Quick definition: Churn rate is the percentage of a defined customer, account, user, or recurring-revenue population that stops an agreed relationship during a stated period.
What is churn rate?
Churn measures loss, but the lost thing must be named. Customer churn counts customers that cancel or fail to renew. Account churn counts organizations. User churn may mean formerly active people who become inactive. Revenue churn measures recurring revenue lost, while net revenue churn also incorporates expansion or contraction among retained customers. These measures can move differently: a company may lose many small customers while retaining most revenue, or lose one large account while customer churn remains low.
A churn definition needs a population, a loss event, a period, and a rule for temporary inactivity. “Monthly customer churn” might mean customers active at the beginning of a calendar month who cancel before month end, divided by customers active at the start. For a non-contract product, “inactive” needs a behavior threshold based on a typical usage cycle. A user who has not returned for seven days is not necessarily churned from a product normally used quarterly.
Churn is related to retention but not always simply one minus retention. Cohort retention asks what proportion of an acquisition cohort remains active at a later time. Period churn commonly starts with an existing base and measures losses during a period. New customers added during the month should not normally enter a beginning-base churn denominator, because they had no full risk exposure during that period.
Churn-rate formula and denominator choices
A common gross customer-churn formula is:
customer churn rate = customers active at period start who churn during period / customers active at period start × 100
For recurring revenue, use a consistent revenue basis:
gross revenue churn = recurring revenue lost from starting customers / starting recurring revenue × 100
Net revenue retention instead begins with starting recurring revenue, subtracts downgrades and churn, adds expansions, then divides by starting recurring revenue. It answers a different question and can exceed 100%. Never label net revenue retention as customer churn or hide expansion inside a gross-churn claim.
| Measure | Numerator | Denominator |
|---|---|---|
| Logo churn | Starting accounts lost | Accounts active at start |
| Customer churn | Starting customers lost | Customers active at start |
| Gross revenue churn | Recurring revenue lost or contracted | Starting recurring revenue |
| Inactivity churn | Previously active units crossing inactivity threshold | Units at risk under the rule |
Churn in A/B testing
Churn is an important guardrail for pricing, paywall, onboarding, engagement, reliability, and feature-removal tests. It is often too slow and rare to be the sole primary outcome for a short experiment, but a treatment that raises immediate conversion while increasing cancellations can be harmful. Predefine a follow-up window that covers a credible opportunity to cancel, renew, or become inactive. Avoid calling a seven-day read “monthly churn” before the cohort has matured.
Analyze users or accounts by assigned variant, including those who did not engage with the changed feature. Excluding people who never opened an email or never completed setup can bias the estimate when treatment affects engagement. Use a stable assignment identity, and consider account-level analysis when multiple users share a plan or can influence each other’s cancellation.
Churn results should include counts, rates, absolute differences, uncertainty, and the exposure time available to each arm. When churn is rare, an observed zero difference is not proof that a treatment is harmless; the interval may still permit material harm. Plan sample size and duration for the smallest meaningful retention effect. Our sample-size guide and test-duration guide explain the planning trade-offs.
Worked churn calculation
A B2B platform tests a new plan-change screen. It defines 90-day customer churn as a starting paid account that cancels all paid service within 90 days of assignment. Control has 8,000 eligible accounts and 320 cancellations; treatment has 7,950 accounts and 358 cancellations.
control churn = 320 / 8,000 = 4.00%treatment churn = 358 / 7,950 = 4.50%absolute change = +0.50 percentage pointsrelative change = 12.5% more churn
The treatment may reduce support volume now, but the 0.5-point increase could matter commercially. The team checks cancellation reasons, downgrades, involuntary payment failures, reactivations, account size, and whether support agents routed treatment customers differently. It does not subtract reactivations unless that rule was part of the pre-agreed definition; cancellation churn and net active-base change are distinct measures.
Churn data-quality caveats
Billing status is not always customer intent. An account can be canceled voluntarily, lost after a failed payment, suspended for fraud, migrated to another contract, or merged into a parent account. Keep these statuses distinct and specify which are churn for each metric. A customer who downgrades to a free tier may be customer churn for paid-service analysis but still be retained as a product user.
Identity and hierarchy errors distort both sides of the rate. One corporate customer can own many workspaces; mergers and reseller arrangements can create apparent losses. Use a stable customer or billing-account key and document account merges. For usage churn, delayed telemetry, bot traffic, seasonal use, and feature outages can mimic inactivity. Do not infer churn from missing events until data completeness is checked.
Calendar alignment matters. A February monthly churn calculation has fewer days than January, and annual plans do not face a renewal opportunity every month. Use cohort and contract-period views where they better represent risk, and make any annualization assumption explicit.
Common churn-rate mistakes
- Using all current customers as the denominator: new additions should not dilute losses from the beginning base.
- Mixing customer and revenue churn: counts and dollars describe different exposure.
- Calling a downgrade a cancellation without definition: revenue and logo outcomes differ.
- Ignoring involuntary churn: payment recovery may be a separate operational problem.
- Comparing immature cohorts: newer customers have less cancellation opportunity.
- Using short inactivity thresholds: normal usage rhythms can look like attrition.
Frequently asked questions
What is the difference between churn and retention?
Churn measures losses from a population at risk; retention measures the share remaining. Definitions and time windows determine whether they are exact complements.
Should new customers be included in monthly churn?
Usually not in a beginning-of-period churn denominator. Report them separately or use a cohort retention measure.
Can revenue churn be negative?
Gross revenue churn cannot be negative under a loss-only definition. Net revenue churn can be negative if expansion exceeds losses.
Is an inactive user churned?
Only if an inactivity rule appropriate to the product and usage cycle says so. Distinguish temporary absence from a durable loss.
Why is churn hard to test?
It may be delayed and rare, requiring longer follow-up and larger samples than a near-term engagement metric.
Summary
Churn rate is loss among a precisely defined population at risk. Its denominator, loss event, time period, and treatment of downgrades, payment failures, reactivations, and account hierarchy must be explicit. In experiments, use churn as a mature outcome or guardrail and interpret it with equal follow-up, uncertainty, and customer-level context.
Sources
- U.S. Securities and Exchange Commission: Annual-report filing example
- Recurly: SaaS metrics guide
- U.S. Census Bureau: Business Dynamics Statistics FAQ