Marginal CAC: What Really Happens When You Double Ad Spend

Every founder I talk to who is about to scale a channel says a version of the same sentence: the ads are working, so we are going to put more money behind them. It is the most reasonable thing in the world to say. It is also a linear statement about a curve that does not go straight, and the gap between those two things is where a lot of profit quietly disappears.

Here is what makes this hard to argue about honestly. The advice to scale what is working is not wrong, exactly. It is just incomplete in a way that costs real money, and the industry that should be able to tell you how much it costs cannot actually tell you. So let me be upfront about what follows: there is no published study that tells you what happens to your customer acquisition cost (CAC, the cost to acquire one new customer) when you double your ad spend. I went looking for one. It does not exist. What does exist is enough to make the shape of the problem clear, and enough to show you that the honest version is sharper than the tidy percentage you were probably hoping for.

"Scale what's working" is a linear rule for a curve that bends

The advice gets repeated in every founder group and on every podcast: find the thing that works, do more of it until it stops working, then go find something else. As a heuristic it is fine. The problem is the middle clause. "Until it stops working" implies a cliff, some obvious moment when the channel breaks and tells you. That is not how ad response behaves.

Ad response is predominantly concave. For most brands it starts bending early, well before your dashboard shows anything wrong, rather than holding flat and then hitting a cliff. Every dollar you add tends to be slightly worse than the one before it, all the way up.

The mechanism is not mysterious. Your first dollars reach the people most likely to buy: the ones who already know you, who searched for you, who look exactly like your best customers. Those people are finite. To spend more, the platform has to show your ads to people who are progressively less likely to convert, and it has to outbid other advertisers for their attention. You are not buying more of the same thing at the same price. You are buying a worse thing at a higher price, and calling it the same line item.

This is where the naive model does its damage. When I talk to founders planning a scale-up, the arithmetic in the plan is usually some version of: take a brand spending $50K and getting 833 customers, the plan says $100K buys 1,666, maybe 1,400 if we are being conservative. That "being conservative" haircut is doing enormous work in that sentence, and nobody has ever checked whether it is the right size. It is a feeling, dressed up as a forecast.

What the real 2026 DTC data actually shows

Start with the best real-world evidence available, which is aggregate panel data rather than a controlled experiment. Northbeam's Media Buyer Monthly Retro for April 2026, covering more than 1,000 ecommerce brands, reported median ad spend up 9.6% year-over-year, median revenue up 4.4%, and median new-customer CAC up 9.7%.

Read that carefully, because it is doing two things at once. The typical brand pushed roughly 10% more into ads and got roughly 10% more expensive at acquiring customers. Close to one-for-one pass-through. And revenue grew less than half as fast as spend. The efficiency did not come along for the ride.

The cohort cut is worse. The $5M-$10M revenue band took the sharpest CAC hit in the report at +17.1%, nearly double the panel-wide rate. That band is often where founders decide it is time to press the accelerator, an uncomfortable coincidence.

Cohort	Ad spend growth YoY	New-customer CAC growth YoY	Note
Panel median (1,000+ brands)	+9.6%	+9.7%	Roughly 1:1 pass-through at the typical brand
$5M-$10M revenue band	Not separately disclosed	+17.1%	Worst CAC hit in the report
$10M-$20M revenue band	Not separately disclosed	Not disclosed	Only band with positive MER; new-customer revenue +14.3%

Two honest caveats, because this data gets misused constantly. First, these are panel medians, not a matched experiment: each figure is a separate median, so they describe the typical brand on each measure rather than one brand's before-and-after. Second, and more important, this measures about 10% spend growth. It is not a doubling. Anyone citing these numbers as evidence for what happens when you double spend is stretching them past what they can carry.

One more thing from Northbeam's January 2026 benchmark cut is worth holding onto, because it complicates the story in a useful way. Comparing the median business to the top quartile, both groups increased spend. The median saw new-customer revenue fall and CAC rise. The top quartile's new-customer CAC actually fell while revenue and MER improved. That is not the same curve producing a nicer number for the winners. It looks more like what the retention-and-margin argument later in this piece describes: operators who raise AOV, retention or margin are not riding the acquisition curve harder, they are moving it, so the same extra dollar buys a cheaper marginal customer than it would for a brand that is only bidding more aggressively. The best operators are not exempt from the curve. They just manage the marginal dollar, and the economics feeding it, more deliberately than everyone else.

Nobody publishes the number everyone quotes

Here is the part that should bother you. Ask around and someone will hand you a tidy percentage for what doubling your budget does to your CAC, usually with total confidence and no hesitation about the decimal point. I went looking for the source behind any version of that number. There isn't one.

