True CAC vs Reported CAC

Customer acquisition cost is a division problem, and both halves are usually wrong.

The numerator is understated because it counts media spend and nothing else. The denominator is overstated because it counts purchases rather than customers. Both errors push the result in the same direction, which is why reported CAC is reliably more flattering than the real thing.

What belongs in the numerator

Everything spent to acquire a customer, not just what the ad platform billed:

Media spend, from billing rather than the reporting interface. Credits, refunds and currency conversion mean the two disagree.

Agency fees, whether a retainer or a percentage of spend. This is money spent acquiring customers by any reasonable definition, and it is often ten to twenty percent of media.

Tax on ad invoices, where your market charges it.

Creative production — photography, video, and product sent to creators. Campaigns do not run without it, and for some brands it rivals media spend.

Acquisition discounts. A welcome offer or first-order code is margin given up specifically to convert a first purchase. Functionally it is identical to spending the same money on an ad, and it is the component most often forgotten.

Affiliate and influencer commissions paid per sale.

What does not belong: retention spend, brand campaigns with no acquisition intent, and the cost of servicing customers you already have.

What belongs in the denominator

Genuinely new customers, counted once.

This sounds trivial and is the source of most of the error. Three problems recur:

Repeat buyers counted as acquisitions. A platform reports purchases. Many of those purchases come from customers who already existed, and retargeting campaigns are disproportionately made of them. Dividing spend by purchases rather than new customers understates CAC by whatever your repeat rate happens to be.

Duplicate customer records. Guest checkout, a second email address, a different phone number. Each duplicate inflates the new-customer count and lowers apparent CAC.

Attribution double-counting. If you count “new customers attributed to Meta” plus “new customers attributed to Google”, the sum exceeds the number of new customers the store actually gained.

The safest denominator is the one that reconciles to the store: customers whose first-ever order fell in the period, counted from your own order data rather than from any platform.

Blended versus channel CAC

Blended CAC takes all acquisition spend divided by all new customers. It cannot be gamed by attribution and it reconciles to real money. It also includes customers who arrived organically, so it flatters paid efficiency.

Channel CAC attempts spend and customers per channel. More actionable when it is right, and it depends entirely on an attribution model that each platform grades generously.

Use blended for the question “is total acquisition spending sustainable”, because that is a question about real money. Use channel figures for relative decisions inside a channel, where consistent bias matters less than direction.

The number to put in front of the business is blended. It is the one that cannot be argued with, and the one that moves when the business actually changes.

Why the gap matters more than its size

A CAC understated by twenty-five percent is not a reporting inconvenience. It is a decision error that compounds.

Acquisition decisions are made by comparing CAC against contribution margin per order. Understate CAC and the comparison shows headroom that does not exist, so spend increases. Increased spend raises real CAC further, because the marginal audience is always less responsive. The reported figure moves less, because the missing components are proportional and stay missing.

The store ends up scaling into a widening loss while every dashboard it consults shows a stable, acceptable number. This is not a hypothetical failure mode — it is the ordinary consequence of comparing a platform metric against a margin figure, which is what most stores do by default.

Measure it by cohort, not by month

A monthly CAC figure divides this month’s spend by this month’s new customers, and both halves belong to slightly different populations.

Spend in the last week of a month acquires customers in the first week of the next. In a steady state that lag cancels out and nobody needs to care. During a scale-up it does not: spend rises before the customers it buys arrive, so CAC looks worse than it is while you are accelerating and better than it is when you slow down. Stores frequently panic at exactly the wrong moment because of this artefact alone.

Cohort framing removes it. Group customers by their first-order month, attribute the spend that acquired them, and follow the group forward. The CAC for a cohort settles once, rather than being restated every time the following month’s spending changes.

It also makes the comparison that actually matters possible: CAC for the January cohort against CAC for the June cohort, on the same basis. Rising cohort CAC is the earliest reliable signal that a channel is saturating, and it appears months before it reaches net profit.

Making the number honest

Three habits fix most of it.

Reconcile spend to the bank, quarterly. Total acquisition cost across every account against the money that actually left for advertising. If they disagree, something is missing.

Count new customers from your own data, by first-order date, deduplicated as far as you reasonably can.

Track CAC against contribution margin per order on the same chart. Not against a target from last quarter, and not against a platform benchmark. The only threshold that matters is your own margin, and putting the two lines together makes the moment they converge impossible to miss.