Customer Acquisition Cost: Why Founders Answer With a Payback Period
Nine founders gave nine differently-built acquisition costs — and several of the CAC figures in the record were arithmetic done on air, not numbers any company reported. Here is what actually goes in the fraction.
On this page
In January 2017, Nathan Latka asked Ry Walker what it cost Astronomer to acquire a customer. Walker did not have that number. What he had was a payback period — nine months, fully weighted, with sales and marketing salaries counted in rather than just ad spend — and he volunteered how it had moved: twelve months at first, down to six, back up to twelve when they doubled the growth team, then down to nine. Latka did the multiplication on air. Nine months of payback on a $3,000-a-month customer implies an acquisition cost somewhere around $27,000. “That might be true,” Walker said. “I don’t know, you’d have to check that.”
That $27,000 is the only customer acquisition cost the interview produced, and nobody at Astronomer computed it. It is a host’s arithmetic applied to a founder’s payback period — a payback period Walker said in the same breath was modelled on “not a ton of data.” The company’s entire paid-acquisition history at that point was a first monthly marketing budget of $600, switched on weeks earlier to get retargeting running. Everything else, he said, had been referral and outbound.
CAC is a definitional problem before it is a benchmark one. Three things have to be settled before the fraction means anything: what goes in the numerator, what counts as an acquired customer in the denominator, and which slice of the business the fraction covers. Founders on the record settle them differently, or not at all — and when the answer is missing, somebody else supplies one by working backwards from a payback period.
CAC = sales and marketing spend in a period ÷ new customers acquired in that periodThe definition nobody disputes. Every term in it is contested.
CAC = payback months × monthly gross profit per customerThe same identity run backwards — how a CAC gets manufactured out of a payback period a founder did state. It inherits every assumption in the payback figure and adds one of its own.
What this explainer owns, and what the neighbours own
Four related guides sit beside this one, and they answer different questions. This one is upstream of all of them: it is about whether two CAC figures are even measuring the same thing.
What the number should be, sorted by sales motion, out of the disclosures founders put on tape. That guide answers “is mine good?” This one answers “is mine the same thing as theirs?”
The other half of the ratio, and the only thing that makes an acquisition cost good or bad. The LTV guide owns the churn-inverse trap and the LTV:CAC bands; neither is repeated here.
Whether the customer you bought stays long enough to pay you back. Churn rate takes apart the four measurement choices under any retention figure; renewal rate does the same for contracts that expire on a date.
The numerator: everything, or just the media budget?
“Fully loaded” is the phrase, and the founders who use it seriously will tell you what is inside it. Inbenta’s Jordi Torras, in February 2019, enumerated his without being asked twice: “marketing, outbound sales, inbound sales, SDRs, account executives, demos, the whole thing.” He then declined to express it as a cost per customer at all, and gave an efficiency instead — 0.8, eighty cents of loaded cost per dollar of new annual contract value.
$0.80Inbenta’s fully loaded cost of adding one dollar of new ACV — marketing, SDRs, AEs and demos included — February 2019
On his roughly $30,000 average deal that works out near $25,000 to sign one, which is our arithmetic and not his. mParticle’s Michael Katz used the same adjective and reached a much larger number: fully weighted, November 2018, he was spending $400,000 to $500,000 to land an account paying $20,000 a month, against a payback he put around 24 months and wanted at 18.
The clearest evidence that the numerator is mostly people rather than media comes from a founder complaining about the opposite. Giles Palmer put Brandwatch’s cost of acquiring one of its roughly $30,000 annual contracts at “tens of thousands, annoyingly” in early 2018 — and in the same conversation said “we don’t do enough paid ads. I think that’s an opportunity.” A company whose acquisition cost is in the tens of thousands and whose stated regret is that it under-buys advertising has told you where the money goes: a marketing team of 30 to 32 out of 420 people, seven or eight of them designers, plus the sellers.
The awkward part is that some of the spend cannot be attributed even in principle. Tomer Levy would not give an absolute figure for Logz.io in June 2017 — only that it was “currently extremely low, well beyond the standard of the industry, because we bring customers through content.” His actual budget lines tell you why an absolute figure was hard to produce.