Northbeam, Triple Whale, Recast and Prescient AI all acknowledge saturation. Recast's own materials say plainly that you cannot double spend and expect sales to double. Prescient AI markets a "% of saturation" metric. Not one of them publishes a cross-brand table mapping a spend multiple to a CAC increase. The academic literature does not have one either. It has elasticities, which are a different object.

So let's build the honest version from the one number the research does agree on. Across the classic advertising-response literature, the average short-run advertising elasticity lands near 0.10. That means a 100% increase in spend produces roughly a 10% increase in short-run sales, on average, across studied categories. It runs higher for new products (around 0.3) and much lower for established products, about 0.01 (Assmus, Farley and Lehmann 1984, via Allenby and Hanssens). Long-run elasticity runs roughly twice short-run, for the subset of campaigns found to move sales measurably in the first place, not for advertising in general.

Hold a constant-elasticity model against that figure and the math is fully reproducible. Sales multiple equals spend multiple raised to the power of 0.10. CAC index equals 100 times spend multiple divided by sales multiple. That is the whole model.

Spend multiple	Implied sales multiple (elasticity 0.10)	Implied CAC index (base 100)	Implied CAC increase
1.0x	1.000x	100	0%
1.5x	1.041x	144	+44%
2.0x	1.072x	187	+87%
2.5x	1.096x	228	+128%
3.0x	1.116x	269	+169%

Notice which direction this cuts. If anything, the rigorous published literature implies a steeper marginal CAC penalty for a genuine 2x spend move than the comfortable round numbers that get traded around. Whatever tidy figure you were handed was not just unsourced. It was optimistic.

I want to be equally careful about what this model is not. That 0.10 elasticity is a cross-category average drawn largely from pre-digital advertising research, extrapolated onto a modern auction environment with real-time bidding and audience-level saturation that behaves differently from broadcast reach. It is directionally informative. It is not a forecast for your account. A published adstock example makes the same point from another angle: doubling gross rating points from 100 to 200, which implies doubling spend only if cost per GRP holds flat, moved the modeled sales effect from 4.6 to 5.3 units, a 100% increase in delivery producing a 15% effect increase. Modern media-mix modeling tools, including Google's Meridian, formalize exactly this with saturation curves built to find the point where marginal return decays toward zero.

The claim	What is actually published	Verdict
"CAC rises X% when you double spend"	No vendor or academic source maps spend multiples to CAC increases	Unsupported. Ask for the source.
"Scaling a channel raises its CAC"	Concave response curves; ~0.10 average short-run elasticity; 2026 panel CAC +9.7% on spend +9.6%	Well supported, three independent evidence types
"Your CAC will rise about 10% like the panel did"	Panel measured ~10% spend growth, not a doubling	Misapplied. Wrong scale of move.
"There's a ceiling number for your channel"	Practitioner guides suggest ranges; no disclosed methodology behind any of them	Treat as opinion, not benchmark

What public companies already tell the SEC

If you want the mechanism confirmed by a company with real accountability for describing it accurately, start with what gets filed under a 10-K's risk factors.

Warby Parker's FY2025 10-K states that as it grows it "may struggle to maintain cost-effective marketing strategies, and our customer acquisition costs could rise substantially, particularly if our customer mix skews towards fewer repeat purchases by existing customers and more new customers that require higher costs to acquire." That is the exact mechanism described above, filed with the SEC by a company that has every incentive to sound optimistic.

Wayfair's 2024 Annual Report to Shareholders, filed with the SEC alongside its Form 10-K, describes something subtler and more useful in the CEO's own shareholder letter: how "a new tranche of profitable advertising spend" keeps reopening as loyalty improves. Efficiency gains do not automatically drop to the bottom line, because a better-performing brand can profitably afford to reach further out. That is the curve working in your favor, and it is worth understanding that the curve moves when your retention improves. Stitch Fix ran the opposite experiment in public: marketing expense rising as a percentage of revenue through 2025 while the active client base kept shrinking.

None of these companies used the phrase "diminishing returns" next to "customer acquisition cost." I checked the full-text search. They used risk-factor language instead. The concept is identical.

Nobody can tell you your curve. Not Northbeam, not Recast, not seventy years of media-mix modeling. That is not a gap in the research you should wait to be filled. It is the entire reason "scale what's working" is a bad rule: it substitutes a number nobody has for a test you could run this month.

How to find your own marginal CAC before you commit the budget

Since no one can hand you the number, measure it. This is less work than it sounds and the arithmetic is grade-school.

