconsistent definitions or published values before using a number in a committed forecast.
• Choose your own target and explain its basis: current cohort, margin, customer value and the measured change you can make.
• Review uncertainty and cohort size. One lost customer out of ten changes the rate by ten percentage points.

In ninety days, a new monthly cohort can provide several renewal opportunities and evidence about onboarding and recovery. A new annual cohort can provide early engagement and intent, while its renewal outcome remains unobserved. Review a mature annual cohort if you need annual renewal evidence now, or continue following the new cohort through its actual due window.

Keep the benchmark, scenario and target separate#

Use a published figure to understand its measured population, a transparent calculation to explore assumptions, and your own comparable cohorts to set a target. That makes vertical comparisons useful without turning a broad label or incomplete annual exposure into false precision.