Skip to content

Churn

The fraction of customers (logo churn) or revenue (revenue churn) lost in a period. Simple to state, unusually easy to compute three different ways.

The idea

Churn splits into two families:

Logo churn counts customers. A customer churns when their last active subscription ends — active subscription count reaching zero, not merely one subscription among several being cancelled.

Revenue churn counts MRR lost. A customer downgrading from $500 to $50 is not a logo churn at all, but is a substantial revenue event.

The two diverge sharply by segment mix: losing many small customers and losing one large one look identical in logo churn and nothing alike in revenue churn.

Voluntary vs. involuntary

Involuntary churn — failed payments rather than a decision to leave — accounts for 20–40% of total subscription churn, and smart retry logic recovers 60–80% of failed charges. Stripe reports ~41% of failed invoices recovered via ML retries.

This distinction is the entire premise of Baremetrics Recover and ProfitWell Retain. Tidemill treats dunning as an explicit non-goal — billing-engine territory — but the analytical distinction between voluntary and involuntary churn remains meaningful and is not currently broken out.

How Tidemill computes it

The scoping rule does the real work: only customers active at period start (\(C_{\text{start}}\)) can appear in the numerator. Customers who both join and churn inside the same period are excluded — otherwise a burst of trial signups would manufacture churn out of nothing.

Revenue churn is gross: churn MRR only, excluding contraction and expansion, since those are visible separately in the MRR waterfall.

Pure-usage customers churn too. A metered-only customer has zero subscription MRR but accrues usage MRR once their first usage invoice is paid; on cancellation the churn amount is subscription_mrr + usage_mrr, so neither component is lost.

Authoritative formulas: definitions.md — Churn. Implementation: tidemill/metrics/churn/.

Contested ground

Churn is the metric where Tidemill most visibly diverges from ChartMogul, and each divergence is documented rather than buried:

Question Tidemill ChartMogul
Churn-then-reactivate in one period Counts as churn Netted out
Revenue churn definition Gross churn only Gross (incl. contraction) and net variants
Recognition timing At status change / period end Configurable across three options
Incomplete periods Actuals only Extrapolated to full-period rate

None of these has a correct answer. The point is that a user reconciling Tidemill against a previous tool can find the difference in a table instead of guessing — see metric transparency.

Sources