Here is a number that has quietly misdirected more marketing budgets than any other: blended customer acquisition cost. Total spend over total new customers, one tidy figure, easy to put in a board deck. It feels like the truth because it is arithmetic. But blended CAC is an average, and budgets are not allocated on averages. They are allocated on margins — on the cost of the next customer, from the next dollar, in the next channel you decide to fund. The gap between those two ideas is where your best channel goes dark.
Imagine the readout. Blended CAC is $180, holding steady quarter over quarter, and everyone exhales. Underneath that calm surface, branded search is converting demand at $40 while a cold prospecting channel is paying $600 for customers who would mostly have arrived anyway. The blend hides both facts. It flatters the expensive channel by borrowing the efficiency of the cheap one, and it punishes the cheap one by making it look ordinary. You cannot steer on a number that averages your sharpest signal into your dullest.
Average CAC answers a question you are not asking
Average CAC tells you what acquisition has cost historically, across everything, all mixed together. Useful for a retrospective. Useless for a decision. The decision in front of a growth leader is never "what did customers cost on average" — it is "if I move $50,000 from channel A to channel B next month, what happens to volume and cost?" That is a marginal question, and only marginal CAC answers it.
Marginal CAC is the cost of the next increment of customers from a specific channel at its current spend level. It is almost never equal to the average. In a healthy channel it rises as you scale, because you exhaust the cheap, high-intent audience first and pay progressively more to reach the next, colder slice. The first $10,000 in a channel might buy customers at $90. The tenth $10,000, chasing a thinner audience through a saturating auction, might buy them at $320. The average across all of it reads as something comfortable in the middle — and that comfortable middle is a fiction you can never actually buy at.
You never spend at your average CAC. You spend at your marginal CAC. The average is a story about the past; the margin is the price of the future.
This is why the channel that looks cheapest on a blended basis is so often the one whose next dollar is most expensive. Branded search, retargeting, and existing-audience email post gorgeous average numbers because they convert demand that already exists. Their marginal curve, though, is nearly vertical — there is only so much branded demand to harvest. Pour more budget in and you do not get more customers, you get more credit for customers you already had. Meanwhile a top-of-funnel channel with an ugly average might have a flat marginal curve and real room to scale. Read the average and you defund the engine; read the margin and you find the bearing.
The harder problem: are you causing the demand, or just billing it?
Marginal CAC still rests on attribution, and attribution rests on an assumption that is usually false — that the touchpoint your platform claims credit for is the reason the customer converted. Meta will happily report a $60 CAC on a campaign whose audience was going to buy regardless. That is not acquisition; that is invoicing for inevitability. The only way through is incrementality testing, which asks the one question attribution cannot: how many of these customers would we have gotten anyway?
You answer it by deliberately withholding the channel and watching what happens. A geo holdout turns the channel off in a matched set of regions and compares conversions against regions where it stayed on; the difference is the incremental lift. A PSA test serves a charity placebo to the control group so both sides experience an ad load and only the message differs. A scaled-spend test pushes budget up sharply in one cell and measures whether net-new conversions move proportionally or barely at all. These cost something to run, and they sometimes deliver news nobody wants — that a celebrated channel is largely harvesting organic demand. That news is the most valuable output your measurement program will ever produce.
Payback window is the second axis nobody plots
Even a true, incremental marginal CAC is only half the picture, because two channels can acquire customers at the identical cost and still deserve completely different budgets. The difference is how fast the money comes back. A channel acquiring at $200 with a four-month payback can be scaled far more aggressively than one acquiring at the same $200 with a fourteen-month payback, because the first recycles cash into the next cohort three times faster. CAC tells you the price of a customer. Payback window tells you how quickly that price is refunded — and therefore how hard you can lean on the channel without starving the business of working capital.
This is where blended thinking does its quietest damage. Blend the payback windows together and a channel that returns cash in a quarter looks identical to one that returns it in a year. You end up funding the slow channel at the expense of the fast one, throttling your own compounding for no reason a spreadsheet would ever surface. Plot CAC against payback per channel and the right moves get obvious: protect the fast-payback, low-marginal-cost, high-incrementality channels and feed them first.
Reallocating on real signal
Put the instruments together and budget allocation stops being a debate and becomes a reading. For each channel you want three numbers, not one: marginal CAC at current spend, an incrementality factor from a real holdout, and a payback window. The channels worth funding are the ones with a flat marginal curve, an incrementality factor near one, and a short payback. The ones to cap are those whose marginal cost is climbing fast, whose incrementality is well under one, or whose cash takes a year to return — no matter how lovely their blended average looks.
Then move deliberately. Shift budget in increments of ten to twenty percent, hold it long enough for the payback window to report, and re-measure the marginal curve at the new spend level — because the curve moves the moment you do. This is slower than chasing the lowest number on the dashboard, and it is the only method that compounds. Blended CAC will keep sitting at the top of the report, calm and authoritative. Treat it as a vanity metric for the board, not a navigational instrument. The signal you actually steer by is one layer down, channel by channel, in the margin.