Cost per acquisition is the easiest number in paid media to optimise, and one of the easiest to get wrong. Teams chase it because it’s simple, comparable, and sits neatly in a weekly report. The problem is that a low CPA doesn’t tell you whether the media caused the sale. It only tells you a sale happened somewhere near it.
We see this most in accounts we inherit from another agency. CPA has been driven down for months, sometimes years, and the account looks efficient on paper. Scratch the surface and the “efficient” spend is chasing branded search, retargeting warm audiences, and claiming credit for people who were already going to buy. The volume that’s actually attributable to the media, the part that’s genuinely incremental, is usually a lot smaller than the report suggests.
What CPA actually measures
CPA measures cost divided by conversions in a given window. Full stop. It says nothing about what would have happened without the spend. A branded search campaign can report a $4 CPA and still be close to worthless, because most of those clicks would have converted anyway, direct or organic. Optimise toward CPA alone and the algorithm does exactly what you asked: it finds the cheapest conversions available, which are very often the ones you already had.
This is how budgets quietly drift toward the bottom of the funnel over time. Prospecting gets starved because it never looks as efficient as retargeting on a CPA basis, even though prospecting is the only part of the account actually growing the customer base.
“A dashboard can tell you the number went down. It can’t tell you whether the media made that happen.”
Why it compounds as accounts scale
The bigger the account, the more this matters. [Placeholder — swap in a real before/after example from an account we’ve audited, with the client’s sign-off: e.g. “we’ve taken over accounts posting a strong blended CPA that looked healthy until a holdout test showed the true incremental CPA was multiples higher, once branded and retargeting volume were excluded.”] The number on the dashboard hadn’t moved. What the money was actually doing had.
What we do instead
We treat CPA as one input, not the scoreboard. Before we touch a media plan we run a holdout, geo test, or lift study wherever the account is big enough to support one, so we know which channels and campaigns are actually adding customers rather than claiming credit for them. From there, budget gets reallocated toward what’s incremental, even when that means a headline CPA gets worse before the business gets better.
That’s the efficiency-before-scale principle we build every account on: cut the spend that isn’t earning its place before asking for more budget, and be honest about which spend that is, even when the easier story is the one the dashboard already tells. It’s the same discipline behind the results we’ve reported for clients like Karmo, where return on ad spend improved year on year not by spending more, but by being stricter about what the existing budget was allowed to claim credit for.
Where to start
None of this makes CPA useless. It’s a fine number for day-to-day optimisation once you know which part of the account it’s safe to trust. The mistake is treating it as the whole picture. If your reporting can’t tell you what would have happened without the spend, it can’t actually tell you whether the spend is working. That’s usually the first thing worth checking before adding another dollar to the plan.