How to Calculate Churn Rate: 3 Formulas, 3 Answers

Customer churn is the share of customers you lost during a period, divided by how many you had at the start, times 100. That's the definition, and it's the one everyone agrees on. What nobody warns you about is that "the churn rate formula" is really several standard formulas, and on the same underlying data they hand you different numbers — some flatter you, some panic you.

The short version

Pick your formula based on how fast your customer base is growing, and pick your unit — logos, users, or dollars — before you touch a spreadsheet. Flat base? The simple formula is fine. If you're growing more than about 5% a month, simple churn will lie to you, and you want a midpoint or cohort calculation instead. Everything below is the elaboration on those two decisions.

The canonical formula, and why the denominator is the whole fight

Start with the plain version. Customer (logo) churn = customers lost during the period ÷ customer count at the start of the period × 100. Lose 3 out of 100 in a month, and that's 3% monthly churn.

The part people skip is the denominator. It is your customer count at the beginning of the window, before any losses and before any new signups. New acquisitions are deliberately left out. As the analytics reference site Count puts it, "the denominator is your customer count at the beginning of the measurement period, before any churn or new acquisitions occurred. This baseline ensures you're measuring churn against a consistent starting point."

Hold onto that word baseline. Almost every argument about churn rate is really an argument about what you're allowed to put in the denominator.

The three formulas, one dataset

Let me make this concrete with a base small enough to check by hand. Picture a 3-week window — treat it like a compressed month — that opens with 12 customers. During the window, 4 new customers sign up and 2 of the original 12 cancel. So we end with 14 customers, having lost 2.

Same story, three answers.

Formula 1 — Simple (start-of-period denominator)

Lost ÷ start-of-period count. That's 2 ÷ 12 = 16.7%. Clean, defensible, and the one most tools show by default.

Formula 2 — Midpoint / adjusted (average customer count)

Here you divide by the average of your start and end counts to account for the base moving under you. Paddle's guidance is to "take the midpoint of the customer count for the month, rather than using its value on the 1st of the month." Our midpoint is (12 + 14) ÷ 2 = 13. So 2 ÷ 13 = 15.4%.

Formula 3 — Cohort-based (track the start-cohort only)

Ignore signups entirely and follow only the 12 customers you had on day one. How many of those specific 12 were gone by the end? Two. So 2 ÷ 12 = 16.7% — the same as simple here, but the number now means something narrower. It's the survival of a fixed group, and it can't be moved by whatever the sales team booked mid-month.

Formula	Churn on our dataset	When to use
Simple (start-of-period)	16.7%	Flat or slow-growing base; quick internal read
Midpoint / adjusted	15.4%	Meaningful mid-period growth to normalize for
Cohort (start-cohort only)	16.7%	Board reporting; isolating behavior from growth

Same 2 losses, same 12 starters. The spread between them isn't measurement noise. It's three different questions wearing the same name, and once a fast-growing month enters the picture, that spread widens fast.

The denominator problem: how growth fakes your churn

Now scale up and watch the distortion get ugly. Paddle's engineering write-up gives the cleanest example I've seen: "In August, 125 churned customers were added to the numerator, but the 5,000 new customers that joined in August didn't get added to the denominator — which means that the churn rate is artificially high."

Read that mechanically. The customers who cancel this month land in your numerator immediately. The thousands who joined this month sit outside the start-of-period denominator, because they weren't there on day one. So a great growth month — the kind you'd celebrate — makes simple churn read worse, because the numerator swells against a denominator that never got the good news.

The same lever runs in reverse, and this is where I get quietly annoyed. Slide this month's new signups into the starting count and you inflate the denominator, and the churn rate drops. ZoomInfo names it flatly: "including new customers in the denominator artificially lowers your churn rate by inflating the starting customer count." Some teams do this by accident. Some do it because a lower number is easier to present. Same math, different intent, both wrong.

A decision rule: which formula at which growth rate

Here's the ladder I actually use.

If your base is flat or growing slowly — say under 5% a month — simple churn is fine, and reaching for anything fancier just adds explanation overhead. Once monthly growth crosses roughly 5%, switch to the midpoint formula or a cohort so the incoming crowd stops distorting the read. And when you're reporting to a board, or anyone who'll make a decision on the number, use a cohort. Growth physically cannot hide behavior inside a fixed group.

