SaaS metrics are the handful of numbers that show whether a subscription software business is growing and whether that growth pays for itself. The core set is short: monthly recurring revenue (MRR) and the four ways it moves, customer churn, gross and net revenue retention (GRR and NRR), customer acquisition cost (CAC) with its payback period, and customer lifetime value (LTV). The formulas are simple. Most wrong numbers come from loose definitions.
We build SaaS products and data pipelines for clients, and this is the set we'd start a product's first dashboard with.
The core metrics on one page
Metric | How to calculate it | What it tells you |
|---|---|---|
MRR | Sum of normalized monthly subscription revenue | How big the recurring business is today |
Net new MRR | New + expansion - contraction - churned MRR | Whether you grew this month, and why |
Logo churn | Customers lost ÷ customers at the start of the period | How many customers leave |
GRR | A cohort's MRR now, excluding upgrades and price rises ÷ its MRR a year ago | How much revenue leaks out |
NRR | A cohort's MRR now ÷ its MRR a year ago | Whether existing customers grow on their own |
CAC | Sales and marketing cost ÷ new customers won | What a customer costs to win |
CAC payback | CAC ÷ (a new customer's MRR × gross margin) | Months until a customer has paid back its CAC |
LTV | Average MRR per account × gross margin ÷ monthly revenue churn | Rough gross profit a customer brings over its life |
MRR and the four ways it moves
MRR is the recurring subscription revenue you would bill in a normal month, measured today. It is a forward-looking number built from active subscriptions. Cash in the bank is a different report.
Settle these rules before anyone builds a chart:
- Annual and multi-year plans count at their monthly value. A $12,000 yearly contract is $1,000 of MRR every month, even though the cash arrived in one go.
- Subtract recurring discounts. A customer on 20% off pays you 20% less.
- Leave out taxes (VAT, sales tax), setup fees and one-off services.
- Trials count as zero until the first paid period starts.
- Decide how failed payments behave. Stripe, for example, keeps
past_duesubscriptions in MRR and counts them as churn only once they are cancelled or markedunpaid(Stripe's Billing analytics definitions). That is a reasonable default if your dunning period is short. - Pick one rule for usage-based charges and stick to it. Stripe leaves metered usage out of standard MRR. If usage is a large part of your revenue, report it as its own line, for example as a trailing three-month average, so a quiet month for one big customer doesn't look like churn.
- If you bill in several currencies, convert at a fixed rate for trend charts and show the currency effect as a separate line. Otherwise an exchange-rate move looks like customers leaving.
Every change in MRR then falls into one of four buckets:
- New MRR: customers who started paying this month. Track reactivations separately if you can.
- Expansion MRR: existing customers paying more (upgrades, extra seats, price increases).
- Contraction MRR: existing customers paying less but still paying.
- Churned MRR: customers who stopped paying altogether.
A worked example. MRR was $40,000 on 1 March. During March you signed $3,000 of new MRR and $1,500 of upgrades, and lost $500 to downgrades and $1,200 to cancellations. Net new MRR is 3,000 + 1,500 - 500 - 1,200 = $2,800, so MRR on 1 April is $42,800.
Chart the four buckets, because the total hides the story. $2,800 of growth made of $4,500 in and $1,700 out is a calm business. The same $2,800 made of $9,000 in and $6,200 out is a sales team refilling a leaking bucket, and it will stall the month sales slows down.
ARR is MRR × 12 and nothing else. Folding services revenue or one-off fees into "ARR" overstates the business, and any investor doing diligence will take them back out.
Churn and retention: count customers and count revenue
Logo churn is the share of customers who left: customers lost in the period divided by customers at the start. Lose 6 of 300 in a month and monthly logo churn is 2%.
Tools disagree on the denominator. Stripe divides churned subscribers by the subscribers at the start plus the new ones added during the period, which gives a slightly lower rate than the textbook version. Neither is wrong. Write down which one you use and keep it.
Monthly rates compound, so 2% a month is not 24% a year. You keep 0.98¹² ≈ 78.5% of customers, which makes annual churn about 21.5%.
Customer counts treat a $50 account and a $5,000 account the same, so you also need revenue retention. Both revenue measures look at a cohort: the customers who were paying at the start of the window, usually twelve months ago.
- GRR is what that cohort pays now, counting cancellations and downgrades but ignoring upgrades and price increases, divided by what it paid then. It can never exceed 100%.
- NRR is what that cohort pays now with every movement included, divided by what it paid then. Above 100% means existing customers grow faster than they leave.
Say the customers who paid $40,000 of MRR in March 2025 pay $41,600 in March 2026. Without their upgrades and price increases they would pay $36,400. NRR is 104% and GRR is 91%.
How does that compare? SaaS Capital's 2025 retention survey of more than 1,000 private B2B SaaS companies (those under $1M ARR excluded) found a median GRR of 90 to 91% in every annual-contract-value band below $250K, and 95% above it. Median NRR ranged from 98% for contracts under $12K a year to 106% for contracts over $250K. The same data shows why retention gets so much attention: companies with NRR above 130% had a median growth rate of 50%, against 15% for those below 90%. Our example sits close to the median.
