ARR, MRR and churn — SaaS metrics, formulas, benchmarks and measurement traps

ARR, MRR and churn are three numbers an investor, co-founder or bank will ask about within the first minutes of any conversation about a SaaS product. ARR (annual recurring revenue) is recurring revenue expressed on a yearly basis — and this is where the confusion starts, because the same term means two different things in the industry, and neither of them is annual invoiced revenue. SaaS metrics look like simple arithmetic, but every tool computes them slightly differently, and comparing your own number against a benchmark computed a different way leads to bad decisions.

This article orders the metrics the way they actually follow from one another: from MRR and ARR, through MRR movements and churn, to revenue retention (GRR, NRR), customer economics (LTV, CAC), and Rule of 40. The formulas come from ChartMogul, Stripe and Baremetrics documentation; the benchmarks come from reports that state their sample, period, and whose data it is. Wherever we show a worked calculation, it's an illustration, not market data — and it's labelled as such.

ARR and MRR — definitions and formulas

MRR: recurring revenue in a month

MRR (monthly recurring revenue) is the sum of the monthly value of every active subscription. Stripe's documentation is precise about what's excluded: MRR is "the sum of the monthly-normalized value of all your active and past_due subscriptions. MRR calculations exclude any taxes applied to the subscription, any subscribers on free plans, and any metered (usage based) products" (Stripe, Analytics).

The word "normalized" matters here. Baremetrics puts it plainly: if a customer pays $1,200 a year, "you simply divide that number by 12" (Baremetrics, MRR). The simplest formula on that same page is "MRR = number of customers * average billed amount" — number of customers times average monthly amount.

ARR: two definitions under one acronym

ChartMogul flags the ambiguity directly: "ARR stands for either Annualized Run Rate or Annual Recurring Revenue. In modern SaaS, ARR usually means Annualized Run Rate: your Monthly Recurring Revenue (MRR) multiplied by 12. The stricter definition, Annual Recurring Revenue, counts only contracts of 12 months or longer" (ChartMogul, ARR).

So there are two formulas:

• ARR (annualized run rate) = MRR × 12 — the way ChartMogul itself calculates it in-app;
• ARR (annual recurring revenue) = contract value ÷ number of years, where contracts shorter than 12 months fall outside this definition entirely.

Why ARR isn't annual invoiced revenue

ARR is a snapshot, annualized. Annual invoiced revenue is the sum of what you actually billed over twelve months — including onboarding fees, training, usage overages, and minus customers who left in March. An illustration, not data: a product with MRR of €20,000 in January and €40,000 in December has a December ARR of €480,000, even though it invoiced far less than that across the whole year. The reverse also happens: a company that sold a lot of one-off implementations can show high annual revenue and low ARR. If "ARR" comes up in an investor conversation or a business plan, say which definition you mean.

MRR movements: where revenue change actually comes from

The level of MRR alone says little. What matters is what makes up the change. ChartMogul splits it into five movements (ChartMogul, MRR):

• New Business MRR — "recurring revenue from leads converting to new, paying customers";
• Expansion MRR — "extra revenue from existing customers upgrading to a higher tier or buying add-ons";
• Contraction MRR — "revenue lost when customers downgrade to a lower-priced plan or drop an add-on";
• Churned MRR — "revenue lost when customers cancel their plan during the period";
• Reactivation MRR — "revenue from previously churned customers returning to a paid plan".

Stripe describes MRR growth with the same set of components — starting MRR, plus new MRR, reactivation, and expansion, minus contraction and churn, then adjusted for foreign-exchange impact (Stripe, Analytics). That last item matters if you sell in USD or GBP and report in EUR — more on that below.

Baremetrics gives a shorter version: "Net New MRR = New MRR + Expansion MRR – Churned MRR," where on that page "Churned MRR" covers both cancellations and downgrades (Baremetrics, MRR) — the first example of a small definition difference producing a different "churn" number from the same underlying data.

