Marginal CAC Decides How Far You Can Scale Paid Ads

In 2024 the median SaaS company paid 2 dollars in sales and marketing for every dollar of new customer revenue, up from 1.76 a year earlier, per Benchmarkit. In the same data the blended CAC ratio, the one most teams watch, fell 13 percent to 1.40, diluted by cheaper expansion revenue. That is the whole problem in one statistic, the cost of a new customer rose while the headline number fell. The diminishing-returns curve, not bad creative or targeting, caps how far you can scale paid acquisition, and the only number that shows where the ceiling sits is marginal CAC at your current spend, not average or blended. The rule is to scale until marginal CAC equals LTV, not until average CAC hits target.

Why every channel saturates, and faster than you think

Ad response is concave, and the effect is small. The largest meta-analysis in the field, Sethuraman, Tellis and Briesch's 2011 study of 751 elasticities, put the average short-term advertising elasticity at 0.12, down from 0.22 two decades earlier, with about half not significantly different from zero. A 1 percent rise in advertising buys about 0.12 percent more sales, so doubling spend nowhere near doubles sales, diminishing returns made concrete. Marketing-mix models encode it with a saturation curve, usually a Hill function whose half-saturation point K marks where you get half the maximum response. Beyond about three times K, per MetricGate, you hit sharply diminishing returns, below K you are near-linear. The profit version is an inverted U, a marketing return of about 150 percent on the first 10,000 dollars, 40 percent on the next, and negative beyond. For direct-response digital channels the curve is safely concave from the first dollar.

Why blended CAC hides the ceiling

Blended CAC is structurally the lowest number you can report, because organic and referral customers with near-zero marginal cost dilute it, 20 to 40 percent below paid-only CAC. That is why it misleads. Brian Balfour, ex-VP of Growth at HubSpot, titled his 2017 essay Your Average CAC Is Lying to You, because averages hide what decides whether scaling helps or hurts. A Saras Analytics case shows it. A skincare brand spending 180,000 dollars a month gets a viral moment, blended CAC drops from 52 to 37 dollars, and the team adds 45,000 dollars. When the organic spike fades, that budget turns out to have acquired customers at a paid CAC above 130 dollars, while blended only ticked up to 61, more than double the dashboard number. David Skok's 2018 survey of about 385 SaaS companies shows the gap in real data, a median new-customer CAC ratio of 1.32 against a blended ratio of 1.11 and expansion at just 0.38. The Saras numbers are illustrative, the Skok medians are reported survey data.

Marginal CAC equals LTV is the real stop line

Scale each channel until the last dollar stops paying for itself, until marginal CAC equals LTV, or marginal ROAS hits breakeven. Average CAC equals target is the wrong line because it blends your cheap responsive core with your expensive marginal customer. A worked example makes it vivid. You have 100 customers for 20,000 dollars, an average CAC of 200. You spend 1,000 more to win the 101st. If that customer's lifetime value is 750, you just lost 250 dollars on the marginal acquisition, even though average CAC barely moved to about 208 and still looks healthy. Skok's rule is that a business with sound unit economics should add as many units of growth as it can while the marginal economics hold up. A great blended number is not permission to keep spending, it is the figure built to hide the marginal customer your next dollar buys.

Why CAC climbs as you scale

Selection against a finite audience is the reason. You reach your highest-intent prospects first, and as spend grows the algorithm pushes into lower-probability audiences that convert worse, so marginal CAC climbs even as volume flattens. The agency Impremis calls this the CAC trap, a saturation curve playing out in real time, not a campaign problem, and the CPM curve compounds it as the cheaper auctions run out. This rests on the Ehrenberg-Bass finding that brands grow mainly by penetration, acquiring more buyers, not loyalty, against a finite pool, with IPA data attributing about 82 percent of growth cases to penetration and about 2 percent to loyalty. Andrew Chen of a16z generalized the time dimension as the Law of Shitty Clickthroughs, the first banner ad in 1994 ran a 78 percent clickthrough rate against Facebook's 0.05 percent by 2011, a 1,500-fold decline. Automated broad targeting accelerates the exhaustion, Stackmatix notes a serviceable audience under 500,000 cycles through faster under Advantage+ than under manual campaigns.

