You can grow recurring revenue fast and still be running a business that destroys cash with every new customer. The headline number, annual recurring revenue, going up and to the right, hides that completely. The metrics that catch it are less famous and far more useful, and a commercial leader who knows which one answers which question is hard to mislead.

The quick version

  • The ARR/MRR waterfall breaks recurring revenue into its moving parts, new, expansion, contraction and churn, so "we grew 20%" turns into a story you can actually manage.
  • The Rule of 40 (growth rate + profit margin ≥ 40%) is a one-line health check that stops a board rewarding growth bought at any cost.
  • The magic number and CAC payback both ask the same uncomfortable question: is the money you spend to grow coming back fast enough?
  • Each metric is a lens, not a verdict. They are easy to game and vary widely by segment, use a small set together, and decompose any number that moves the wrong way.

The idea in depth

Subscription revenue behaves differently from one-off sales, so the traditional income statement reads it poorly. A normal business books a sale and moves on; a recurring-revenue business books a relationship that can grow, shrink, renew or cancel every month. That is why a separate vocabulary grew up around it, and why ARR (annual recurring revenue) and MRR (monthly recurring revenue) became the base unit. As Andreessen Horowitz set out in their reference piece "16 Startup Metrics" (2015), ARR should count only the genuinely recurring component, strip out one-off setup fees, hardware and professional services, or you flatter the number and mislead yourself.

The ARR/MRR waterfall: stop reading the net number

A single growth percentage is the average of forces pulling in opposite directions, and averages hide the things you most need to see. The waterfall (sometimes called the MRR or ARR bridge) splits the period's change into four flows: new revenue from new customers, expansion from existing customers buying more, contraction from customers downgrading, and churn from customers leaving. Start of period, plus new, plus expansion, minus contraction, minus churn, equals end of period.

flowchart LR
    S(["Starting ARR"]) --> N(["+ New
new logos"])
    N --> E(["+ Expansion
upsell / cross-sell"])
    E --> C(["− Contraction
downgrades"])
    C --> H(["− Churn
cancellations"])
    H --> F(["Ending ARR"])
The ARR waterfall, one net number broken into four flows you can each manage. Leaders Loop

Two businesses can both report 20% net growth. One does it with 30% new, almost no churn and healthy expansion, a durable machine. The other does it with 60% new and 40% leaking out the bottom, a leaky bucket sprinting to stay level. The net number calls them identical; the waterfall tells them apart. It is also where retention metrics come from: net revenue retention is just expansion minus contraction minus churn on the existing base, the subject of our companion piece on net and gross revenue retention.

Make the waterfall your default growth report, not the net figure. When growth dips, you want to point at one flow, "expansion stalled," "contraction spiked in the SMB segment", because a named, located cause is fixable, and a number going down is just a mood.

The Rule of 40: growth and profit on one line

Fast growth and strong profit pull against each other, you can almost always buy more of one by spending the other. The Rule of 40 is the heuristic that holds both in view at once: a healthy software business's revenue growth rate plus its profit margin should add up to at least 40%. Grow 40% and you can afford to break even; grow 20% and you should be running a 20% margin; grow 10% and a 30% margin is the price of staying healthy. The rule was popularised by venture capitalist Brad Feld in his 2015 post "The Rule of 40% For a Healthy SaaS Company", who heard it from a late-stage investor and used year-over-year recurring-revenue growth with EBITDA as the profit measure.

The rule earned its place because it tracks value rather than vibes. McKinsey's analysis, "SaaS and the Rule of 40", reports that companies at or above the line tend to command higher enterprise-value-to-revenue multiples than those below it. Worth keeping in proportion, though: McKinsey also finds only about a third of software companies clear the bar in any given year, so treat it as a stretch goal, not a baseline every healthy business hits.

Recurring-revenue metrics are lenses, not verdicts, each is true and incomplete, and the skill is knowing which one answers the question in front of you.

Treat it, then, as a portfolio test on your spending rather than a target to optimise in isolation. Use the Rule of 40 to force the real conversation, not to set a quota. If you're below the line, the question it asks is sharp: are we under-growing, or over-spending to grow? The number won't answer that. It just makes you ask, which is most of the value.

Magic number and CAC payback: is the growth paying for itself?

Growth has a price, and the only honest question is whether it comes back. Two metrics answer it from slightly different angles. The SaaS magic number, a measure popularised by Scale Venture Partners, takes the net new ARR you added in a quarter, annualises it (multiply by four), and divides by the sales-and-marketing spend of the prior quarter, the lag is deliberate, since spend takes time to convert. Run it over a full year and the four-quarters of spend and net new ARR cancel the annualising step out, which is the simpler way to read it. Above roughly 0.75 is usually taken as efficient enough to keep investing; well below it suggests you're pouring money into a funnel that doesn't return it.

