Churn and retention definitions that stick

Someone says “churn is fine” while someone else says “we are bleeding logos.” Both open dashboards. Both are sincere. One is looking at revenue retained after expansion. One is counting customers who cancelled. The room treats the words as synonyms. They are not.

This is Part 2 of Finance analytics for non-finance. Part 1 covered ARR and MRR vocabulary: the recurring stock and how it moves. This part covers the retention side of the dictionary: logo churn, revenue churn, gross retention, net retention, and the boring definition stack that makes those words stick under pressure.

What you’ll learn

How customer (logo) churn differs from revenue churn
What gross retention and net retention usually mean, and why both matter
A four-step definition stack: who counts, what active means, which window, how you report
How expansion can hide logo problems in “healthy” net numbers
A worked example with a small customer table you can recompute by hand

Retention is a family of metrics, not one KPI

In everyday speech, retention means “they stayed.” In analytics, retention always needs an object: stayed as a customer, stayed as a paid seat, stayed as a recurring dollar, or came back to open the app this week. Product analytics often means activity retention (did the user return?). Finance analytics often means commercial retention (did the subscription continue?). Both are legitimate. Confusing them is how a product team celebrates weekly actives while finance watches MRR walk out the door.

For this series we stay in the commercial lane that sits next to ARR and MRR. Activity retention still matters for product work, and it can lead commercial churn, but the board pack usually wants subscription and revenue language. If you need the broader habit of defining metrics cleanly, the metrics series pairs well with this post.

Rule of thumb: Never say “retention is 92%” without naming retained what, among whom, over which window, after which rules.

The retention definition stack

Before formulas, lock four decisions. Skip one and two honest analysts will invent two different numbers.

Filled retention stack: who counts, what active means, window, report

1. Who counts

Is the unit a person, a company account, a billing parent, a subscription, or a contract? B2B companies often want logo retention at the account level even when one account has five subscriptions. Marketplace businesses might care about supply-side and demand-side retention separately. Internal “customers” (another business unit) may be excluded from external SaaS metrics.

Write the unit in one sentence: “A logo is a distinct paying company account with at least one active recurring subscription.” If you cannot write that sentence, you are not ready to report logo churn.

2. What active means

Active might mean paying, not cancelled, not delinquent past N days, or still within contract even if payment failed. Paused subscriptions are a classic trap. Some teams treat pause as retained. Some treat pause as churned. Some invent a third status and exclude pauses from both numerator and denominator. Any of those can be fine. Silent inconsistency is not fine.

3. Window

Monthly, quarterly, and annual windows change the story. Logo churn of 3% per month is not the same economic reality as 3% per year, yet people drop the time unit in conversation. Cohort windows (customers acquired in January, measured at month 3) differ from portfolio windows (all customers active at the start of the quarter). Part 3 goes deep on cohort charts. Here, just force the window onto the slide title.

4. Report shape

Will you report a rate, a count, a dollar amount, a bridge, or a curve? Will you show starting base in the denominator? Will you annualize monthly rates? Annualizing a monthly churn rate with a naive formula can mislead when churn is not constant. Prefer showing the raw period rate and the window label.

Logo churn versus revenue churn

Logo churn (customer churn) counts customers who stopped being customers under your definition. If you started the month with 100 logos and lost 5 with no new logos, logo churn is 5%. New logos do not usually reduce churn rate in the classic starting-base formulation; they show up in growth metrics separately. Confirm whether your company uses starting base, average base, or another denominator. Starting base is common and easy to audit.

Revenue churn looks at recurring dollars lost from the base that existed at the start of the period, typically from cancellations and sometimes from downgrades depending on whether contraction sits inside “churn” or beside it. A company can lose few logos and lots of revenue if a large account leaves. A company can lose many small logos and little revenue if whales stay and expand.

Both metrics answer useful questions:

Logo churn asks: are we keeping relationships?
Revenue churn asks: are we keeping recurring dollars from the base we already had?

Gross retention and net retention

These two are board favorites and definition landmines.

Gross revenue retention (GRR, sometimes “gross retention”) usually measures how much of starting recurring revenue you kept after churn and contraction, without counting expansion as a rescue. In simplified form for a period:

GRR ≈ (starting MRR − churned MRR − contraction MRR) / starting MRR

Capped thinking: GRR cannot exceed 100% in standard formulations because expansion is excluded from the numerator benefit.

Net revenue retention (NRR, NDR, “net retention”) usually allows expansion to offset losses:

NRR ≈ (starting MRR − churned MRR − contraction MRR + expansion MRR) / starting MRR

NRR can exceed 100% when expansion outweighs losses. That can be genuinely excellent. It can also hide a hole: a few expanding enterprise accounts paper over a small-customer exodus. Healthy companies often track NRR and GRR and logo churn together so the story cannot hide in one number.

Churn card defining customer churn as stopped being a customer, revenue churn as lost recurring dollars, and net rete…

Keep a churn card like the one above next to your ARR definition. If someone only quotes net retention, ask for logo churn and gross retention in the same breath.

Worked example: small base, big story split

Toy company, start of April MRR by account:

Account	Start MRR	End MRR	What happened
Acme	$10,000	$14,000	Expansion
Beta	$2,000	$0	Churned
Coral	$1,500	$1,000	Contraction
Delta	$500	$500	Flat
Echo	$1,000	$0	Churned

Starting logos: 5. Ending logos among the start set: 3 retained (Acme, Coral, Delta). Logo churn = 2 / 5 = 40% for the month among the starting set. That is rough, and the small n makes it noisy, which is a teaching feature not a bug.

