What is CPA (cost per action)?

Why CPA is attractive and expensive

On its face CPA removes all the advertiser's risk: you pay for results. That is genuinely valuable, particularly for businesses that cannot absorb a campaign that does not work.

It is also the most expensive way to buy media, and the reason is structural rather than exploitative. The publisher is now underwriting your conversion rate, your pricing, your landing experience and your product — none of which they control. They price that exposure in, and the premium over an equivalent CPM is often substantial.

The other cost is selection. A publisher will only offer CPA where they are confident of the outcome, which means it is available on inventory that already performs and unavailable on the inventory where you would most value the guarantee.

Defining the action

• Make it verifiable by both parties. A disputed definition becomes a billing dispute.
• Make it deep enough to matter. CPA on a form fill produces form fills; quality is not the publisher's problem unless you make it one.
• Define the attribution rules in the contract — window, model, and what happens to actions with multiple sources.
• Agree a fraud and refund process in advance, because CPA is the model with the strongest incentive to manufacture actions.

Why AI inventory rarely offers CPA

The obstacle is measurement rather than reluctance. A publisher can only sell an outcome they can observe, and on conversational surfaces a large share of outcomes are invisible to both sides: no click, no referrer, no session, sometimes no browser at all.

Neither party can agree which actions the placement caused, so there is nothing to price. That is why AI inventory today is bought on CPM and CPC, with the advertiser supplying their own outcome measurement through a holdout.

Outcome-based AI pricing needs a shared measurement layer that does not exist yet. It is a reasonable thing to expect and not a reasonable thing to wait for.