CAC payback asks the same thing in the unit a CFO feels: how many months of a customer's gross-margin contribution it takes to earn back what you spent to acquire them. The widely-used reference here is David Skok's "SaaS Metrics 2.0", which frames CAC, lifetime value and payback as the unit economics that decide whether a subscription business is fundamentally sound. Skok's rule of thumb is blunt: aim to recover acquisition cost in under twelve months, because anything longer puts heavy strain on cash unless investors keep refilling the tank.

flowchart TD
    Q(["Question: is growth
paying for itself?"]) --> M(["Magic number
net new ARR ÷ prior-Q S&M"])
    Q --> P(["CAC payback
months to recoup CAC"])
    M --> A(["Portfolio view:
efficiency of the whole engine"])
    P --> B(["Unit view:
cash strain per customer"])
Two angles on the same question, efficiency at the company level, and cash strain at the customer level. Leaders Loop

Never report a growth number without an efficiency metric next to it. Growth on its own can be bought; growth with a healthy magic number or a sub-twelve-month payback is growth you can sustain. The cross-sell and upsell motions in upsell, cross-sell and land-and-expand matter here precisely because expansion revenue carries almost no acquisition cost, it's the cheapest fuel for every one of these ratios.

The honest limitation: every one of these is gameable

Here is where the playbook breaks down. None of these metrics is a fact of nature; each is a definition you chose, and most have a soft spot. ARR can be inflated by quietly counting one-off fees as recurring. The Rule of 40 mixes a growth rate with a margin, two numbers on different scales, so a company can clear it on paper while masking thin unit economics. The magic number swings hard with the timing of bookings and spend, and a single lumpy quarter distorts it. CAC payback depends entirely on which costs you load into CAC and which margin you apply, so two teams can compute wildly different numbers for the same business.

They also vary enormously by segment and stage. Enterprise customers expand readily and churn rarely; SMB customers do the opposite. A benchmark that fits a mature enterprise vendor can be actively misleading for an early-stage product selling to small businesses. Treat published benchmarks as direction, never destiny, and measure your own cohorts over time.

So write your definitions down, use a small basket of metrics together, and never let one number carry a decision on its own. A waterfall that looks healthy, a Rule of 40 above the line, and a payback under a year tell a consistent story precisely because they're hard to fake all at once. Any single one of them, alone, can be made to lie.

A worked example

The figures below are illustrative, chosen to show the mechanics rather than to report a real company.

A subscription business starts the year at £10m ARR. Over the year it adds £3m of new ARR from new logos and £2m of expansion from existing customers buying more. It also loses £0.5m to contraction (downgrades) and £1m to churn (cancellations). The waterfall: 10 + 3 + 2 − 0.5 − 1 = £13.5m ending ARR. Net growth is 35%, but now you can see how: strong new and expansion, with a manageable £1.5m of leakage. That's a durable shape, not a leaky bucket.

Now the efficiency check. Suppose it ran a 12% EBITDA margin. Rule of 40 = 35% growth + 12% margin = 47, comfortably above the line, so the spend behind that growth looks justified. To grow, it spent £4m on sales and marketing last year; this year's net new ARR was £5m (the £3m new plus £2m expansion), giving a magic number around 1.25, efficient, worth feeding. And if acquiring an average customer cost £6,000 against £800 a month of gross-margin contribution, CAC payback is roughly 7.5 months, inside Skok's twelve-month guideline.

Four numbers, one consistent verdict: this is efficient, durable growth. Change one input, say churn jumps to £3m, or S&M doubles with no extra ARR, and the metrics disagree with each other, which is exactly the early warning a single headline figure would have buried. The point of the basket is that disagreement is the signal.

Frequently asked questions

What's the difference between ARR and MRR?

None, mathematically, ARR is just MRR multiplied by twelve, the same recurring revenue annualised. Teams selling monthly often manage in MRR; teams selling annual contracts usually talk in ARR. The trap is the same for both: only count genuinely recurring revenue. Setup fees, one-off services and hardware are real revenue, but they're not recurring, and folding them in inflates the number you're using to judge the health of the subscription itself.

Is the Rule of 40 a hard target?

No, treat it as a health check, not a KPI to optimise. It's a heuristic popularised for later-stage software businesses, and a very early company growing 150% will sit far above 40 while burning cash, while a mature, profitable one may sit just under it and be perfectly healthy. Use it to start the right conversation, are we under-growing or over-spending? rather than as a line you must clear every quarter regardless of stage.

Magic number or CAC payback, which should I use?

Both, because they answer slightly different questions. The magic number is a top-down efficiency read on your whole go-to-market engine in a quarter; CAC payback is a bottom-up, per-customer read on how long cash is tied up before it returns. The magic number is quick but lumpy quarter to quarter; payback is steadier but depends heavily on how you define CAC. Watching them together catches problems either one alone would miss.

Should expansion revenue count in these metrics?

Yes, and it's where most of the upside hides. Expansion (upsell and cross-sell to existing customers) is part of net new ARR in the waterfall and feeds the magic number, yet it carries almost none of the acquisition cost a new logo does. A business with strong expansion can post excellent efficiency numbers because so much of its growth is cheap. That's why retention and expansion economics sit so close to these metrics, see net and gross revenue retention.

How often should I review these?

Monthly for the waterfall (it's your operating dashboard), quarterly for the magic number (it needs a full quarter to be meaningful and is noisy below that), and quarterly-to-annually for the Rule of 40 and CAC payback, which move slowly and reward a longer view. Reviewing efficiency metrics too frequently invites overreaction to noise; reviewing the waterfall too rarely lets a leak run for months before anyone names it.

Related in the Toolkit

Where to go next