Two cautions before you celebrate an NRR figure:
- Price increases count as expansion. Raise prices 10% across the board and NRR climbs by up to 10 points without anyone buying more. Always read GRR next to NRR. If you bill in a high-inflation currency, deflate the figure too. Turkish consumer prices rose 29.73% in the year to September 2026 (TÜİK, Turkey's statistics office), so a lira-billed product reporting 125% NRR is at about 96% in real terms (1.25 ÷ 1.2973).
- Small customer bases make percentages noisy. With 20 customers, a single cancellation is 5% logo churn. Below about 50 accounts, read the list instead: who left, why, and who is up for renewal next quarter.
Unit economics: CAC, payback and LTV
CAC is everything you spent on sales and marketing in a period divided by the customers you won in it. Count it fully loaded: salaries and commissions of the people doing sales and marketing, ad spend, tools, agencies and events. Leaving salaries out makes every paid channel look cheap. If your sales cycle is long, lag the spend: with three-month deals, compare this quarter's new customers with last quarter's spend.
Example: you spent $90,000 on sales and marketing in a quarter and won 15 customers, so CAC is $6,000.
That is blended CAC. CAC per channel needs every signup tagged with where it came from, and that has to happen at the first visit. On a US paid-media agency's attribution platform that we work on, everything downstream depends on one piece: a tracking script that captures all five UTM parameters and ad click IDs such as gclid and fbclid, keeps them through the session and attaches them to the conversion. If you don't store the source at signup, you can't reconstruct it later, and blended CAC is all you'll ever have.
CAC payback is CAC divided by the monthly gross profit a new customer brings. If new customers pay $500 a month at 80% gross margin, each one brings $400 of gross profit a month, and $6,000 ÷ $400 = 15 months. Use gross profit here; revenue flatters products with heavy hosting or support costs.
For reference, the 2026 benchmark report from Aleph and Benchmarkit puts median CAC payback at 16 months for full-year 2025, with the top quarter of companies at six months or less and the bottom quarter at 24 months or more (Aleph's summary). David Skok's long-standing rule of thumb in SaaS Metrics 2.0 is stricter: many of the best SaaS businesses recover CAC in 5 to 7 months, and once payback runs past 12 months, profitability gets thin. So 15 months is normal, but you are financing every new customer for more than a year before it pays you back.
LTV estimates the gross profit a customer brings over its life: average MRR per account × gross margin ÷ monthly revenue churn. With $500, 80% and 2%, LTV is $400 ÷ 0.02 = $20,000, and LTV:CAC is 3.3, just above the 3x that Skok's article sets as the bar for the best SaaS businesses.
Treat that number with suspicion:
- Churn sits in the denominator. Cut it to 1% and LTV doubles to $40,000. Early churn estimates are noisy, so early LTVs are mostly guesses.
- Once NRR is above 100%, revenue churn is effectively negative and the formula stops making sense.
- Tools calculate it differently. Stripe's subscriber LTV is revenue per subscriber divided by subscriber churn, with no gross margin at all.
Two fixes work in practice. Cap the lifetime at three years (36 × $400 = $14,400, an LTV:CAC of 2.4, which is less flattering and more believable), or wait until you have two or more years of cohorts and add up what each cohort actually paid. Until then, payback is the more honest number because it needs fewer assumptions.
Activation and usage: the early warnings
Revenue metrics lag. By the time a cancellation shows up in MRR, the customer made the decision weeks or months earlier. Usage data is where you see it coming.
Define activation as the first time an account gets what it is paying for. Take Oplist, a client we have been building for since November 2023. Its hospital rostering product uses a solver to generate department rosters, and managers review and approve them. For a product like that, a good candidate for activation is the first roster a manager approves; a login tells you very little. Track the share of new accounts that reach the moment within a fixed number of days, and the share that repeat it the following month.
In B2B, the unit to count is the account. DAU/MAU suits products people open every day; a scheduling or reporting tool used weekly will look broken on it. Measure at the product's natural rhythm, weekly or monthly active accounts.
One exercise is worth doing early: take the customers who cancelled last year and look at their usage in the 90 days before they left. Whatever dropped first belongs in your customer health score.
Getting the numbers right
Most arguments about metrics are really arguments about data. A setup that holds up:
- Use the billing system as the source of truth for revenue. The CRM knows about pipeline and contracts in negotiation; it shouldn't feed MRR.
- Build one table with one row per customer per month and that month's normalized MRR, using the rules above. Every revenue metric comes from it.
