What is Churn Rate?

Churn rate is the share of customers or users who, in a period, stopped paying or stopped counting as active, relative to the base at the start of that period. It answers “who left.” The pair is retention rate (“who stayed”). With the same activity definition and the same cohort:

retention rate = 1 − churn rate

(or both add to 100% when written as percents). Different denominators break the identity. Do not add calendar-month account churn to app-install retention and treat them as complements.

Base formula on a customer base:

Churn rate = Customers lost in the period ÷ Customers at the start of the period

Example: 1,000 paying accounts on 1 March, 70 do not renew and do not return in March → 7% churn. April’s new subscribers do not belong in that ratio — they dilute churn by construction.

Logo, revenue, and what “left” means

Logo / customer churn counts heads: contract closed, subscription not renewed, account deleted. Revenue churn counts money: leavers × their ARPU. One whale can be a small logo churn and a large revenue churn.

Gross revenue churn is revenue lost to downgrades and leavers, ignoring expansion among those who stayed. Net revenue churn subtracts expansion (upgrades, add-ons). Net can be negative: remaining accounts pay more than leavers took. For LTV and payback on CAC, read both: logos can hold while the bill shrinks — that is still economic churn.

Voluntary churn (cancel) and involuntary churn (expired card, chargeback, ban) should not share one KPI. The second is billing and fraud; the first is product and price. In subscriptions and rebill offers, involuntary churn is often larger than the “cancel reason” field implies.

Churn and retention are one coin

Retention rate for the same period T on the same base:

Retention rate = Still-active at period end ÷ Customers at period start

Hence churn = 1 − retention if “still active” = “not lost” under one rule. In apps that is often false. D7 retention is the share of an install cohort that opens the app on day 7. “D7 churn” as 1 − D7 is only a label on that same cohort; it is not SaaS paying-base churn for a calendar week.

Calendar churn (everyone active on the 1st) and cohort churn (January acquirers gone by March) are different reports. Buying economics needs the second: that is cohort analysis. The calendar view is a board slide and lies when the base is growing or shrinking fast.

In mobile games and utilities, “left” is often N days of inactivity, not uninstall. The app can stay installed and already be churned in the LTV model.

Why churn sits in unit economics

A simplification for a stable subscription:

LTV ≈ ARPU ÷ churn

(if churn is the period’s fraction, ARPU is for the same period, margin = 1). At 10% monthly churn and $20 ARPU, crude LTV is ~$200 before margin and tax. A 2 pp rise in churn hits LTV harder than the “small” percent suggests: the denominator shrank. That is why CAC cannot be compared with a day-0 payout alone in subscription, gambling revshare, or nutra rebill.

Payback lengthens when early churn is high: the user never reaches the second or third charge. In affiliate that shows up as rebill falling after trial. A buyer who scales on CPA first purchase / FTD parks churn risk with the advertiser — until the network cuts payout or moves to hybrid.

Churn also sets the ceiling on bid: the same CPA on a leakier cohort (other GEO, incent, “freebie” creative) implies a lower allowable CAC. Cutting CPC does not fix that if cohort quality is unchanged.

Where it is counted in ads and product

SaaS and subscriptions are the textbook case: logo + net revenue. Banks and telecom count contract churn. E-commerce with repeats more often says “did not buy in 90 days” than “churn” in the narrow sense; the formula is the same, the inactivity threshold is a convention.

In e-commerce, churn without a silence window is undefined: there is no monthly charge. Set a window (60/90/180 days without an order) and count the cohort share that exited it. Otherwise a one-off purchase is misread as loyalty.

Apps: uninstall, inactivity, and subscription cancel are three churns. UA reads the retention curve (D1/D7/D30 are sections of the retention article, not separate terms) and, separately, subscription churn after trial.

Do not confuse with bounce rate: a single-page session is not customer churn.

Limits and mistakes

A denominator of “anyone who appeared in the month” understates churn versus “base on the 1st.”
Reactivation: a return in the same period either cancels churn or double-counts — write the rule first.
Benchmarking industries without the same period (month vs year) and without logo vs revenue.
“Cut churn with a discount for everyone” holds heads and cuts margin — LTV can fall.

Churn describes base loss. The “buy / don’t buy more traffic” decision still closes on LTV, CAC, and — if the fight is whether media made “our” users leave faster — incrementality, not the churn percent on a slide.