CPA

Spend divided by conversions is not a contested formula. What is contested, every single time, is which event counted as a conversion.

If the platform counts a form submission and the client counts a signed contract, the two are measuring different things and no reconciliation will make them agree. The agency reports a CPA of €40, the client’s sales director knows each customer cost €600 to acquire, and both are right. That conversation has ended more retainers than poor performance has.

It gets worse when the definition is unstable. Adding a second conversion action mid-quarter, a phone call, a brochure download, silently increases the denominator and drops reported CPA overnight with nothing in the account having improved. The trend line breaks for administrative reasons and you are the one asked to explain it.

The fix is unglamorous and takes twenty minutes at kickoff: write down which single event is the conversion the account is judged on, what fires it, and what would have to happen for that to change. Anything else is a secondary conversion, tracked and reported separately.

Target CPA bidding takes your figure and treats it as the price you are willing to pay per conversion, then bids per auction according to how likely that auction is to convert. It will pay well above the target for a click it predicts is valuable and well below for one it does not, aiming for the average to land on your number.

Two consequences follow, and both surprise people. First, cost per click stops being a meaningful signal on its own: a rising CPC alongside a stable CPA means the system is buying more expensive clicks that convert better, which is exactly what it was told to do. Second, the target itself changes what the account buys. Lower it and the system becomes conservative, bidding only on the auctions it is most confident about, and volume falls sharply.

The learning phase is where most damage gets done. After any significant change, such as a new target, a new conversion action or a budget shift, the system needs a period and a volume of conversions before its predictions stabilise. Judging performance during that window and reacting to it restarts the clock, and an account that gets "optimised" weekly can spend months never leaving learning.

Google’s own guidance is that Target CPA needs meaningful conversion volume to work well. On a low-volume account, a target that is technically available is not necessarily a target the system can hit.

CPA and conversion volume pull against each other for the same structural reason ROAS and volume do: the cheapest conversions are captured first. Branded search, remarketing, and audiences already close to buying are all inexpensive, and once they are exhausted every additional conversion comes from someone further away.

This means a campaign that doubled conversions at a 30% higher CPA may well be the better commercial outcome, and CPA reported on its own makes it look like a failure. An agency judged purely on CPA has an incentive to cut back to the efficient core, which reports beautifully and grows nothing.

The number that answers the real question is marginal CPA: what the next slice of spend would cost per conversion, rather than what the whole account averaged. It is harder to produce, and it is the difference between "our CPA is €45" and "we can add forty more customers a month at €70 each, and here is whether that is worth it".

Marginal CPA is also the honest way to handle a client asking for more volume at the same CPA. Usually the answer is that both are not available, and saying so with a number is better than promising and missing.

A target CPA should be derived, not chosen. Start from what a customer is worth: average order value or lifetime value, multiplied by gross margin, gives the profit a customer generates. That is the absolute ceiling. Spend more than that to acquire one and the sale loses money.

For lead generation there is a step in between. A lead is not a customer, so the ceiling on cost per lead is the profit per customer multiplied by the rate at which leads become customers. A business making €800 of margin per customer and closing one lead in ten can pay up to €80 per lead before it stops working.

Then subtract room for overheads and actual profit. A target set exactly at break-even means the advertising pays for itself and the business earns nothing from it, which is not usually the objective.

Every one of those inputs belongs to the client, which is why this is a kickoff conversation rather than an optimisation task. It is also the conversation that most often reveals the client has never calculated their close rate. Useful to discover in week one rather than in month six.

Duplicate conversions are the most common and the most embarrassing. A purchase event firing from both the confirmation page and a tag manager trigger doubles the conversion count and halves the reported CPA, and it can run for months before anyone reconciles against the client’s orders. Any CPA that looks unexpectedly good deserves a count check before it goes in a report.

Attribution differences are the second. A CPA calculated from platform-attributed conversions will always beat one calculated from the client’s own order data, because the platform credits itself generously and cannot see the rest of the journey. Neither figure is fraudulent; they answer different questions.

Averaging across dissimilar things is the third and the hardest to spot. An account-level CPA blending branded search at €8 with cold prospecting at €90 describes neither, and the blended figure moves whenever the budget split moves, which looks like performance and is arithmetic.