The short version

Target CPA tells the bidding system what you want to pay per conversion on average. It then bids toward impressions it predicts will convert at that cost.

Two things follow that people underestimate:

  1. It is an average, not a cap. Individual conversions cost more and less. A tCPA of 30 with conversions at 45 and 18 is working as designed.
  2. It changes who sees your ads. By selecting for predicted-cheap conversions, a low target steers toward a different audience than a high one. You are not buying the same people more cheaply.

The second point is where most tCPA surprises come from.

The audience effect

Lowering the target does not negotiate a discount. It restricts the system to auctions it predicts are cheap, which skews toward users who are easier to convert now — often people closer to the bottom of the funnel, or simply cheaper to reach.

This is why a lowered tCPA can improve reported CPA while worsening business results. You have shifted toward capturing demand that may have converted anyway, and away from creating new demand. The metric improves; incremental revenue does not.

Raising the target has the mirror effect: more volume, wider audience, higher reported CPA, and potentially more genuinely new customers.

Setting it

Start from what the account currently achieves, not from what you wish it achieved.

  • At current CPA. Stabilizes without disrupting delivery. The right starting point in most cases.
  • Below current. Reduces volume. Expect it, and decide whether the trade is acceptable.
  • Far below current. Delivery collapses. The system serves nothing rather than exceed the constraint.

If you need a lower CPA than the account has ever produced, the target is not the tool. Conversion rate, creative, and offer are.

Volume requirements

tCPA needs conversion volume to predict with. Below a meaningful weekly count, predictions are noisy and the target behaves erratically.

If volume is short, the options are to optimize toward an earlier and more frequent event, consolidate campaigns so the volume pools, or use a simpler strategy until there is enough data.

Common mistakes

  • Treating it as a cap and escalating when a conversion exceeds it. It is an average by design.
  • Adjusting it every few days. Each change disrupts learning; the campaign never stabilizes long enough to evaluate.
  • Lowering it to hit a reporting target. Reported CPA improves while incremental volume falls.
  • Setting it on a brand-new campaign. No history means no basis for the prediction.

tCPA or target ROAS?

Use tCPA when conversions are worth roughly the same amount — lead generation, app installs, sign-ups, subscriptions at one price.

Use target ROAS when order values vary meaningfully, because a tCPA treats a small order and a large one identically.

Practical rule

Set the target at current performance and move it in small steps, and remember that lowering it changes who you reach rather than what you pay for the same people. Judge the change on business results, not on reported CPA.

Related reading

Automate marketing in less than 2 minutes