- Paid search, deliberately tiny — “a few thousand dollars a month… five to fifteen, depends on the month.” Measurable, and almost irrelevant.
- Events and evangelists — hundreds of thousands of dollars in 2017, under a million, and in his own word “very hard to measure.”
- Content, from before the product existed — the channel he credits for the whole advantage, ranking first for ELK searches in a community Logz.io did not start. Years of it, with no line item at all.
Put the events money in the numerator and Logz.io’s CAC is one number; leave it out because it cannot be tied to a customer and the same company reports a different one. Neither choice is dishonest. Only the silence about which was made is.
The denominator: what counts as an acquired customer
Astronomer’s list price in January 2017 was $6,000 a month for the shared cloud and $10,000 for a private instance. The average customer paid about $3,000. Walker was explicit about why: “if a customer’s like, hey, I just need this little thing done, we’ll say sure, how much money you want to give us for it” — and then work them up to the standard deal as their use cases expand. The thing counted in the denominator was a foot in the door.
Which is what makes the $27,000 fragile in a second way. Latka inverted the nine-month payback using the $3,000 realised average. Invert it against the $6,000 list price Walker gave in the same interview and the implied cost is about $54,000 — our arithmetic on his numbers, and neither figure is margin-adjusted. Same founder, same tape, same payback period, two defensible answers a factor of two apart.
Cropin’s Krishna Kumar described the same shape from the other end in January 2019: “Customer comes with 10,000. In three months they expand 200k, because they proven the product in one location, they take it to 20 location.” A denominator that counts logos at signature prices the acquired thing at its smallest moment.
Three further choices move the count, and none of them is usually stated:
- Which tier is in — Levy’s payback claim explicitly excluded Logz.io’s self-serve tier, which he treated as its own animal. Self-serve signups are the largest available denominator and the cheapest available customers; whether they are in or out swings the result further than any other single decision.
- Whether the period matches the sales cycle — Katz put mParticle’s enterprise cycles at six to nine months, and Daniel Saks described nine-month cycles at AppDirect. Divide this quarter’s spend by this quarter’s new logos in a business like that and you are pricing last year’s pipeline with this year’s budget.
- Whether expansion counts — a customer who triples inside the contract was acquired once. Every dollar of that growth lands in the numerator’s denominator-free zone, which is precisely why founders with strong expansion tolerate acquisition costs that look indefensible on the initial deal.
Blended, paid, or per channel
Odoo is the cleanest disclosure in the archive on this, because Fabien Pinckaers gave two numbers rather than one in May 2019. New business arrived either directly or through partners, and the two cost very different amounts.
What Odoo paid to land a new customer through its own funnel, against an average of roughly $235 a month in recurring revenue across 11,000 customers.
What a new deal cost through the partner channel — under half the direct figure, for a customer the blended number treats as identical.
A blended CAC for Odoo sits somewhere between $1,200 and $2,500 and describes no deal the company actually did. Worse, it moves with mix: a quarter in which partners close a larger share drops the blended number without either channel having changed at all. That is the same failure the blended churn rate has, and it has the same fix.
Odoo also shows how little a CAC settles on its own. At $235 a month, a $2,500 direct acquisition takes roughly eleven months of revenue to recover — our arithmetic. Pinckaers reported payback immediately, because Odoo bills annually upfront, so about $2,820 of cash lands on day one against that $2,500. Same numerator, same denominator, same price, two different answers. The billing term did all the work.
