Blended CAC is total acquisition spend divided by all the new customers it produced. It rises for four reasons: the mix of traffic shifted toward expensive sources, conversion rate fell, price or offer changed who buys, or auction costs rose.
Mix is the usual answer and the last one anyone suspects. Every channel can hold its price steady while the blended number climbs.
The meeting goes the same way every time. The chart has climbed for three quarters, the founder puts it on the screen and asks what happened, and four people answer.
Paid search says competitors bid up the same terms all spring. Paid social says a well funded rival showed up in June and took the price of a thousand ad impressions with them.
Analytics says the attribution window changed. Somebody mentions that the first quarter is always the cheapest and this is the third.
Every one of those statements is checkable, and each survives the check. The meeting ends with a decision to watch it closely. Next quarter it is higher again.
Every channel owner has a true explanation for their own number. Rising blended CAC is what happens when all of those true explanations sit in one room.
More explanation will never close that gap. It closes when somebody pulls the number apart.
Why is our CAC going up when every channel looks fine?
Because you never set blended CAC. It is a weighted average of every source you buy from, weighted by how many customers each one delivered. Move the weights and the average moves, whether or not a single underlying price does.
Here is the smallest version, with figures invented for the arithmetic.
In the first quarter you have two sources. Branded search, meaning people who type your company name into Google, brings 300 customers at $200 each. Paid social prospecting brings 200 customers at $700 each.
500 customers, $200,000 spent, blended cost of $400.
By the fourth quarter branded search does exactly what it did before. The number of people searching your name is finite and you were already capturing them, so it delivers 300 customers at $200 again. Paid social has no such ceiling, so it took all the growth: 500 customers, still at $700 each.
Run it. 800 customers, $410,000 spent, blended cost of $512.50. Up 28%, and both channel owners are right that their own price never moved.
Push it harder. Suppose paid social genuinely improved and its cost per customer fell 15%, to $595, on the same volumes.
The blend lands at $447, still 12% above where the year started. Every channel flat or better, and the output still climbs.
Decompose before you diagnose. Blended CAC moving is a fact about your mix at least as often as a fact about your ads.
What actually moves blended CAC?
Only four things underneath it can move, so the decomposition has four questions. You can answer three from data you already have.
The first is mix: what share of your customers each source produced, this period against last. It moves without anyone deciding it should, because sources have wildly different ceilings.
Your cheap sources are capped by demand you did not create this quarter: organic, referral, word of mouth, branded search. Paid prospecting is uncapped and absorbs whatever growth they cannot supply. Grow faster than they can grow and the average drifts up on its own.
The second is conversion rate: of the people who arrive, how many turn into customers. Pay the same $4 for a click, convert half as many of those clicks, and your cost per customer doubles with the auction untouched.
The third is offer and price. Raise your price and fewer people convert, so acquisition costs more, and it costs more for a good reason. Shortening a trial, adding a qualification step and moving upmarket do the same.
Each lands in the CAC line a quarter or two after the decision, by which point the room has stopped connecting the two.
The fourth is affordability. Retention and margin set what a customer is worth, and therefore the most you can pay to get one. A CAC up 20% against retention up 40% describes a healthier business than the one you had.
That is the argument for judging acquisition against lifetime gross profit.
Is the ad auction really the reason?
It is the most comfortable answer available, the only one of the four requiring nobody present to have made a mistake. It is also the fastest to check.
Pull the input price for that channel across both periods: cost per click, or cost per thousand impressions if that is how you buy. That is what the auction actually sets.
Then measure it against that channel's own cost per customer. If clicks got 10% dearer while customers got 55% dearer, the auction owns 10 points of a 55 point problem. The other 45 live between the click and the sale.
Auction inflation is real and nobody's fault: fourth-quarter retail bidding, a competitor with fresh funding, a platform opening its inventory to more advertisers.
The error is arithmetic. The auction gets accepted as the whole explanation, and the room stops looking at the rest.
What if the worst channel is holding up the good ones?
Rank your sources by cost per customer and one finishes last by a humiliating margin. Usually it is a top-of-funnel source: the podcast read, the campaign aimed at strangers who could not name your company.
Every reporting model you own will make it look indefensible. Credit lands on the last thing a customer touched, and the last touch was your own brand name in a search box.
That survives perfect measurement. It sits underneath the separate and louder problem where the numbers themselves are wrong.
Fix all of your tracking and the top of your funnel still looks expensive, because the demand it creates gets harvested downstream by something cheaper.
The channel with the worst CAC is sometimes the one holding the others up. Cut it and watch the good numbers get worse.
So buy evidence before you buy a decision. A holdout is the cheap version: turn the channel off across a handful of markets, or everywhere for three weeks.
Then watch the sources that never touched it: branded search impressions, direct traffic, the answers people give when you ask how they heard about you. If those sag while it is off, it was paying for them all along.
Skip that test and you walk into the trap, the same arithmetic running backwards. Cut the worst line and the blended figure improves next month, because you removed the most expensive customers from the average.
It looks right for about four weeks. Then customers fall, the cheap sources cost more because you are bidding for demand nobody is creating, and the figure sails past where it started.
How do I run the decomposition myself?
One table, about an hour of somebody's afternoon. Rows: every source that produced a customer. Columns, for each of two periods: customers, spend, cost per customer, and share of total customers.
Two quarters is usually the right window. A month is noise, and a year hides the turn inside itself.
One construction rule saves you from the table most teams build first. Spend lands in the period you spent it. Customers land in the period they converted.
If your sales cycle runs 45 days, those columns describe different groups of people and the ratio wobbles on timing alone. Either lag the spend column to match, or pick periods long enough that the lag stops mattering.
Read the share column before anything else. It tells you how much of the total move is even available to fix inside an ad account.
Two of those rows can be true in the same quarter. Do the mix arithmetic first regardless: if two thirds of the move came from weights, two thirds of it is a question about what you will pay for the 800th customer.
Whatever comes out, give it a standing line on your weekly growth review. A mix shift caught in the week it starts is a budget decision with an owner. The same shift found three quarters later is a crisis with four defenders.
This is the first table I build on any account, before touching a bid.
It is one lens of the six in a full growth teardown, the only one needing nothing but a spreadsheet and a free afternoon.
If the chart has climbed for three quarters and every explanation in the room is true, start with the table. Let's talk.