Set a target CPA that holds

The target is not an advertising number. It comes out of the business and it comes out of one calculation.

Take the contribution a conversion actually produces: the revenue, minus the cost of delivering it, minus whatever else varies with volume. That is what a conversion is worth to you. Subtract what the business needs to keep from each one, and what is left is the most you could pay before the conversion stops being worth having.

That figure is a ceiling, not a target. Setting the target at the ceiling means half your results are unprofitable by construction, because a target is an average and half of an average sits above it. The target sits below the ceiling by enough room for the distribution to be a distribution.

Work it on your own numbers rather than a benchmark — the CPA calculator takes your figures with no account attached, and the ROAS calculator does the same arithmetic from the return side if that is how your business thinks.

Two adjustments worth making before you commit the number.

Repeat value. If a conversion is worth more than its first transaction, the ceiling rises — but only by an amount you can evidence, over a horizon you can wait out. A target set on a lifetime value that arrives in eighteen months is a target the business funds for eighteen months.

What the platform can see. The system optimises towards conversions it is told about. If the valuable half of your conversions is invisible to it, a target derived from blended value asks the system to hit a number using only the data it has, which is a different and usually harder number.

This is the section that prevents the most arguments.

A target is a goal, not a ceiling. Results will land above and below it, and the promise is about the average over a period rather than about any result. Comparing a week's achieved cost against the target and concluding the platform has failed is comparing a sample against a population parameter.

Three consequences for how you set and read it.

Pick the window before you look. Long enough that the average has formed, and written down in advance so it cannot be chosen after the fact to suit the conclusion.

Expect the early period to be worse. A campaign is working out who to show ads to, and its cost during that stretch is not evidence about the target.

Read the volume beside the cost, always. A target that is being met on a tenth of the volume you wanted is not being met in any sense that matters.

Too tight. Delivery thins out. The system declines auctions it cannot win at your number, spend falls below the budget, and every reported rate looks excellent. This is the dangerous one, because the dashboard rewards it: efficient and invisible reads as a win until somebody notices the campaign has spent a third of its budget. The tell is spend well under budget with the target being met comfortably.

Too loose. The system takes the easy volume at a price you did not need to pay. Results arrive, the cost sits near the target because the target was generous, and the money you left on the table is invisible because nothing reports a counterfactual. The tell is achieved cost tracking the target almost exactly with plenty of headroom in delivery.

Neither failure announces itself. Both are found by reading spend and volume against the plan rather than by reading the rate on its own.

Every change to how a campaign bids is a change in conditions the delivery system reacts to, which makes a weekly nudge a campaign that is permanently readjusting. Fewer, larger, deliberate changes beat many small ones — a claim worth stating without a threshold attached, because nobody outside a platform can tell you where the thresholds are and the vendor figures in circulation are not sourced facts.

What is checkable is your own history: how long past changes took to settle, and what the record says you changed and when. Every applied change is a row with what it moved and from what, so the question "what did we do before this period" has an answer.

The mechanics in Tempio. A target is carried on a bid-shaped change request, and money fields on it are integer minor units of the account's currency while a target return is a ratio rather than money. Guardrails are checked when the change request is written, again at approval, and once more immediately before a platform is called.

Stated once: your connections read your accounts, a change is written as a draft on the record with the state it would change from and the checks it passed, applying it is gated on approval — by you, or by a rule you wrote — and no platform write reaches a customer account until that platform has granted write access. Applying a cleared change to a platform is in private beta.

A rule you wrote can clear a decrease under a cap it states in a currency it states. Raising a target — the fix when it is too tight — always waits for a person, and so does any change whose direction cannot be read against the previous value.

If the business cannot absorb results above the number — a fixed margin per unit, a cash position that will not tolerate a bad fortnight — then what you want is a refusal rather than a goal, and that is a different instrument with a different cost. When to use a cost cap is that decision.

If you do not yet know what a conversion is worth, the honest move is to find out before handing anybody a number. A target derived from a guess is a guess with a system optimising towards it.

How to choose a bid strategy is the decision one level up, and How budget changes restart learning is what each of these changes costs in delivery.