On the record
Founders report a payback period; the CAC gets computed for them
Search the archive for a founder-stated customer acquisition cost and what you mostly find is something adjacent: a payback period, a ratio, a headcount, or a refusal. Nine disclosures, in the form each founder actually gave them:
| Company (interview) | What the founder gave | Implied CAC | Whose arithmetic |
|---|---|---|---|
| Astronomer (Jan 2017) | Nine-month fully weighted payback; ~$3,000 average monthly revenue per customer | ~$27,000 | Latka’s, on air — “you’d have to check that” |
| Logz.io (Jun 2017) | “Extremely low”; payback well below a 12–24-month norm, self-serve excluded | Not given | Nobody — he declined the number |
| Brandwatch (early 2018) | “Tens of thousands, annoyingly”; $30,000 ACV at about 85% gross margin | $28,000–$29,000 | Latka’s, put to him; Palmer: “exactly” |
| AppDirect (mid-2018) | A “small, elephant-hunting type team”; 49 salespeople against 557 employees in the dataset | Not given | Nobody — he declined a payback period too |
| Movable Ink (mid-2018) | Payback under 12 months; LTV:CAC “surpassing” 5x | Not given | “We haven’t had a hard target on CAC” |
| mParticle (Nov 2018) | $400–500K fully weighted, per $20K-a-month account | Founder-stated | Katz’s; payback around 24 months |
| Cropin (Jan 2019) | $5,000–$6,000 and an eight-month payback, on a ~$13,000 contract | Founder-stated | Kumar’s, both figures |
| Inbenta (Feb 2019) | $0.80 of fully loaded cost per dollar of new ACV | ~$25,000 on a $30,000 deal | Ours |
| Odoo (May 2019) | $2,500 direct and $1,200 partner, stated separately | Founder-stated, per channel | Pinckaers’s |
Founder disclosures from the GetLatka archive, each carrying its interview date and the form the founder used. Where the last column names someone else, the underlying disclosure was a period or a ratio, not a cost per customer.
Three of the nine gave a cost per customer in the plain form the formula asks for. Two declined outright. The other four reported a period, a ratio or a team — and the cost figures that now circulate for two of them were assembled by a host multiplying out loud on a podcast, with the founder either hedging or agreeing in a single word.
None of that is misconduct. A payback period is arguably the more useful number: it already contains the price, and operators run on it. Vivek Sharma said so directly at Movable Ink in mid-2018 — asked whether he ran to a six-, twelve- or twenty-four-month CAC rule, he answered “we haven’t had a hard target on CAC”, and named payback under twelve months as the constraint he does hold the go-to-market engine to. The problem is only what happens next, when a payback period gets inverted into a cost and filed in a benchmark table beside numbers built from entirely different ingredients.
Reporting a CAC that survives diligence
- Write the numerator’s boundary down. Salaries, commissions, tools, content, events and management overhead — in or out, declared once and never quietly changed. Torras’s enumeration is the model: name the line items out loud, including the ones you excluded because nothing could be attributed to them.
- Say what a customer is. A first small land is not a standard deal. If your realised average revenue per customer is half your list price, as Astronomer’s was, the fraction is measuring something other than the deal you sell.
- Report it per channel and per segment. Odoo’s direct and partner numbers differ by more than double; a blended figure between them tracks sales mix rather than performance.
- Match the period to the sales cycle. On a nine-month enterprise cycle, this quarter’s spend bought next year’s logos, and a same-quarter fraction is measuring nothing.
- If what you have is a payback period, say that instead. It is a legitimate answer, and it is the one most operators actually run on. Just do not let it be inverted into a cost per customer without the revenue-per-customer figure and the margin assumption travelling with it.
Walker had less data than any founder in the table and was the most honest of them about it. He never claimed the $27,000 and never rejected it. What he said about the payback period underneath it belongs on every acquisition cost in every benchmark table ever published.
“Our math on that is based on not a ton of data.”
Ry Walker, co-founder and CEO, Astronomer
SourcesFounder disclosures from the GetLatka archive: Astronomer (Ry Walker, January 2017), Logz.io (Tomer Levy, June 2017), Brandwatch (Giles Palmer, early 2018), AppDirect (Daniel Saks, mid-2018), Movable Ink (Vivek Sharma, mid-2018), mParticle (Michael Katz, November 2018), Cropin (Krishna Kumar, January 2019), Inbenta (Jordi Torras, February 2019) and Odoo (Fabien Pinckaers, May 2019); AppDirect’s salesperson and headcount figures from GetLatka rows dated 30 November 2018. The $27,000 and the $28,000–$29,000 are Nathan Latka’s arithmetic on air, as marked in the table. The $25,000 implied by Inbenta’s ratio, the $54,000 implied by Astronomer’s list price, and the eleven months of revenue behind Odoo’s direct cost are our arithmetic applied to the disclosed figures.
