What it means
Churn rate measures leakage. Over some period, some share of what you had at the start is gone by the end, and churn is that share expressed as a percentage. The apparent simplicity is deceptive, because "what you had" can mean customers or revenue, "gone" can mean cancelled or merely reduced, and the period boundary can be drawn in several defensible places. Almost every disagreement about a churn number is a disagreement about one of those three choices rather than about the data.
There are four numbers worth computing, and mature teams publish all four because each answers a different question. Customer churn asks how many relationships you lost. Gross revenue churn asks how much money walked out, counting downgrades as well as cancellations. Net revenue churn nets expansion against that loss. Net revenue retention is the same figure stated the friendly way round.
The relationship between the first two is diagnostic in itself. When revenue churn runs higher than customer churn, the accounts leaving are larger than your average, which points at your ideal customer profile or at a problem that only appears at scale. When it runs lower, you are losing small accounts, which is usually an onboarding or pricing issue at the entry tier.
The formulas, written out
Customer churn = (customers lost during the period / customers at the start of the period) x 100
Gross revenue churn = ((churned MRR + contraction MRR) / MRR at the start of the period) x 100
Net revenue churn = ((churned MRR + contraction MRR - expansion MRR) / MRR at the start of the period) x 100
Net revenue retention = 100 - net revenue churn
Variables:
- Churned MRR: recurring revenue from customers who cancelled entirely during the period.
- Contraction MRR: recurring revenue lost from customers who stayed but shrank, through downgrades, seat reductions or discounts.
- Expansion MRR: recurring revenue gained from customers who were already in the starting base. New customers are excluded, which is the entire point.
- MRR at the start: the recurring revenue of the base as it stood on day one of the period. Frozen, not recomputed at the end.
To convert a monthly rate into an annual one:
Annual churn = 1 - (1 - monthly churn)^12
And the reason churn dominates unit economics:
Expected customer lifetime in months = 1 / monthly customer churn
A worked example, end to end
A subscription business starts the month with 1,240 customers and 74,400 of MRR, an average of 60 per customer. During the month:
- 58 customers cancel, carrying 4,180 of MRR.
- Existing customers downgrade or reduce seats, losing 910 of MRR without cancelling.
- Existing customers upgrade or add seats, adding 3,640 of MRR.
- 143 new customers arrive with 8,730 of MRR. This number appears in no churn calculation at all.
Customer churn = 58 / 1,240 = 4.68%
Gross revenue churn = (4,180 + 910) / 74,400 = 5,090 / 74,400 = 6.84%
Net revenue churn = (5,090 - 3,640) / 74,400 = 1,450 / 74,400 = 1.95%
Net revenue retention = 100 - 1.95 = 98.05%
Four numbers from one month, ranging from 1.95% to 6.84%, all correct and all describing something different. A company that reports only net revenue churn looks like it is losing almost nothing. A company that reports only gross revenue churn looks like it is bleeding. Both are the same business.
The most informative line is one nobody computes. The 58 cancelled customers carried 4,180 of MRR, which is 4,180 / 58 = 72.07 each, against a base average of 60. Pure cancellation churn as a share of revenue is 4,180 / 74,400 = 5.62%, against customer churn of 4.68%. The accounts leaving are twenty percent larger than the typical account. That is not a general retention problem, it is a specific problem with a specific segment, and a broad re-engagement campaign will not touch it.
Annualising, correctly:
Annual churn = 1 - (1 - 0.0468)^12 = 1 - 0.9532^12 = 1 - 0.5626 = 43.74%
And the lifetime implication: 1 / 0.0468 = 21.4 months, so at an average 60 per month the expected lifetime revenue per customer is about 1,282. Improve monthly churn to 2.5% and the lifetime becomes 40 months and the figure becomes 2,400. Roughly two points of monthly churn nearly doubles the value of every customer you acquire, which is why churn beats almost every acquisition optimisation available.
Four ways the number goes wrong
Annualising by multiplication. 4.68 x 12 = 56.2% against a correct 43.7%. The multiplication assumes the base stays the same size all year, when in fact each month churns a smaller absolute number of customers than the last. Twelve and a half percentage points of error flows straight into lifetime value, payback period and every acquisition decision built on them.
Putting new customers in the denominator. 58 / (1,240 + 143) = 4.19% instead of 4.68%. It looks like a rounding decision and it is not: it makes churn appear to improve whenever growth accelerates. Add 300 new customers next month instead of 143, lose the same 58 from the same starting base, and the reported rate falls to 58 / 1,540 = 3.77% while retention has not changed at all. Any metric that improves when you spend more on acquisition is not measuring retention.
