Plenty of companies grow themselves into trouble. Revenue climbs, the board claps, and the bank balance quietly drains, because each new customer costs more to acquire than they ever pay back. Unit economics is the discipline that catches this early, by asking a deceptively small question: forget the whole company, does one customer make money?

The quick version

  • CAC (customer acquisition cost) is the fully-loaded sales-and-marketing spend it takes to win one customer. Track it paid as well as blended, the blended figure flatters you with free organic customers.
  • Contribution margin is what's left from a customer's revenue after the variable cost of serving them. LTV should be built on this margin, not on revenue, revenue you bank, margin you keep.
  • LTV (lifetime value) is the margin a customer throws off over the whole relationship. The rule of thumb is LTV:CAC of roughly 3 or more, and recovering CAC inside about 12 months.
  • These are guidelines, not laws. LTV rests on a guess about how long customers stay, and a flattering ratio can hide a cash-flow problem that payback would have caught.

The idea in depth: cost, worth, margin, time

The four metrics answer four different questions about the same customer, and they only make sense as a set. Start with cost. CAC is, in Andreessen Horowitz's words, "the full cost of acquiring users, stated on a per user basis", salaries, ad spend, tooling, commissions, the lot, divided by the customers won. The trap is which customers you count. a16z draws the distinction sharply: blended CAC is "total acquisition cost / total new customers acquired across all channels," while paid CAC is "total acquisition cost / new customers acquired through paid marketing." Their warning is worth pinning to the wall: blended CAC isn't wrong, but on its own "it doesn't inform how well your paid campaigns are working and whether they're profitable."

Report both, and watch the gap between them. If your blended CAC looks healthy only because word-of-mouth is quietly subsidising an unprofitable paid channel, the day you lean on that channel to grow faster, your real cost surfaces, and the economics you presented to the board were never the economics you were about to scale.

Next, worth. Before you can value a customer you have to know how much of their money you actually keep, which is contribution margin, revenue from a customer minus the variable costs of serving them (hosting, support, payment fees, cost of goods). This is the number LTV must be built on. a16z is blunt about the most common error: estimating LTV "as a present value of revenue or even gross margin of the customer instead of calculating it as net profit of the customer over the life of the relationship." A £100-a-month customer who costs you £40 to serve is worth £60 a month to the business, not £100. Build your lifetime value on the £100 and every downstream decision inherits the lie.

So anchor every customer-value conversation on margin, not headline revenue, and know your variable cost per customer well enough to defend it. Margin is also the lever you control most directly: shave the cost to serve and you lift LTV without selling a single extra unit.

LTV is then the contribution margin a customer generates across the whole relationship, discounted for time. The cleanest practical form is margin-per-period multiplied by how long they stay (or, equivalently, margin divided by the churn rate). This is where the famous benchmark lives. David Skok, whose SaaS Metrics 2.0 work on the For Entrepreneurs blog set much of the industry's vocabulary, puts it directly: "The best SaaS businesses have a LTV to CAC ratio that is higher than 3, sometimes as high as 7 or 8." Below 1 you lose money on every customer; at 3 or more you have room to fund the rest of the company.

"I should stress that these are only guidelines, there are always situations where it makes sense to break them.", David Skok, SaaS Metrics 2.0, For Entrepreneurs

Finally, time, the metric founders feel in their bank account. A 5:1 LTV:CAC means nothing if you wait three years to see the money while wages fall due monthly. CAC payback is the number of months a customer's contribution margin takes to repay what you spent acquiring them: CAC divided by monthly contribution margin per customer. Skok again: "many of the best SaaS businesses are able to recover their CAC in 5-7 months," and the model turns "anemic if the time to recover CAC extends beyond 12 months." Payback is the cash-flow metric; LTV:CAC is the profitability metric. You need both, because they fail in different directions.

flowchart TD
  Cust("One customer") --> CAC(["CAC, what you spent to win them"])
  Cust --> Rev(["Their revenue, minus cost to serve = contribution margin"])
  Rev --> LTV(["LTV, that margin, over the whole relationship"])
  CAC --> Ratio(["LTV ÷ CAC: is the customer worth more than they cost? (aim ≥ 3)"])
  LTV --> Ratio
  CAC --> Pay(["CAC payback = CAC ÷ monthly margin: how long is the cash locked up? (aim < 12 mo)"])
  Rev --> Pay
Four readings off one customer, two measure profit (margin, LTV:CAC), one measures cash (payback). Leaders Loop

Why the ratio everyone quotes can mislead you

CAC and contribution margin are close to mechanical, you spent the money, you counted the customers, the cost to serve sits in the ledger. LTV is the soft one, because it depends on a number you cannot yet know: how long the customer will stay. The venture investor Bill Gurley made this the centre of his 2012 essay "The Dangerous Seduction of the Lifetime Value (LTV) Formula." His charge is that the formula's danger is its "simplicity and certainty", it produces a confident, precise-looking number out of an assumption (the retention rate) that early-stage companies are in the worst possible position to estimate. Get the churn assumption slightly wrong and LTV, which divides by it, swings wildly.

The fix is to treat LTV as a forecast with error bars, not a fact. a16z's own discipline is the tell: rather than projecting a lifetime, they say they "prefer to measure 12 month and 24 month LTV" from historical cohort data, the value a real group of customers has actually delivered so far. On top of that, lead with payback, which leans on far less guesswork: it only needs this month's margin and this month's CAC, both of which you already know.