How to find your knee, and the trap of guessing

The saturation point is measurable, so stop guessing it. Fit a marketing-mix model with adstock and a saturation curve and read the knee where marginal ROAS meets breakeven, or run burst tests that vary daily budget by 20 to 30 percent to observe marginal CAC directly, then allocate so the last dollar in every channel earns the same. An AI workflow with code execution earns its place here. You can have Claude fit a Hill or negative-exponential curve with scipy, take its first derivative for incremental conversions per dollar, invert that to marginal CAC, solve for the spend where it equals your target, and run a constrained optimization to split a fixed budget so marginal returns equalize. Open-source tools, Google Meridian, Meta Robyn and PyMC-Marketing, do this with uncertainty built in, and Robyn's allocator maximizes next-dollar response. The limits matter as much as the method. Small-data brands cannot fit a reliable curve, extrapolating past observed spend is dangerous, and the optimal point is a wide distribution. Naive curves confuse correlation with causation, since demand and spend rise together, which is why you anchor with incrementality and plan with intervals.

The scaling decision checklist (copy this)

Run this before any budget increase.

• Decide on marginal CAC, not blended. Compute the cost of the customers your next dollar buys, and report blended only as a health number.
• Set the stop line at marginal CAC equals LTV (marginal ROAS equals breakeven), and stop adding budget once a channel crosses it, however good blended looks.
• If you have no fitted curve, run a burst test, vary one channel's daily budget by 20 to 30 percent in a clean window and record marginal CAC at each level.
• Allocate by equalizing marginal returns across channels, not by hitting an average target.
• Never extrapolate past observed spend, and validate any large reallocation with a holdout or geo test.

The full workflow, fitting the saturation curve, computing marginal CAC at your current spend, solving for the spend where marginal CAC equals your target, and allocating budget across channels by equalizing marginal returns, is packaged as a reusable Claude skill. Get the free skill.

Before you call it saturation, rule out the cheaper fixes

Rising CAC is not always saturation, and the misdiagnosis is expensive because the fix is the opposite. Creative fatigue looks similar but a refresh cures it. Meta's analytics show the mean user sees the same creative about 4.2 times in 30 days, the chance of a click drops about 45 percent by the fourth exposure, and fresh creative lifted conversion about 8 percent in high-fatigue cases. But Meta is explicit that fatigue is distinct from audience saturation, which refreshing creative would not resolve. The test is clean, if new creative does not help and frequency is high, you have saturation, not fatigue. Rule out seasonality with year-over-year deseasonalized data, and auction pressure by checking whether CPMs are up market-wide rather than your frequency climbing. Once you confirm real saturation, the ceiling can be pushed outward, not just respected. New audiences, geographies, channels and offers reset the curve, what Chen calls finding the fresh powder, and lifting your landing-page conversion or average order value shifts the whole curve down, cutting CAC at every spend level without adding anyone to the audience.

What to do Monday

Change the number you scale on. Report blended CAC to the board as a health metric, but decide every incremental-spend on marginal CAC by channel, and stop adding budget once a channel's marginal CAC approaches your LTV. No fitted curve yet? Run a burst test this week and record marginal CAC at each level, then stand up the AI-assisted fit, single-channel scipy to start, graduating to Meridian, Robyn or PyMC-Marketing at a year of weekly data with real spend variation. And keep pushing the ceiling out, because the curve is not a wall, it is a position you can move.

Sources: Sethuraman, Tellis and Briesch, advertising elasticity meta-analysis, Journal of Marketing Research, 2011; Benchmarkit, 2025 SaaS Performance Metrics (new vs blended CAC ratios); David Skok, Matrix Partners, 2018 SaaS survey of about 385 companies; Brian Balfour, Reforge, Your Average CAC Is Lying to You, 2017; Saras Analytics, marginal vs blended CAC scenarios (illustrative); Ehrenberg-Bass Institute and Byron Sharp on penetration, with IPA effectiveness data; Andrew Chen, a16z, the Law of Shitty Clickthroughs; Meta analytics on creative fatigue vs audience saturation; Google Meridian, Meta Robyn and PyMC-Marketing; Measured and Google Research on response-curve reliability.