Counting only full cancellations. A customer who drops from 120 a month to 40 on the twelfth of the month has not churned, and has taken 80 of MRR with them. In this month contraction totalled 910, which is 910 / 74,400 = 1.22 percentage points of revenue erosion invisible to any customer-count metric. In products sold by seat, contraction is frequently larger than cancellation and it almost always arrives first, which makes it the earlier warning of the two.
Counting failed payments as churn immediately. Of the 58 cancellations, 21 were involuntary: expired cards and declined charges. Nine of those recovered inside the fourteen-day dunning window. Count churn at the moment the payment fails and the month reads 58 lost, 4.68%. Count it at the end of the recovery window and it reads 49, or 3.95%. Both are defensible, they differ by 0.73 points, and the choice must be written down and kept, since switching between them mid-year creates a trend that never happened. Report involuntary churn separately in either case, because its remedy is a card updater and a retry schedule rather than anything to do with the product.
How churn is actually derived from CRM and billing data
The calculation reads three tables and a policy document:
- Subscription records: start date, end date, plan, billing term, and MRR normalised to a monthly figure regardless of whether the customer pays monthly or annually.
- Plan-change events: each with an effective date, a previous MRR and a new MRR. Without these, contraction and expansion cannot be separated from each other or from new business.
- Payment and dunning state: failed charges, retry attempts, recovery, and the point at which a subscription is considered ended.
- Cancellation reason codes, from a closed list, which is the only part of the whole exercise that tells you what to do about it.
Edge cases that materially change the answer:
- Mid-period plan changes. A downgrade effective on the twelfth affects cash this month and MRR from the next billing date. Movement accounting normally uses the plan value at period end, so the month's contraction figure and the month's invoice total will not agree. Both are right; document which one the churn report uses.
- Reactivations inside the period. A customer who cancels on the third and resubscribes on the twentieth has churned and been won back. Netting them out silently understates both churn and new business, and hides a recovery process that may be working well.
- Downgrade to a free plan. That is 100% revenue churn and 0% logo churn. Decide which report each one belongs in, because a free-tier product can otherwise show excellent customer retention alongside collapsing revenue.
- One company, several subscriptions. Three subsidiaries with separate contracts count as three in a subscription-level rate and one in an account-level rate. Enterprise reporting usually wants the account view, and the two rates can differ substantially in a base with many multi-entity customers.
- Trials that never converted. Not churn. They were never customers, and including them inflates the rate while mixing an acquisition problem into a retention metric.
- Period boundaries and timezones. A cancellation at 23:50 on the 31st belongs to this month in one timezone and the next in another. Fix a reporting timezone once and apply it to every date in the pipeline, or two reports built from the same database will disagree by a handful of customers every month.
- Refunds and credits. A full refund after cancellation affects revenue but not the customer count, and a retention credit reduces MRR without any plan change. Decide whether credits count as contraction.
Why it matters
Churn compounds, and compounding is unintuitive. The difference between 2% and 4% monthly churn does not feel like much in a dashboard, and it is the difference between an average customer lifetime of fifty months and one of twenty-five. Every acquisition decision, every payback calculation and every valuation runs through that number, which is why a two-point improvement in retention typically beats a twenty-percent improvement in conversion.
It is also the honest scoreboard for everything upstream. A good ICP shows up as lower churn in the high-fit cohort. Good support shows up as lower churn among accounts with recent tickets. A misleading sales pitch shows up as a spike in churn at month three or four. Almost every quality problem in a business eventually arrives at this metric, which is why segmenting it matters more than the headline value.
Common mistakes
- Reporting one number. Customer, gross revenue and net revenue churn tell three different stories. Publish all three.
- Never segmenting. A blended rate across plans, tenures and segments hides every actionable pattern in the data.
- Ignoring the first ninety days. Early churn is usually an onboarding failure and behaves nothing like late churn. Report them separately.
- No reason codes. Without them, churn is a number you can watch and cannot change.
- Comparing to published benchmarks. Rates depend so heavily on price point, contract length and segment that cross-company comparison is close to meaningless. Compare against your own trend and your own cohorts.
Related concepts
- ICP: the profile whose accuracy churn is the final test of.
- NPS: the early sentiment signal that often precedes a churn event.
- CSAT: the interaction-level score that accumulates into a retention outcome.
- First response time: one of the few operational levers that visibly moves retention.
- Sales pipeline: renewals and expansions belong in a pipeline of their own.
- Conversion funnel: the acquisition side of the same equation.
How Pinlyx handles it
Pinlyx records plan changes as dated events rather than overwriting a plan field, so contraction and expansion can be separated from cancellation and from new business. Cancellation reasons come from a configurable closed list, involuntary churn from failed payments is tracked through its recovery window instead of being counted on the first decline, and churn can be segmented by plan, tenure, source and account-fit score so the headline rate is a starting point rather than the whole report.