Breaking revenue into movements lets you tell apart two products with identical growth. One grows because sales keeps bringing in new customers while existing ones leave. The other grows more slowly on new customers, but existing ones buy more. Total MRR is the same; the health of the business is not — and that's exactly what the MRR bridge shows.

In this example — the numbers are invented for illustration — MRR grows from €50,000 to €64,000, a 28% rise over the year. That looks healthy. But the entire €14,000 of growth — and more — came from new-customer sales (€18,000), while the base present at the start of the year shrank by €4,000. For simplicity, we're assuming all expansion, contraction and churn here applies to customers who were already with you at the start of the year — we'll come back to that assumption with NRR.

Churn: losing customers versus losing revenue

"Churn" in SaaS means departures — but you always have to say departures of what. In our article on freemium and trial models, we showed with a worked example how the difference between 2% and 6% monthly revenue churn plays out in MRR over a year. Here, the focus is on how to calculate churn in the first place.

Customer churn rate (logo churn)

Baremetrics gives the classic formula: "(Lost customers during a set time period / Total customers at the start of a time period) x 100 = Customer churn" (Baremetrics, churn).

Stripe calculates it differently. Its subscriber-churn metric is "the number of total churned subscribers in the past 30 days, divided by the number of active subscribers 30 days ago, plus the total new subscribers in the past 30 days" — new subscribers from the same period sit in the denominator too. Stripe's worked example gives "100 / 1100 = 9.1%" (Stripe, Analytics). Under the Baremetrics formula, the same raw numbers (100 losses against 1,000 customers at the start) would give 10%. Neither is wrong — they're simply different.

Revenue churn (MRR churn)

Revenue churn measures the same thing in money. Per Baremetrics: "(Lost revenue / Total revenue at the start of a time period) x 100 = Revenue churn," and a worked example on a separate page gives "MRR Churn Rate = $500 / $10,000 = 5%" (Baremetrics, churn MRR). Stripe's panel shows "churned revenue" as "the sum of churned MRR plus total contraction MRR in the period" — revenue lost from cancellations and from downgrades combined.

Why do you need both numbers? Because they say different things. If mostly your smallest customers leave, customer churn is high and revenue churn is low — the problem is product-market fit at the cheapest tier. If one large customer leaves, customer churn barely moves while revenue drops sharply. ChartMogul's data also shows that at companies above $1 million ARR, cancellations account for roughly 70% of lost ARR and downgrades for the remaining 30% (ChartMogul, SaaS Benchmarks Report, 2,100+ companies, 12 months to March 2023) — contraction is not a line item you can ignore.

How much churn is a lot

The same ChartMogul report says median monthly customer churn, as companies mature, "stabilizes at 3-4% per month," and a company at 1–2% monthly churn is already in the top 25% (ChartMogul, SaaS Benchmarks Report). Caveats: this is data to March 2023, from ChartMogul's own customer base — companies that use its subscription analytics tool. Treat it as an order of magnitude.

Net revenue churn and retention: GRR and NRR

Baremetrics also defines net revenue churn: "[(Churn MRR - Expansion MRR) / Total MRR at the start of the time period] x 100 = Net revenue churn." When expansion outweighs losses, the result goes negative — so-called negative churn. The same mechanism, seen from the other side, gives two retention metrics now standard in reporting:

• GRR (gross revenue retention) — per ChartMogul, "GRR equals starting MRR, minus contraction and churn, divided by that same starting MRR" (ChartMogul, GRR). It counts only losses, never expansion, so "GRR can never exceed 100%".
• NRR (net revenue retention, also NDR) — "NRR equals starting MRR, plus expansion and reactivation, minus contraction and churn, divided by that same starting MRR" (ChartMogul, NRR). It's usually measured year-over-year for the same customer cohort.

In our MRR-bridge example: GRR = (50,000 − 2,500 − 7,500) ÷ 50,000 = 80%, NRR = (50,000 + 6,000 − 2,500 − 7,500) ÷ 50,000 = 92%. The €18,000 from new customers enters neither figure — that's the point. As ChartMogul notes, "because expansion only ever adds, NRR ≥ GRR always holds".

