Churn rate = lost in the period ÷ what you started the period with
Simple, and almost every reported churn number is wrong in one of three places: what "lost" counts, what the denominator holds, and whether the two describe the same population.
Logo churn and revenue churn
Logo churn counts customers. Ten of two hundred accounts left, so 5%.
Revenue churn counts money. Those ten carried $40,000 of $2,000,000, so 2%.
Both are true and they say opposite things about the same quarter. A business losing many small accounts has high logo churn and low revenue churn, which looks fine on the revenue line while the customer base erodes underneath. The reverse, losing one enterprise account, barely moves logo churn and can be the worst quarter in a year.
Report both. A single churn number is a choice about which problem to hide.
The denominator decides the number
Start-of-period is the convention: customers or revenue at day one, before anything added during the period.
The common error is including mid-period additions in the denominator. A growing business then reports falling churn without retaining anyone better, because the base inflated. When growth stalls, the same churn rate suddenly appears to spike, and nobody can explain why.
Fix the denominator, fix the series.
Annualising is not multiplication
Monthly churn of 2% is not 24% annually. Churn compounds on a shrinking base:
(1 - 0.02)^12 = 0.785, so about 21.5% annual
The gap widens as the rate rises, and at high monthly churn the naive multiplication is badly wrong. State whether a figure is monthly or annual, and which method converted it.
Read it against expansion
Gross churn alone describes leakage. It does not describe whether the base is growing, because expansion from existing customers can exceed it.
That is what net dollar retention measures, and Stripe's framing is the useful one: net revenue retention combines churn, contraction, and expansion into one figure for the existing base. Report gross churn and NDR together. Gross churn says what is leaking; NDR says whether the bucket is filling anyway.
A worked example
Start the quarter with 400 customers and $1,600,000 of ARR. During the quarter, 12 customers leave carrying $38,000, 30 new customers arrive, and existing customers expand by $52,000.
Logo churn: 12 ÷ 400 = 3.0%
Gross revenue churn: $38,000 ÷ $1,600,000 = 2.4%
Net revenue movement: (−38,000 + 52,000) ÷ 1,600,000 = +0.9%
The 30 new customers appear in none of these. They belong to acquisition, not retention, and putting them in the denominator would report 12 ÷ 430 = 2.8% and call the improvement retention.
The traps
- One churn number with no label for logo or revenue.
- Mid-period additions in the denominator.
- Annualising by multiplying.
- Excluding downgrades, which are contraction and belong in revenue churn.
- Counting a churned account that later returns as never having left.
- Comparing to a benchmark without checking its segment. Self-serve and
enterprise churn are not the same metric in practice.
Where this sits
Churn is Measurement in the Tenbound Pipeline Architecture Standard, and it is where a Market problem shows up last. Accounts acquired outside the ICP churn at a rate no customer success motion can fix, which is why a churn spike is often a targeting finding arriving nine months late.