Churn Rate

Churn is a family of metrics, not one number

Four independent choices sit underneath every churn rate, and each is defensible on its own terms:

• Unit — customers or revenue. Logo churn counts accounts; revenue churn counts money. They diverge whenever your small customers behave differently from your large ones, which is always.
• Gross or net — whether expansion from surviving customers is allowed to offset losses. Gross churn can only be positive; net churn can go below zero.
• Denominator — customers at the start of the period, the average across it, or the count at the end.
• Scope — whether trials, free plans, delinquent accounts and same-month signups are in the base at all.

Multiply those out and a single month of billing data supports well over a dozen honest churn rates. This is why cross-company churn comparisons are close to worthless unless both sides publish their definition, and why the first useful question about any churn figure is not "is it good?" but "what is in the denominator?"

The denominator problem

The strictest definition is a closed-cohort one: take the customers who existed on day one, ask how many of them were gone by day thirty, and ignore everyone acquired in between. It is the only version that measures retention of a fixed population, and it is the version that supports cohort analysis.

Most dashboards do something looser — all cancellations in the month, over the starting count. That is fine and widely used, but it mixes populations: a customer who signed up on the 4th and cancelled on the 20th appears in the numerator having never been in the denominator. In a fast-growing business with a leaky first month, this systematically overstates churn. Using an average of the start and end counts partially compensates, at the cost of a denominator that no longer corresponds to any real moment in time.

None of these is wrong. Switching between them silently is wrong, and it happens whenever a company changes analytics tools.

When a cancellation counts

A customer who cancels on 3 March with a subscription paid through 31 March has not stopped paying you. Two conventions exist. Cancellation-date churn records the loss on the 3rd: it is the earliest honest signal, and it is what you want for detecting a deteriorating month early. Expiry-date churn records it on the 31st: it matches the revenue you actually lose and reconciles cleanly to MRR.

Pick one per metric and be consistent. The specific failure to avoid is booking the customer loss on the cancellation date and the revenue loss on the expiry date, which puts customer churn and revenue churn permanently out of step and makes the two impossible to reconcile.

Annual plans break monthly churn

A customer on an annual contract cannot cancel in most months — they only get the opportunity once a year. Run a standard 30-day churn calculation over a base containing annual subscriptions and you will report an artificially low rate for eleven months and a spike in the twelfth, driven entirely by when contracts happen to renew.

The standard correction is a 365-day lookback for annual plans: measure how many customers who were active a year ago are still active now. Baremetrics applies exactly this rule, and it is a documented source of surprise for their users, because the churn number stops responding to recent events. Better practice is to report monthly and annual cohorts as separate series rather than blending them into one rate that describes neither.

Trials, free plans and delinquents

Three population decisions change churn more than most people expect. Trials should normally be excluded from the base entirely — a trial that ends without converting is a failed trial conversion, not churn, and counting it as churn makes a successful top-of-funnel campaign look like a retention crisis. Free plans likewise: a downgrade to free is real revenue loss and belongs in revenue churn, but whether the account is a churned customer depends on whether you consider free users customers at all. Delinquent accounts — cards failing, invoices unpaid — sit in limbo until a policy decides. The common convention is that delinquency past 30 days counts as churn; the important part is that this is a configurable window, not a law, and moving it from 30 days to 60 visibly changes your reported churn without a single customer behaving differently. See involuntary churn.

Annualising, and why multiplying by twelve is wrong

Monthly churn compounds. A 2% monthly rate does not produce 24% annual churn; it produces 1 − (0.98 ^ 12), or about 21.5%. The error looks small at 2% and becomes serious quickly: 5% monthly is 46% annually, not 60%. Multiplying by twelve also produces impossible results above 8.3% monthly, where the linear version exceeds 100% and the compounded version is 64%.

The same compounding is what makes small churn improvements worth so much. Cutting monthly churn from 3% to 2.5% raises the implied average customer lifetime from 33 months to 40, and every metric derived from lifetime — LTV, LTV:CAC — moves with it.