NRR — why it matters more than new-customer count

NRR answers one question: what would happen to revenue if you acquired zero new customers starting tomorrow? Above 100% NRR, a company grows even without selling anything new. At 80%, it has to replace a fifth of its revenue every year before it can grow at all. That's why ChartMogul writes that B2B SaaS companies "should aim for over 100%".

The link between NRR and growth shows up in the data. In ChartMogul's 2023 report (2,100+ companies, 2022 data), companies with NRR above 100% grew an average of 43.6% a year, versus 13.1% for companies below 60% NRR (ChartMogul, SaaS Retention Report). In the 2024 edition (2,500+ companies, H1 data 2021–2024), median growth for companies at 100%+ NRR was 48% year over year (ChartMogul, SaaS Retention: The New Normal). That's correlation, not proof of causation — but it's consistent across editions.

NRR benchmarks — and why they disagree

Two recent public datasets give noticeably different pictures:

• ChartMogul, December 2025 — the "AI churn wave" report covers 3,500 software companies (roughly 2,700 B2B SaaS, 600 B2C SaaS, and 200 "AI-native" companies), with annual retention for 2025, restricted to companies with ARR of at least $250,000. Median NRR in B2B: 82% (top quartile 97%). In B2C: 49% (ChartMogul, The AI churn wave).
• High Alpha 2025 SaaS Benchmarks Report — a survey of 800+ respondents (69% US, 17% Europe), data as of the end of Q2 2025. Median NRR by ARR band: under $1M — 100%, $1–5M — 104%, $5–20M — 103%, $20–50M — 103%, above $50M — 101% (High Alpha, SaaS Benchmarks 2025).

A 20-point difference between the two medians isn't an error — it's a different sample and a different method. High Alpha is a survey: companies self-report, and most respondents are from the US. ChartMogul covers a broader slice of companies above the $250,000 ARR threshold and separately shows B2C, where — as the authors note — "B2C is much less sticky with minimal upsell." The practical takeaway: there's no single "good" NRR. Compare yourself against a benchmark whose sample resembles your company — customer segment, scale, market.

NRR depends on price

The single strongest driver is price level. In ChartMogul's SaaS Benchmarks Report (2,100+ companies, data to March 2023), "nearly half of SaaS businesses with an ARPA over $1k/month have net retention over 100%," while among companies with ARPA under $25/month, only 2% reach that level (ChartMogul, SaaS Benchmarks Report). ARPA is average monthly revenue per account — ChartMogul excludes free-plan and trial accounts from the calculation (ChartMogul, ARPA). We cover the consequences of this for a cheap product more fully in our article on freemium.

On the loss side: per ChartMogul, the best-performing companies at every stage hold GRR above 86% (ChartMogul, SaaS Retention Report), and the High Alpha 2025 survey gives median GRR across its ARR bands at 88–92%.

LTV, CAC and LTV:CAC — customer economics

LTV: the value of a customer over time

LTV (lifetime value) is, per ChartMogul, "the estimated revenue an average subscriber generates over their entire relationship with your business," with the basic formula LTV = ARPA × gross margin ÷ customer churn (ChartMogul, LTV). Baremetrics and Stripe give a version without margin: ARPU divided by churn (Baremetrics, LTV; Stripe, Analytics).

The difference matters. An illustration, not data: at ARPA of €200 a month and 2% monthly churn, the no-margin version gives an LTV of €10,000. At 80% gross margin, ChartMogul's version gives €8,000. If you're comparing LTV against customer acquisition cost — which is itself a cost, not revenue — the margin-adjusted version is the fairer one.

A second point concerns the nature of LTV itself: it's "usually a forward-looking estimate, not a historical measurement." ChartMogul states outright that the basic formula "produces an overly optimistic result," because it assumes linear churn, and cites its own analysis of 35,512 customer cohorts: in 28.3% of them, actual revenue diverged from the LTV forecast made at sign-up by more than 50%. In a young product with only a few months of history, LTV therefore carries a very wide margin of error.