Marginal CAC is the change in ad spend divided by the change in new customers. That is it. If you go from $50K and 833 new customers to $70K and 900 new customers (a calculation example, not a prediction from the elasticity model above), you spent an extra $20,000 to buy an extra 67 customers, so your marginal CAC is about $299. Meanwhile your average CAC moved from $60 to $78 and looks almost fine. The average is a fact about your history. The marginal is a fact about your next decision.

Watch how violently those two diverge in the doubling case. Take the illustrative model above and put dollars on it: a brand spending $50,000 a month at a $60 new-customer CAC is buying 833 customers. Double spend to $100,000. Under the 0.10 elasticity, customers rise to 893 (that is 833 times 2^0.10). Average CAC becomes $112, which looks survivable. But the extra $50,000 bought only just under 60 extra customers (59.8), so those customers cost about $836 each.

The $60 starting point is an assumption I picked to make the arithmetic concrete, and the ratio does not depend on it. Under this model, for a doubling of spend, the marginal CAC works out to the starting CAC divided by (2^0.10 minus 1), which is about 13.9 times whatever you started at. If your real elasticity is better than 0.10, and for a growing brand it may well be, the number is gentler. The point is not the 13.9. The point is that no plausible elasticity makes average and marginal CAC anywhere near each other, and the average is the one on your dashboard.

This is why blended CAC is such a trap here. It is genuinely the right way to run the business day to day, and I say that to founders constantly, because blended is what you actually control. But blending is averaging, and averaging is precisely the operation that hides the cost of the incremental dollar. The blend keeps looking healthy while the decision you are making right now is underwater. Both things are true at once. Use blended CAC to run the business and marginal CAC to make this one decision. If you want to go deeper on the reporting side, our blended ROAS vs breakeven MER breakdown covers where the blend helps and where it lies.

So: step the budget up in increments small enough that you can still read the result against normal noise, and hold each step long enough to clear your CAC payback period before you judge it. Then compute the marginal CAC on that step and compare it to your contribution margin per order, not to a target ROAS. The pattern I keep seeing is founders handing an agency a bigger budget with a revenue target and no gate. Give them the gate instead: here is the marginal CAC we will tolerate, here is the frequency ceiling, and if we break through either one on the way up, we hold before we push further. Agencies are generally happy to work to a gate. They just need one that exists.

What to do when you find your ceiling

Finding the ceiling is not a failure. It is information you paid for, and most brands never actually buy it.

The first move is to check whether it is a real ceiling or a creative one. Rising frequency against flat incremental reach means you are paying to talk to the same people again, and that is an audience problem. Falling conversion on stable reach is usually a creative or offer problem. Those look identical in a CAC number and they have completely different fixes.

The second move is to attack the other side of the equation. Every one of those risk factors is really a statement about mix. Warby Parker's filing warns about a mix skewing toward new customers who cost more. Wayfair's describes a new tranche of profitable spend opening up as loyalty improves. Both are saying the same thing from opposite ends: your acquisition ceiling is set by what a customer is worth, so improving retention, AOV, or margin literally raises the ceiling. It moves the curve rather than fighting it. That is the lever founders reach for last and should reach for first, and it is why we spend so much time on the contribution margin side of this rather than the ad side.

The third move is the least satisfying and most often correct: accept that the channel is at its efficient size and stop trying to make it bigger. A channel producing profitable customers at a stable volume is not a problem to be solved. The instinct to keep pushing is what turns a good channel into a bad quarter.

Related reading. For the same CAC and payback math applied elsewhere, see CAC payback benchmarks and blended ROAS vs breakeven MER. For the weekly decomposition that tells you which stage the extra spend is breaking, see how to find the funnel stage that is actually raising your CAC. For how we help brands model margin and cash, see our fractional CFO work.

Sources and methodology

No public dataset maps ad-spend multiples to CAC increases, and this piece does not pretend otherwise. We looked specifically for a cross-brand benchmark tying a spend doubling to a CAC outcome across the major attribution and media-mix vendors. Each acknowledges saturation conceptually. None publishes the table. Where this piece gives a number for a doubling, it is a clearly labeled model built from published elasticity research, not observed data.

The 2026 DTC panel figures come from Northbeam's published benchmark summaries, which are vendor benchmark data rather than peer-reviewed research. Median ad spend growth (+9.6%), revenue growth (+4.4%), new-customer CAC growth (+9.7%) and the $5M-$10M band's +17.1% CAC increase are from the Media Buyer Monthly Retro for April 2026, aggregated across 1,000+ ecommerce brands. The median-versus-top-quartile comparison is from Northbeam's January 2026 benchmarks. The underlying brand-level data is not public, and these are panel medians rather than matched pairs.