- Derive the movements with SQL. This works in PostgreSQL. It pairs each customer's MRR with the previous month's by adding a copy of every row shifted one month forward:
-- mrr_monthly(customer_id, month, mrr): one row per paying customer per month,
-- month = first day of the month (a date), mrr already normalized
-- (annual / 12, discounts subtracted, tax excluded)
WITH shifted AS (
SELECT customer_id, month, mrr, 0 AS prev_mrr
FROM mrr_monthly
UNION ALL
SELECT customer_id, (month + INTERVAL '1 month')::date, 0, mrr
FROM mrr_monthly
),
per_customer AS (
SELECT customer_id, month, SUM(mrr) AS mrr, SUM(prev_mrr) AS prev_mrr
FROM shifted
GROUP BY customer_id, month
)
SELECT
month,
SUM(mrr) AS total_mrr,
COALESCE(SUM(mrr) FILTER (WHERE prev_mrr = 0), 0) AS new_mrr, -- includes reactivations
COALESCE(SUM(mrr - prev_mrr) FILTER (WHERE prev_mrr > 0 AND mrr > prev_mrr), 0) AS expansion_mrr,
COALESCE(SUM(prev_mrr - mrr) FILTER (WHERE mrr > 0 AND mrr < prev_mrr), 0) AS contraction_mrr,
COALESCE(SUM(prev_mrr) FILTER (WHERE mrr = 0), 0) AS churned_mrr
FROM per_customer
WHERE month <= (SELECT MAX(month) FROM mrr_monthly)
GROUP BY month
ORDER BY month;The first month in the table shows everyone as new, so start your charts from the second. The month column must hold the first day of each month: shift 31 January by a month and PostgreSQL gives you 28 February, which matches nothing.
- Reconcile against invoices every month. For four years we worked on Savings United's BI team, where commission data from each affiliate network had to agree with the warehouse before anyone could trust a report, and we automated that comparison. Subscription revenue needs the same habit. Compare MRR with what was actually invoiced; every gap should be explained by annual prepayments, credit notes, refunds or currency. A gap nobody can explain usually means a broken pipeline.
- Write the definitions down, edge cases included. Is a paused account churned? Is a switch from monthly to annual billing expansion? (With normalized MRR it isn't, and an annual-plan discount even shows up as contraction.) Does a refund in the first week cancel the new MRR? Keep that page one click away from the dashboard.
On tools: if you bill through Stripe, Chargebee or Recurly, start with analytics connected to the billing system. Stripe Billing has its own dashboard, and ChartMogul and Paddle's free ProfitWell Metrics connect to the common billing platforms. Product usage goes into PostHog, Mixpanel or Amplitude. When you need to join billing, usage and CRM data, for example churn by activation cohort, move into a warehouse with a BI tool on top; our modern data stack write-up covers that setup. If you bill through a local payment gateway that doesn't calculate subscription metrics for you, store its subscription webhooks in your own table and compute MRR in SQL as above, or push the data into ChartMogul through its Import API, which ChartMogul offers for billing systems it doesn't integrate with natively.
Which metrics matter at which stage
- Up to roughly 20 paying customers: activation, retention by monthly cohort and a lot of customer conversations. Percentages at this size are noise.
- Up to about $1M ARR: the four MRR movements every month, logo churn, GRR and a rough CAC payback.
- Beyond that: NRR by segment, CAC by channel, LTV from real cohort curves, and the Rule of 40 (revenue growth rate plus profit margin should add up to at least 40%) once investors ask for it.
If NRR stays flat however well customers are served, look at pricing. A plan with nowhere to grow can't produce expansion revenue, which is one of the trade-offs in our post on SaaS pricing models.
Frequently asked questions
What is a good NRR for a B2B SaaS company? Above 100% means your existing customers grow on their own. In SaaS Capital's 2025 data the 75th percentile sits between 106% and 115% depending on contract size, so 110% or more puts you in or near the top quarter in most bands. Read it next to GRR: 110% NRR on 85% GRR means upsells are covering a leak.
What is the difference between logo churn and revenue churn? Logo churn counts customers who left; revenue churn counts the MRR they and downgrading customers took with them. Losing many small accounts gives high logo churn and low revenue churn, and that pattern often means the product fits larger customers better.
Is MRR the same as revenue in the accounts? No. MRR is a run rate built from the subscriptions active today. Accounting revenue follows IFRS 15 or ASC 606: it is recognized as the service is delivered and also includes services, setup fees and usage. Cash received is a third number. Expect all three to differ, and make sure someone can explain the gap each month.
What is negative churn? Net negative revenue churn means expansion from existing customers is larger than the MRR lost to cancellations and downgrades, so the base grows even with no new sales. It is the same condition as NRR above 100%. If you raise prices every year, check it with the price increases taken out.
If you want help building the subscription table and dashboards on your own billing data, that is the kind of work our data and AI team does.
Related Articles
WMS vs TMS: What Each Does and Which One You Need First
A WMS runs what happens inside the warehouse; a TMS runs what happens after the truck leaves. Where they meet, which to buy first, and when you need neither.
Read More: WMS vs TMS: What Each Does and Which One You Need FirstSaaS Pricing Models: How to Choose One and Bill for It
The main SaaS pricing models, when each one fits, and the decision to make first: your value metric. Plus what your pricing asks of the billing code.
Read More: SaaS Pricing Models: How to Choose One and Bill for ItSoftware Project Management: Scope, Estimates, Risk and Reporting
A working method for software projects: a scope with a won't-do list, estimates as ranges, a scored risk register, one route for changes and a one-page weekly report.
Read More: Software Project Management: Scope, Estimates, Risk and Reporting
