CAC payback, and the argument hidden in its denominator
Whether you divide by revenue or by gross profit changes the answer by the whole gross margin, and the two conventions are both in live use — so quoting a payback number without saying which one you used is not a number at all.
CAC payback is the metric a design case reaches most naturally and gets wrong most quietly. Naturally, because onboarding, self-serve setup and demo friction are design problems that plainly cost sales time. Quietly, because the number has a fork in it, both branches are in live use, and almost nobody says which one they took.
What the metric asks
Payback asks how long it takes to earn back what you spent to acquire a customer. Cost on top, per-month return underneath, answer in months. The argument is entirely about what goes underneath.
One convention divides by the customer’s monthly recurring revenue. The other divides by the monthly gross profit that revenue carries, on the reasoning that a dollar of revenue costing forty cents to deliver does not pay back a dollar of acquisition cost. Both are defensible. Both are in use. a16z Growth states it flatly: “some companies adjust CAC payback for gross margins and some don’t”, and, describing what it had to do to compare its own portfolio, “to compare performance consistently across companies, we gross margin-adjusted all our CAC paybacks”.
Sit with what that second quote is evidence of. A venture firm with direct visibility into many companies’ internal reporting found the numbers not comparable until it imposed one convention. This is not a pedant’s hypothetical about definitions. It is a firm describing work it had to do because the ambiguity is real. And a16z does not say the adjusted version is the correct one — it says practice is split, and that it picked one for its own purposes.
How much the fork is worth
The following arithmetic is the course’s own, built to size the gap. The acquisition cost and the per-customer revenue are invented. The margin is not: 67% is the gross margin read off Snowflake’s fiscal 2026 results in the P&L lesson, used here because it is a real software company’s real margin rather than a round number chosen to make a point.
Assumptions (illustrative, not a benchmark)
Cost to acquire one customer $12,000
Revenue per customer, monthly $1,000
Gross margin 67%
Unadjusted
$12,000 / $1,000 12.0 months
Gross-margin adjusted
$12,000 / ($1,000 x 0.67) 17.9 monthsThe same company, the same spend, the same customers. Twelve months or about eighteen, depending on a convention nobody stated. Against the published bands below, twelve sits right on the line between the middle tier and the one above it, while 17.9 sits at the far edge of the widest and least flattering one. Nothing about the business changed between those two readings.
The second fork, in the numerator
The denominator is not the only ambiguity. a16z’s 16 Startup Metrics separates paid CAC — paid acquisition spend divided by the customers that spend actually won — from blended CAC, which divides total acquisition cost by all new customers including the ones who arrived organically. The blended number is always the flattering one, because free customers are dividing a cost they did not incur. The same page warns that CAC calculations commonly leave out referral fees, credits and discounts that are plainly part of what winning the customer cost.
So a payback figure has at least two undeclared choices in it before anyone argues about whether your work moved it. That is not a reason to avoid the metric. It is the reason to state your version of it in the same breath you quote it.
The benchmark table, and what it does not say
Bessemer publishes a good, better, best band for payback: 12 to 18 months, 6 to 12 months, and 0 to 6 months. It gets quoted constantly, including by people building a case for spending money.
The version of that page read while this lesson was written does not state which denominator convention the table assumes. That absence is worth more to you than the numbers are. A widely-cited benchmark, from a firm that had every reason to be precise, published without the one clause that would make it comparable — which is exactly the omission this lesson is teaching you not to reproduce. Quote the bands if you like. Quote them with that caveat attached, or you have imported the ambiguity into your own case and made it yours.
And name the party. Bessemer is a venture firm publishing the standards the companies it backs are measured against, in a report that also markets its own view of the market. The bands are a convention set by investors, not a fact about businesses.
What a design change can honestly claim here
Payback has two terms and you can only reach one of them cleanly.
- The numerator is reachable. If a surface removes specialist hours from a sales cycle, or lets a customer complete setup without a services engagement, the cost of acquiring that customer falls. That is a mechanism you can state in hours and headcount, and somebody can check it against a calendar.
- The denominator is mostly not yours. Revenue per customer moves on pricing and packaging. Gross margin moves on infrastructure and delivery cost. A UI change reaches those only through long chains, and claiming one means claiming the chain.
- Faster is not the same as cheaper. Shortening a sales cycle is genuinely valuable and it is not automatically a payback effect: if the same hours are spent in less elapsed time, cost per customer has not moved. That distinction is the one a sales leader will make within about four seconds, so make it first.
Check your recall
Answer from memory — no scrolling back.
Retrieval check
A slide says “CAC payback: 11 months.” Name the two questions that make that number usable, and say what you would do if the answer to either is “I would have to check.”
Check your answer
Ask whether it is gross-margin adjusted, and whether the CAC is paid or blended. Those two answers move the figure more than any change you could ship.
If the answer is that somebody would have to check, that is a good outcome and not an awkward one. Write your case against the convention you were told, name the convention in the case, and add a line saying the figure would need restating if the company uses the other one. A case that survives a definitional correction without being rewritten is worth more than one that quietly assumed the flattering branch.
Hands on
State the convention before you state the number
Done when: Any case in VALUE-CASES.md touching acquisition cost names paid or blended CAC, names adjusted or unadjusted payback, and expresses its claimed effect in hours or headcount rather than in months of payback.
- Pick the case with the most plausible sales or onboarding effect — likely the client work rather than the review gate — and write its acquisition claim in hours of paid human time per customer. Hours are checkable against a calendar. Months of payback are not.
- Add the two convention labels on one line: paid or blended, and adjusted or unadjusted. If you do not know which this company uses, write both possibilities and mark it as the question to ask.
- Write the elapsed-time trap sentence explicitly: say whether your change removes hours or merely compresses them, and be honest if it is the second.
- If you want to reference a benchmark band, write it with the caveat in the same sentence, or leave it out. A benchmark quoted without its convention imports somebody else’s ambiguity into your document under your name.
- Bring the hours estimate into the chat. I will ask who counted them, because a saved-hours figure nobody sourced is the most common invented number in a design case.
What this does not cover
This lesson stayed on a single customer’s economics. The next one goes up to the company-level score that gets used to judge whether the whole business is healthy at all — the Rule of 40, where it actually came from, what its author said its scope was, and the critique now published by the firm that spent years promoting it. Then the module closes on the move that converts all five of these metrics into credibility: naming the ones your change cannot touch.
Read this next — primary source
Introducing a16z Growth’s Guide to Growth MetricsAndreessen Horowitz — free; a venture firm describing how it normalised its own portfolio’s numbers
One sentence in this post is the whole reason this lesson exists: a firm that can see many companies’ real reporting says plainly that some adjust CAC payback for gross margins and some do not, and that it had to impose a single convention before any comparison was possible. Read the rest for what it is — an investor explaining the normalisation it performs before benchmarking anyone. That normalisation step is invisible in every chart you will later be shown, and knowing it happened changes how much weight a benchmark can carry.
Stuck, curious, or think this lesson is wrong? Ask your teaching agent. The lessons are the scaffold; the conversation is where the learning gets unstuck.