The elasticity model is illustrative and built from one published input. The approximately 0.10 average short-run advertising elasticity, the ~0.3 figure for new products, the ~0.01 figure for established products, and the roughly 2x long-run multiplier come from Allenby and Hanssens, "Advertising Response," MSI Special Report (2004/2005) and Hanssens, "Long-Term Impact of Advertising" (MASB, 2008). Every modeled figure in this piece re-derives from that single elasticity: sales multiple = spend multiple ^ 0.10, CAC index = 100 x spend multiple / sales multiple. The estimate is a cross-category average drawn largely from pre-digital research and applied to a digital auction context where mechanics differ, so treat it as a way to see the shape of the curve, not as a forecast for a specific account.

The saturation-curve background comes from published media-mix modeling methodology. The adstock example (doubling GRPs from 100 to 200 lifting the modeled sales effect from 4.6 to 5.3 units) is from the Munich Personal RePEc Archive paper on advertising adstock transformations. Google's Meridian documentation on media saturation covers how Hill-function saturation curves and adstock decay are used to model declining marginal return. This is methodology background, not a benchmark dataset.

Company disclosures are quoted from annual reports filed with the SEC. The customer-acquisition-cost risk-factor language is from Warby Parker's Form 10-K for FY2025 (CIK 1504776). The advertising-tranche discussion is from Wayfair's 2024 Annual Report to Shareholders, the CEO's shareholder letter on page 5, filed with the SEC alongside its Form 10-K for FY2024 (CIK 1616707, accession 0001616707-25-000041), not from the 10-K's own text. Both are retrievable via SEC EDGAR full-text search. The Stitch Fix marketing-expense trend is reported in CNBC's August 2025 coverage of the company's return to growth. A full-text search for "customer acquisition cost" alongside "diminishing returns" returns no filings using both phrases together; the relevant disclosures use risk-factor phrasing instead.

Operator patterns described here are qualitative and carry no statistics. Where this piece describes what founders say or what we see repeatedly in scaling conversations, those are described patterns, not measured findings, and no figure anywhere in this article is drawn from client work. Every number traces to one of the named sources above or to the labeled illustrative model.

Frequently asked questions

why does the advice to scale what's working break down?

Because it treats every additional ad dollar as identical to the first one, and that is not how response curves work. The curve bends from early on, not at some obvious breaking point your dashboard will flag, and no vendor or academic source publishes the exact point where a specific channel stops paying. The rule is not wrong as a starting instinct. It just skips the step where you check whether your marginal dollar is still buying what the label on the budget says it is buying.

what is marginal cac vs average cac?

Average CAC is total spend divided by total new customers. Marginal CAC is the extra spend divided by the extra customers it bought. They diverge fast when you scale. In an illustrative model using the published 0.10 elasticity, a brand going from $50K to $100K a month sees average CAC move from $60 to $112, while the marginal CAC on the added $50K is about $836. The average is the number you report. The marginal is the number that decides whether the increase was a good idea.

is there a real study on cac when you double ad spend?

No, and that is worth knowing. We looked specifically for one. Northbeam, Triple Whale, Recast and Prescient AI all discuss saturation and diminishing returns, but none publishes a cross-brand table mapping spend multiples to CAC increases. The academic literature gives you elasticities, not CAC tables. If someone quotes you a precise percentage for doubling spend, ask where it came from.

how much should i increase my meta budget at a time?

There is no published universal answer, so ignore anyone who gives you one with a decimal point. The practical test is whether the step is readable: big enough that the change shows up above normal week-to-week noise, small enough that if it goes badly you have not spent a quarter's profit finding out. Then hold it long enough to clear your payback window before you judge it.

how do i calculate my marginal cac?

Take the change in ad spend between two periods and divide it by the change in new customers over the same periods. If you went from $50K and 833 new customers to $70K and 900 new customers (a calculation example, not a prediction from the elasticity model above), your marginal CAC is $20,000 divided by 67, or about $299, even though your average CAC only moved from $60 to $78. Compare that marginal number to your contribution margin per order, not to your target ROAS.

does blended cac hide the real problem when i scale a winning campaign?

Yes, structurally. Blended CAC averages your expensive new dollars together with your cheap existing ones, so a genuinely bad incremental spend gets diluted by all the efficient spend underneath it. The blend keeps looking fine while the decision you are actually making, whether to add the next $20K, is quietly unprofitable. Blended CAC is the right number for running the business and the wrong number for this specific decision.

how do i know if my ad account is saturated?

Watch marginal CAC and frequency together rather than ROAS alone. If your marginal CAC on the last budget step came in near or above your contribution margin per order, that step did not pay. Rising frequency with flat or falling incremental reach means you are paying more to talk to the same people again. Those two signals together are your ceiling showing up.