Churn Rate: Why 2% a Month Is Not 24% a Year
Four choices sit under every churn rate anyone quotes — customers or dollars, which period, which cohort, and what goes in the denominator. Each one moves the number, as the figures mParticle, Funnel.io and AirDNA put on the record show.
On this page
In September 2019, Funnel.io CEO Fredrik Skantze put a churn number on the record: gross monthly churn “between 1.5% and 2% a month.” The annual figure reported alongside it was “24% gross annual churn at max” — twelve times the top of the monthly range, the conversion nearly everyone reaches for. Compound the same 2% instead and the year comes out at 21.5%. Two and a half points of churn appear and vanish depending on the arithmetic, inside a disclosure both sides believed was exact.
That is the mildest of the four ways a churn rate can mean different things. The other three move it further.
“Churn rate” is a family, not a metric. Four choices sit under every number anyone quotes: customers or dollars, which period, one cohort or all of them, and what goes in the denominator. Each one moves the answer, and a rate quoted without all four attached cannot be compared to anything — including your own number from last quarter.
Choice one: customers or dollars
Customer churn = customers lost in the period ÷ customers at the start of the periodAlso called logo churn. Every account counts once, whatever it pays.
Revenue churn = recurring revenue lost in the period ÷ recurring revenue at the start of the periodDollar-weighted: one large cancellation outweighs a crowd of small ones.
mParticle’s Michael Katz gave Nathan both halves of that split in November 2018, at a $36M run rate off 150 customers. Logo retention ran 95–97% per quarter, and the departures were concentrated in the smaller growth accounts. “We actually don’t see any churn in the enterprise,” Katz said, putting enterprise gross revenue churn under 5% a year against net revenue retention of 150%.
Compound that quarterly logo figure and mParticle was shedding roughly 12–18% of its customer count a year while the dollars that mattered barely moved. Both readings are true. A founder quoting the first sounds like a leaky SMB business; a founder quoting the second sounds like infrastructure. Which is why the unit belongs in the sentence every time.
It counts accounts, so it is the honest read on whether people keep using the thing — and the input to lifetime value, where dividing by churn is how expected customer life gets derived in the first place.
It counts dollars, so it is what boards, lenders and acquirers underwrite. Losing ten $100 accounts and losing one $10,000 account is the same logo-churn event and a completely different quarter.
Webflow ran both personalities inside one product in July 2018: $40-a-month designer plans churned 4–5% of customers monthly while the hosting line ran essentially zero, and blended net revenue retention still landed around 100%. ClickDimensions, by contrast, disclosed a 15% logo churn rate with no period attached at all — a figure that describes a healthy enterprise book or a dying one depending entirely on whether it means a month or a year.
The dollar side of the family has arithmetic of its own: gross against net, expansion netted back in, and the benchmark table of what founders actually disclose. Our revenue churn guide is the deep dive there. This one stays on the four measurement choices sitting underneath both numbers.
Choice two: the period
Annual churn = 1 − (1 − monthly churn)12Monthly rates compound; they do not add. Multiplying by twelve always overstates the loss, and the error grows with the rate.
Funnel.io is the mild case: 1.5–2% a month compounds to 16.6–21.5% a year, against the 18–24% straight-line multiplication produces. At the high end the gap stops being a rounding difference. AirDNA told Nathan in 2019 that it was losing 20% of its monthly recurring revenue every month. Multiply by twelve and you get a meaningless 240%. Compound it and you get a number that is merciless but real.
93%of a cohort gone within twelve months, at the 20% monthly churn AirDNA disclosed in 2019
The conversion runs the other way too, and that direction gets fumbled more often. Zwift reported a 70% annual retention rate — 30% annual churn — while noting that many of those users were pausing rather than permanently leaving. Divide 30 by twelve and you would call it 2.5% a month; the monthly rate that actually compounds to 30% a year is 2.9%. Trivial in isolation, load-bearing the moment it feeds a lifetime-value model.
On the record
Churn as founders actually disclose it
| Company | Disclosed rate | Period stated | Unit measured |
|---|---|---|---|
| Collibra (2018) | 3–4% | Annual | Not specified; net revenue expansion 120–125% |
| Funnel.io (Sep 2019) | 1.5–2% | Monthly | Gross revenue |
| mParticle (Nov 2018) | 95–97% retention | Quarterly | Customers |
| Tractian (2023) | 18% | Annual | Revenue, implied by 36% expansion and 118% net dollar retention |
| Webflow (Jul 2018) | 4–5% | Monthly | Customers, designer plans only |
| AirDNA (2019) | 20% | Monthly | Recurring revenue |
Founder-disclosed figures from the GetLatka archive, carrying the period and unit each founder attached — or, where none was attached, what the surrounding math implies.