Most vendors default to simple churn. I'll disagree with the industry here and say it plainly: for a growth-stage company, cohort churn should be your default, not simple. Simple churn's convenience is exactly what makes it dangerous during the months you most want a clean signal. If you want the underlying mechanics of tracking a fixed group over time, the difference between acquisition and behavioral cohorts is worth reading before you pick your cohort definition.

Logo vs. user vs. revenue churn — don't average across them

A churn percentage means nothing until you say of what. Logos, individual users, and dollars are three different units, and they routinely disagree.

Say you lose two customers this month, and they happen to be two of your biggest accounts. Your logo churn looks mild. Your revenue churn is a bloodbath. Count states the trap directly: "a few high-value customers churning might create low customer churn but significant revenue churn." Report only the logo number in that situation and you've technically told the truth while hiding the actual story.

Revenue churn itself comes in two flavors. Gross revenue churn counts only the money you lost. Net revenue churn subtracts expansion — upsells and upgrades from customers who stayed — from those losses. And net can go negative. As ChartMogul explains, "if the monthly recurring revenue (MRR) gained from existing customers (Expansion + Reactivation) exceeds the MRR lost (Churn + Contraction), your net MRR churn rate will be negative." Negative net churn means your existing base grows on its own, before you acquire a single new customer. It's one of the strongest signals a subscription business can show, and it's invisible if you only ever count logos.

Don't annualize by multiplying

A recurring mistake, and an easy one: taking 5% monthly churn and calling it 60% a year. Churn compounds on a shrinking base, so you can't just multiply. Wall Street Prep gives the correct conversion: Annual churn = 1 − (1 − monthly churn)^12.

Run 5% through it: 1 − (0.95)^12 ≈ 0.46, or about 46% annually — not 60%. The 14-point gap is entirely from ignoring compounding, and it's the difference between a plausible retention story and a fictional one.

What's a normal churn rate?

For a reality check, Recurly's subscription network research (updated July 2026) is the cleanest primary source I'd point to. Their numbers put "a median overall annual churn rate of 3.04% for software businesses, with top-quartile performers at 1.78% or below."

A few caveats before you screenshot that. It's annual, not monthly, so don't compare it to your monthly dashboard without converting. It also blends voluntary churn (people who chose to leave) with involuntary churn (failed payments), and those need different fixes. If you want more benchmarks by category, our 2026 product analytics benchmarks roundup breaks them down further.

Where this goes wrong

The failures cluster, and I've watched every one of these happen on a real dashboard.

The first is stuffing new customers into the starting count, which inflates the denominator and quietly drags churn down — the ZoomInfo mistake above, and the most common one by a wide margin. Second is comparing simple churn across months of very different growth without noticing the denominator moved beneath you. A "rising" churn rate is sometimes just a great acquisition month, per the Paddle mechanics. Third is walking into a board meeting with logo churn when the two accounts you lost were your biggest. Revenue churn was the real story, and you buried it.

The fourth is a sourcing problem. A lot of churn "benchmarks" floating around are secondary blogs recirculating survey stats nobody can trace, sometimes with quotes misattributed to experts who never said them. Go to the primary vendor research — Recurly's own numbers, a tool's engineering docs — and skip the telephone game. While you're auditing, check that your churn isn't hiding a mix shift the way Simpson's paradox hides trends inside aggregates.

Doing this in practice

Building a monthly churn cohort is, mechanically, a one-question job in any modern product-analytics tool. You can ask a chat-first platform like Kixo to build the cohort in plain language and get the chart back, and that's genuinely the easy part now. The number is still only as honest as the formula and the unit you chose before you asked.

So the real takeaway isn't a tool or even a formula. It's the decision you make first. Pick your denominator on purpose, pick logos versus dollars on purpose, and default to cohort once you're growing. Get those two right and the different answers stop being a trap. They start being honest views of the same customers leaving.

FAQ

What is the churn rate formula? Customers lost during a period ÷ customers at the start of the period × 100. New customers are excluded from the denominator to keep a consistent baseline.

Why do I get different churn numbers from different formulas? Because they use different denominators. Simple uses the day-one count, midpoint uses the average across the window, and cohort tracks only the starting group — so growth moves each one differently.

How do I convert monthly churn to annual? Use 1 − (1 − monthly churn)^12. Don't multiply by 12; that ignores compounding and overstates the annual figure badly.

What's a good annual churn rate for software? Recurly's network research puts the median around 3.04%, with top-quartile companies at 1.78% or below.