The number nobody had computed. CPA separated into branded and non-branded search. Branded terms — people typing the company's own name — convert cheaply, because those customers had already decided. Non-branded prospecting costs several times more per customer, and some campaigns were costing many multiples of the average. Blended together, the cheap branded volume subsidized all of it and the account looked fine. Nobody was hiding anything. The structure of the reporting simply made it invisible.
We also set a bar the account had never been judged against: a $300 CPA, which was what a six-month gross-margin payback allowed for that business. Not an opinion about efficiency — a cash constraint, derived from the economics.
The test and how we made it readable. We split branded from non-branded, computed CPA campaign by campaign, and cut every campaign coming in above $300. The prediction was stated before the cut: because much of the removed spend was buying customers inefficiently rather than buying incremental ones, total acquisition should fall by far less than the proportion of budget removed. A larger drop would have meant the analysis was wrong.
Result. Roughly $300,000 of annual paid search spend came out of the account. The cost was about 300 customers a year — the campaigns being cut were acquiring customers at close to $1,000 each, several times the bar. Around $200,000 a year was freed and redirected into a testing program that had not previously existed: money that had been leaving the business every month, now buying information about channels that had never been tried.
What transfers. Branded search is demand capture; non-branded is demand creation. They are different activities with different economics, and averaging them together conceals the expensive one behind the cheap one. Separate them before judging any paid search account.
And the CPA bar is not a matter of taste. It comes out of the economics: what gross margin per customer supports at a six-month payback. Once that number exists, cutting is arithmetic rather than argument — which is why this kind of decision tends to stall indefinitely in businesses that have never computed it.