There's a second way the ratio lies, and it's about timing rather than maths. Two businesses can post an identical, healthy 4:1 LTV:CAC. One recovers its CAC in six months; the other takes thirty. The ratio rates them equal; the bank does not. The slow-payback business has to fund a far deeper hole before the money returns, and the faster it grows, the deeper the hole gets, every new cohort starts under water. This is why a venture can look profitable per customer and still run out of cash. a16z uses "3x LTV:CAC as a rough benchmark" precisely as a rough benchmark; it is a screening heuristic, not a verdict.

The honest limitation cuts both ways. These metrics are sharpest for subscription and repeat-purchase businesses with clean cohort data; for a young company, a long-sales-cycle enterprise model, or anything with lumpy, infrequent purchases, the inputs are too noisy to support the false precision the formulae invite. The discipline is still worth running, the act of estimating CAC and margin per customer surfaces problems a P&L hides, but the outputs are a compass, not a map. Knowing roughly which direction the economics point beats a decimal place you can't defend.

A worked example

Take a small B2B software company, a composite, and every figure below is illustrative, selling a £100-a-month plan. The founders are about to raise on the strength of "great unit economics." Let's check.

Last quarter they spent £180,000 on sales and marketing and won 600 customers, so blended CAC is £300. But 200 of those came free from referrals; the £180,000 actually bought the other 400, so paid CAC is £450, half as good again. Each customer pays £100 a month, and it costs £30 to serve them (hosting, support, payment fees), so contribution margin is £70 a month, a 70% margin. Customers stay, on average, three years (36 months) before churning, so a customer's lifetime margin, a simple, undiscounted LTV, is £70 × 36 = £2,520.

Now read the verdicts. On blended CAC, LTV:CAC is 2,520 ÷ 300 = 8.4:1, spectacular, well past Skok's "7 or 8." On the paid CAC the company actually has to scale, it's 2,520 ÷ 450 = 5.6:1, still strong, but a third weaker, and the honest number to take to investors. CAC payback on paid customers is 450 ÷ 70 = 6.4 months, comfortably inside the 12-month line. So far, a healthy business.

Here's where the soft number bites. That £2,520 LTV rests entirely on customers staying three years. Suppose the real figure, once enough cohorts mature, is 20 months, not 36. LTV drops to £70 × 20 = £1,400, and paid LTV:CAC falls from 5.6:1 to 3.1:1, right at the edge of the rule of thumb. Nothing about the cost side changed; one revised assumption nearly halved the headline number. Payback, notice, didn't move at all, it never depended on the lifetime. That's exactly why a cautious operator leads with payback and treats LTV:CAC as the optimistic bookend.

flowchart LR
  subgraph Blended["Story for the pitch deck"]
    BC("Blended CAC £300") --> BR(["LTV £2,520 → 8.4:1"])
  end
  subgraph Paid["Economics you actually scale"]
    PC("Paid CAC £450") --> PR(["LTV £2,520 → 5.6:1, payback 6.4 mo"])
    PR --> PX(["If real lifetime is 20 mo: LTV £1,400 → 3.1:1"])
  end
Illustrative: the same company looks 8.4:1 on blended CAC and an optimistic lifetime, and a much tighter 3.1:1 once you use paid CAC and a realistic churn assumption. Leaders Loop

The lesson isn't a number, it's a habit. Always quote paid CAC alongside blended, build LTV on margin and a defensible lifetime, and let payback, the metric that needs the least guesswork, be your tie-breaker. That habit is where this topic plugs into the rest of the system: the funnel and conversion work that lowers your CAC, and the retention work that lengthens the lifetime LTV depends on, are the two ends of the same equation.

Frequently asked questions

What's the difference between blended CAC and paid CAC?

Blended CAC divides your total acquisition spend by all new customers, including the free ones from referrals, SEO and word of mouth. Paid CAC divides spend only by customers won through paid channels. Blended looks better because organic customers cost (almost) nothing to acquire, but as a16z notes, it tells you nothing about whether the campaigns you'd scale are actually profitable. Report both; scale on the paid number.

Should LTV use revenue or margin?

Margin, specifically contribution margin (revenue minus the variable cost to serve the customer). Building LTV on revenue overstates a customer's worth by whatever it costs to deliver the product, and a16z flags using revenue, or even gross margin, instead of net profit per customer as a common mistake. A customer is worth the money you keep, not the money they hand over.

What's a good LTV:CAC ratio?

The widely-used rule of thumb is 3:1 or higher, with the strongest businesses reaching 7:1 or 8:1 (David Skok). Below 1:1 you lose money on every customer. But Skok himself stresses these are guidelines, not laws, and a16z treats 3x as a "rough benchmark." A ratio that's too high (say 10:1) can even signal under-investment, you may be leaving growth on the table by not spending enough on acquisition.

Why does CAC payback matter if my LTV:CAC is healthy?

Because they measure different things. LTV:CAC asks whether a customer is profitable over their whole life; payback asks how long your cash is locked up before it comes back. Two companies with the same ratio can have wildly different payback, and the one with slow payback has to fund a much deeper cash hole, which gets deeper the faster it grows. Profitable-per-customer businesses still run out of money this way.

When do these metrics stop being reliable?

When the inputs get noisy. LTV depends on a retention assumption, and as Bill Gurley argued, early-stage companies are least able to estimate it, yet the formula produces a falsely precise answer. The metrics are sharpest for subscription and repeat-purchase models with mature cohort data; for new companies, long enterprise sales cycles, or lumpy one-off purchases, treat the outputs as a direction of travel rather than a decimal-place truth.

Related in the Toolkit

Where to go next