Read down the last two columns and the argument makes itself. Six numbers between 1.5% and 20%, and not one of them is comparable to another without its period and its unit riding along.
Choice three: one cohort or all of them
A blended churn rate averages a customer’s first month with their fortieth. Since almost every SaaS churn curve is steepest at the start and then flattens hard, the blend describes nobody. The cleanest disclosures in the archive are all curves rather than points.
Churn ran 40% in the first 30 days, 20% between days 31 and 60, and 5% after 90. Most buyers treated short-term rental data as a one-time decision; the ones who did not, stayed.Scott Shatford, 2019
Self-serve signups discovering that SEO is a real job churned heavily in the opening months, while customers who survived past month nine churned less than 5% a year — a rate CEO Sarah Bird called enviable of an enterprise company.Sarah Bird, Moz
7% of customers churned in the first thirty days; after twelve active months, only 1% did. The same shape, in a self-serve funnel business at the other end of the price scale.
The fourth disclosure is about which cohort you sell into in the first place. Kyle Norton, chief revenue officer at Owner.com, told a room of a thousand software CEOs at SaaS Open in March 2024 that when he arrived, 30% of new customers were gone within ninety days of signing. His biz-ops team built a scoring model — estimated gross merchandise volume, a proxy for how profitable a restaurant would be on Owner.com’s usage-based pricing — and churn split cleanly along it. Before he started, 40 to 50% of deals sat in the high-churn cohort; by the time of the talk roughly 80% sat in the low-churn one, after the team raised the floor on what sales was allowed to close three times: $1,500, then $2,000, then $2,500.
A blended rate would have concealed all of that in both directions — first the problem, then the fix.
Choice four: the denominator
Two teams can agree on the unit, the period and the cohort and still publish different numbers, because the bottom of the fraction is a choice almost nobody makes out loud. The four that move it most:
- Start-of-period only — the strict version, and the sane default. Customers who joined and left inside the period never enter the fraction, which errs in the conservative direction.
- Start plus new signups — pads the denominator with customers who have had no chance to churn yet, so a company reports a lower rate simply for growing faster. This is the most common way a churn number flatters itself.
- Contracted versus collected revenue — annual contracts churn on their renewal date, not evenly across twelve months. Spread them and monthly revenue churn looks smooth and wrong; book them on the renewal date and it looks lumpy and right.
- What counts as leaving — a downgrade is revenue churn but not logo churn, and a pause is not a cancellation, as Zwift’s caveat shows. Two honest teams reading the same ledger produce different rates here.
The denominator also decides how a mixed book reads. AirDNA’s 500 enterprise accounts were under 8% of its 6,500 customers, and they supplied $300,000 of monthly recurring revenue against $420,000 from the other 6,000. Count logos and the enterprise book is a rounding error; count MRR and it is more than two-fifths of the company.
Reporting a rate that survives diligence
- Name the unit. “Customer churn” or “revenue churn,” never the bare word. If you have both, lead with revenue and keep logos beside it.
- Name the period, and compound between them. Monthly to annual is 1 − (1 − m)12, not m × 12 — and the reverse is not annual ÷ 12 either.
- Fix the denominator in writing. Start-of-period count, or start-of-period MRR, declared once and never quietly changed. A denominator switch is the cheapest way to manufacture an improving trend.
- Report by cohort and by segment. Signup month across at least twelve months, split by price band. If the curve flattens, that fact is worth more than the average hiding it.
- Show gross before net. Expansion belongs in the story, not in the churn line. Netting them together is how a leak gets reported as growth.
None of this is bookkeeping for its own sake. An undeclared churn rate hides the thing you most need to see, and it usually hides it behind a number that looks fine. Norton described exactly how that feels from the inside — sales hitting its target while the onboarding team watched the same customers walk back out:
“The sales number looked pretty good. And then the onboarding team going, like, what the hell, man? These deals suck. They’re all churning.”
Kyle Norton, CRO, Owner.com
SourcesFounder disclosures from the GetLatka archive: Funnel.io (Fredrik Skantze, September 2019), mParticle (Michael Katz, November 2018), Webflow (July 2018), AirDNA (Scott Shatford, 2019), Collibra (2018), Tractian (2023), Moz (Sarah Bird), ClickFunnels, Zwift and ClickDimensions; Owner.com figures from Kyle Norton’s SaaS Open talk, March 2024. Compounded and annualized conversions are our arithmetic applied to the disclosed rates.