Starting MRR = $15,000.

Churned MRR = $2,000 + $1,000 = $3,000
Contraction MRR = $500
Expansion MRR = $4,000

GRR ≈ ($15,000 − $3,000 − $500) / $15,000 = $11,500 / $15,000 ≈ 76.7%

NRR ≈ ($15,000 − $3,000 − $500 + $4,000) / $15,000 = $15,500 / $15,000 ≈ 103.3%

Net retention looks fine-ish for a toy month. Logo churn looks alarming. Gross retention looks weak. If the executive summary only said “NRR 103%,” you would miss half the plot. That is the whole point of carrying multiple definitions without treating them as rivals.

A sketch that forces the components into columns:

SELECT
  SUM(start_mrr) AS start_mrr,
  SUM(CASE WHEN end_mrr = 0 THEN start_mrr ELSE 0 END) AS churned_mrr,
  SUM(CASE WHEN end_mrr > 0 AND end_mrr < start_mrr
           THEN start_mrr - end_mrr ELSE 0 END) AS contraction_mrr,
  SUM(CASE WHEN end_mrr > start_mrr
           THEN end_mrr - start_mrr ELSE 0 END) AS expansion_mrr
FROM account_mrr_snapshot
WHERE cohort_month = DATE '2026-04-01';

Real systems need careful handling of mid-period changes, partial months, and account hierarchy. The sketch is a teaching scaffold: separate the loss pieces from the expansion piece before you combine them into NRR.

Voluntary, involuntary, and “we paused them”

Not all exits are equal operationally. Voluntary churn is the customer chose to leave. Involuntary churn is failed payment, expired card, or dunning that never recovered. Product and success teams want voluntary reasons. Payments and lifecycle messaging own a lot of involuntary recovery. Mixing them into one “churn” without tags makes the wrong team own the fix.

Pauses and seasonal subscriptions need a written rule. A landscaping SaaS that pauses every winter is not the same as a core product pause that predicts cancel. If pauses are large, report them as their own state in the bridge rather than stuffing them into retained or churned by vibes.

How this connects to ARR movement

Part 1’s MRR bridge and this part’s retention metrics are two views of the same engine. New and expansion grow the base. Contraction and churn shrink it. NRR focuses on the existing base’s net change. Logo metrics track relationship count. When you debug “ARR missed plan,” ask which component moved: acquisition miss, expansion miss, or retention miss. Retention miss further splits into logos versus dollars versus involuntary payments.

Stewardship still matters. If finance, CS tools, and the warehouse disagree on cancel dates, your churn rate is a data quality metric in disguise. The data quality series is the right companion when cancel timestamps refuse to line up.

Common mistakes

Quoting NRR alone. Expansion can mask logo decay and weak GRR.
Changing denominators quietly. Starting base versus average base versus “customers who could have churned” changes rates a lot.
Calling activity retention commercial retention. Last-week login is not paid subscription.
Annualizing monthly churn with a false precision formula. Show the period rate and the window.
Ignoring involuntary churn. You will staff the wrong playbook.
Letting CS “save” definitions in a spreadsheet. Official retention metrics need a warehouse definition and an owner, same discipline as ARR.
Comparing your NRR to a blog benchmark with a different definition. Benchmarks are directional at best without matched methodology.

How to practice

Write your company’s retention definition stack on one page: who, active, window, report shape.
Compute logo churn, GRR, and NRR for one recent month on a sample of accounts (even 20 rows teaches more than a vague dashboard).
Tag last month’s churns as voluntary versus involuntary if the data exists. If it does not exist, file a data request.
Add a slide footnote that states the definition in one sentence under every retention chart.
Preview Part 3: take one acquisition month cohort and sketch retention at month 1, 2, and 3 by hand before you build a pretty heatmap.

Next in this series: cohort retention charts without lying, including axes, sample size, and period alignment. For learning paths across analytics topics, use the Learn hub. If customer lifecycle stages are messy upstream of finance metrics, the customer analytics thinking in related AMS material on funnels and cohorts will help you connect product events to commercial outcomes without pretending they are the same table.

Quick recap

Retention and churn are families of metrics; name the object and window.
Logo churn tracks relationships; revenue churn tracks recurring dollars.
Gross retention excludes expansion rescue; net retention includes it and can exceed 100%.
Use the stack: who counts, active means, window, report shape.
NRR can look healthy while logos leave; carry multiple views.
Voluntary versus involuntary churn points to different owners and fixes.

Sources

ChartMogul. “Net revenue retention” and related SaaS metric explainers. https://blog.chartmogul.com/net-revenue-retention/
Paddle / ProfitWell. SaaS metrics resources on churn and retention definitions for operators. https://www.paddle.com/resources/saas-metrics
Stripe. Subscriptions and customer lifecycle billing docs (status and cancellation behavior that feed churn pipelines). https://docs.stripe.com/billing/subscriptions/overview
Tomasz Tunguz. Essays on net dollar retention and SaaS retention patterns (practitioner commentary; pair with your own definitions). https://tomtunguz.com/
Bessemer Venture Partners. Cloud and SaaS metric frameworks used widely in board conversations (context for why NRR appears so often). https://www.bvp.com/atlas