CAC: the cost of acquiring a customer

CAC (customer acquisition cost) is, per ChartMogul, "total sales and marketing spend in a period divided by the number of new customers won in that same period." The "fully-loaded" version includes "salaries, commissions, tooling, and ad spend, not just ad spend alone" (ChartMogul, CAC). The most common mistake is calculating CAC from ad spend alone — it comes out attractively low, because it omits salaries and tooling.

A related metric is CAC payback period — how many months of margin from a customer it takes to cover the cost of acquiring them. In the High Alpha 2025 survey, medians run 5 months for companies under $1M ARR, 8 months in the $1–5M band, 14 at $5–20M, 20 at $20–50M, and 17 above $50M.

LTV:CAC and the 3:1 rule

The ratio of LTV to CAC tells you whether acquiring customers pays off. Baremetrics states that the "ideal LTV to CAC ratio is 3:1" (Baremetrics, LTV to CAC ratio); ChartMogul calls 3:1 "a common target." It's a rule of thumb, not a standard, for three reasons:

• LTV is a forecast — with a short history, the forecast error can exceed the difference between a 2:1 and a 4:1 ratio;
• the result depends on the formula — the no-margin version inflates LTV, CAC without salaries deflates cost, and both errors push the ratio in the same direction;
• the ratio says nothing about time — 3:1 with a 30-month payback means a completely different cash need than 3:1 with a 5-month payback. That's why LTV:CAC is worth reading alongside CAC payback period.

Rule of 40 — where it came from, and who it's for

The 40% rule entered wide circulation through a blog post by Brad Feld on 3 February 2015. Feld isn't its author — he writes that he heard it at a board meeting from a late-stage investor he doesn't name. He states the rule itself this way:

> "The 40% rule is that your growth rate + your profit should add up to 40%. So, if you are growing at 20%, you should be generating a profit of 20%. If you are growing at 40%, you should be generating a 0% profit. If you are growing at 50%, you can lose 10%. If you are doing better than the 40% rule, that's awesome." (Brad Feld, The Rule of 40% For a Healthy SaaS Company)

Feld takes growth as annual MRR growth rate and profit as EBITDA. More important, though, is a caveat that usually gets dropped from later summaries: "These are for SaaS companies at scale – assume at least $50 million in revenue." For a product that's just starting out, Rule of 40 is not a health metric: with a small base, 100% year-over-year growth is easier, and profit is beside the point.

Even within the group the rule targets, it's closer to an ambition than a norm. In the High Alpha 2025 survey, median Rule of 40 scores run 33 for companies with ARR of $1–5M, 20 at $5–20M, 24 at $20–50M, and 30 above $50M; the authors titled that page "Median Companies Fall Short of the Rule of 40." High Alpha computes it as annual ARR growth plus free-cash-flow margin or trailing-12-month EBITDA margin, and notes for comparison that the median publicly traded SaaS company scores 33%.

How to measure SaaS metrics without fooling yourself

Most SaaS-metrics mistakes don't come from arithmetic — they come from what you put into the formula. A checklist worth setting once and writing down:

• Divide annual subscriptions by 12. An annual payment booked whole into the purchase month's MRR creates a spike, followed by eleven months of apparent "decline." MRR should always carry 1/12 of the amount, regardless of when the cash actually arrived.
• No one-off fees, no refunds. Onboarding, training, and data-migration fees aren't recurring revenue. Refunds and invoice corrections shouldn't move MRR either — they're billing adjustments, not subscription movements.
• No taxes, free plans, or trials. That's how Stripe calculates it; ChartMogul also excludes free-plan and trial accounts from ARPA. A trial user isn't a customer yet — counting them in understates both churn and ARPA at once.
• Watch past-due payments. Stripe includes past_due subscriptions in MRR, and only counts churn once a subscription is cancelled or marked unpaid. With a lot of failed card payments, your reported MRR can briefly run ahead of the cash that will actually arrive.
• Decide on discounts once. Stripe allows subtracting discounts from MRR and calls that "a more conservative approach." Either version is acceptable — switching mid-year isn't.
• Currency. If you sell in USD or GBP and report in EUR, an exchange-rate move shifts MRR with zero movement on the customer side. Stripe breaks this out as its own line in the MRR bridge; do the same in your own spreadsheet, rather than crediting or blaming sales or churn for an FX effect.
• Usage-based products. Stripe excludes them from MRR. If a meaningful share of your revenue is usage-based, MRR alone won't show the full picture.

