What is Target CPA?

A business whose leads close one in five into customers contributing A$900 has a ceiling of 900 × 0.20 = A$180 per lead. A target set at A$180 buys every lead the business can afford and no profit. A target at A$120 leaves a third of the contribution for fixed costs and profit. A target at A$40, set because somebody saw it once in a good week, does not produce A$40 leads — it produces almost no leads, because the bidder declines the auctions it cannot win at that price.

Target cost per acquisition is the figure an automated bid strategy is told to aim for. The platform then bids in each auction according to its estimate of how likely that impression is to produce a conversion, spending more where conversion looks probable and less where it does not, with the intention that the average across conversions lands near the target.

The word "target" is doing precise work and is frequently misread. It is not a maximum. Individual conversions will cost well above the number and others well below it. A campaign whose achieved cost matches the target while individual conversions range across a wide band is behaving exactly as designed.

The practical consequence is that a target set as though it were a cap will be experienced as a cap on delivery instead. The bidder cannot charge less than the auction clears at; what it can do is decline to enter auctions it does not expect to win at a price consistent with the instruction. Set the figure far enough below what the market charges and the campaign simply stops buying.

The only defensible source is the business, and the derivation is short.

Take the contribution a customer produces — revenue less everything that scales with serving them. Multiply by the rate at which the event being counted becomes a customer. That product is the most a conversion can cost before the transaction stops paying for itself.

Ceiling = contribution per customer × conversion-to-customer rate

The working target sits below that ceiling by whatever share of fixed costs and profit the advertising is expected to carry. A business with heavy overheads needs a wide gap; a lean one needs less.

Two common sources of the number are both wrong. The first is a published benchmark, which describes businesses whose margins and close rates are invisible. The second, more insidious, is the account's own best historical week — a figure achieved once under favourable conditions and then adopted as a standing instruction, which guarantees the campaign spends most of its life unable to meet it.

This is the single most common failure with the strategy and it is worth spelling out mechanically.

The bidder holds an estimate of the chance each impression converts. It bids in proportion to that estimate and the target. When the target is low, the implied bid is low for every impression, and the campaign wins only the auctions where competition is weak or the conversion probability is very high — which is a small and shrinking pool.

The result is a campaign spending a fraction of its budget, producing a handful of conversions, and reporting a cost per acquisition that may indeed be near the target. The target was met and the business got almost nothing.

The correct sequence is to set the target near what the campaign has recently achieved, let it stabilise, then lower it in steps of ten or fifteen percent, watching volume after each step. Where volume falls faster than cost, the previous level was the efficient frontier.

An automated strategy is a model, and models need examples.

A campaign with very few conversions in the optimisation window gives the model almost nothing to learn from, and its bids are correspondingly uninformed. Each platform publishes its own threshold, and accounts below theirs are usually better served by optimising toward a shallower and more frequent event — a lead rather than a sale, a registration rather than a purchase — until the deeper event has volume.

Every material change resets this. A large target change, a budget change, a new creative set or a structural edit puts the campaign back into a learning state where delivery is unstable and cost is unrepresentative. Judging a target during that period, and changing it again in response, produces an account that never leaves the unstable state. Changes should be spaced by at least the conversion window, and made one at a time.

The cheapest conversions in any market are bought first. As budget increases, the bidder reaches people further from a decision, and the cost of the marginal conversion rises.

A target held rigid while budget scales therefore acts as a cap: the campaign will spend only as much as it can spend at that cost, and the rest of the budget goes unused. Reports describe this as being limited, and the limit was set on purpose without anyone intending it.

Where growth is the objective, the target has to be allowed to rise toward the ceiling as volume increases. That is not a failure of efficiency; it is what buying more of a finite thing costs. The discipline is to know the ceiling and to approach it deliberately rather than to defend a target that was never derived from anything.

A target aims at the conversion the account defined, and it pursues that definition with complete indifference to whether the definition is any good.

A duplicated conversion tag halves the apparent cost and the bidder will happily optimise toward the phantom. A conversion counting add-to-cart in an account that sells things will produce a great deal of cheap cart activity. A lead form with no qualification will be filled enthusiastically by people who will never buy.

Before a target is tuned, the conversion behind it should be confirmed against the system that actually records outcomes. Most disappointing automated-bidding results are measurement problems wearing a bidding costume.

• Derive the number from contribution and close rate, not from the account's best historical week. A target below anything the campaign has achieved reduces delivery rather than cost.
• Set it near recent achieved cost and move it in steps. Large changes reset the learning and buy a week of erratic delivery for no reason.
• Give the campaign enough conversions to learn from before judging the target. Under its platform's own threshold the strategy has too little signal and its behaviour is closer to noise.
• Expect achieved cost to sit above the target as volume grows. The cheapest conversions are bought first, so a target held rigid while budget scales is a cap on delivery.
• Check the conversion definition before blaming the target. A bidder aiming at a mismeasured or duplicated conversion will hit a number that means nothing.