The most important rule concerns comparisons. Stripe's analytics documentation gives no formula for MRR churn rate — its benchmarking page lists "Gross and net MRR churn rate" as a metric and points to the definition of lost revenue instead (Stripe, Benchmarking). Stripe's customer-churn calculation includes new subscribers in the denominator; Baremetrics' doesn't. Baremetrics, in one place, folds downgrades into "churned MRR"; ChartMogul keeps them as a separate contraction line. The conclusion: only compare numbers computed the same way — your own results against your own past periods in the same tool, and benchmarks against benchmarks from the same provider. If you switch tools, recompute history or mark the break in the series.

Stripe, for its part, does something sensible: instead of publishing a benchmark, it compares you in-panel against a group of "at least 100 similar businesses," each with at least 100 active subscriptions. That's a useful rule for you too — with a few dozen customers, percentage metrics jump around from single decisions and can't be meaningfully compared with the market.

Which metrics matter at which stage

Not every SaaS metric makes sense from day one. Measuring NRR with ten customers produces a number that swings by dozens of points on one customer's decision.

MVP stage: activation and cohort retention

Before paying customers exist, what matters is whether users reach the point where the product delivers value (activation), and whether they return in the following weeks. Cohort retention — groups of users who signed up in the same week or month, tracked separately — shows this best. If the curve for successive cohorts drops to zero, no pricing or marketing change will fix that. We cover how to plan a first version so these measurements are even possible in our article on the MVP, and the choice between freemium and trial in our article on freemium.

First 50 customers: customer churn and MRR

Once payments start, the foundation is MRR broken into movements, and customer churn — tracked monthly, with a conversation with every customer who leaves. With a small base, the reasons behind cancellations matter more than the percentage. This is also the moment to fix the measurement rules above and not change them later. If you're building a small niche product, we've collected realistic assumptions for that scale in our article on micro-SaaS ideas.

At scale: NRR, CAC payback, Rule of 40

With hundreds of customers and a repeatable sales process, cohort and efficiency metrics take over: NRR and GRR measured year over year, CAC payback period, LTV:CAC with margin-adjusted LTV. Rule of 40 only becomes relevant once you're talking to later-stage investors — Feld was writing about companies above $50M in revenue.

For the broader context of the SaaS model, from definitions to the choice between a ready-made tool and your own product, see our guide to SaaS. The section also covers SLA, multi-tenant architecture, and how to build a SaaS application.

What's next

If you're building your own SaaS product, it's worth planning metrics alongside the first version — that's a decision about which events your application records and how it connects to your billing system, not just a reporting spreadsheet. We help startups build an MVP that collects the data these decisions need from day one.

Sources

ChartMogul — definitions of ARR, MRR (movements), NRR, GRR, LTV, CAC, ARPA

Stripe — Billing Analytics and Benchmarking documentation

Baremetrics — MRR, churn, MRR churn, LTV

Baremetrics — LTV to CAC ratio

ChartMogul — The AI churn wave, 10 December 2025 (3,500 companies, ARR ≥ $250k, 2025)

ChartMogul — SaaS Retention Report, 2023 (2,100+ companies, 2022 data)

ChartMogul — SaaS Retention: The New Normal, 2024 (2,500+ companies, H1 2021–2024)

ChartMogul — SaaS Benchmarks Report, 2023 (2,100+ companies, 12 months to March 2023)

High Alpha — 2025 SaaS Benchmarks Report (800+ respondents, survey)

Brad Feld — The Rule of 40% For a Healthy SaaS Company, 3 February 2015