What is a good CPA for an AI ad?

It depends on order value, margins, returns, customer type and measurement. Use comparable account data and business economics.

Does a lower CPA prove AI is better?

Not alone. Assess other changes and whether the test isolates the intended comparison.

How to assess acquisition cost for AI ads (2026)

Build a comparable baseline

Use your own comparable periods and products. Separate new and existing customers where the data supports it. Record discounts, shipping and stock. A sale to existing customers is a weak baseline for a new-product launch to an unfamiliar audience.

Divide relevant media spend by recorded purchases under the agreed definition. Show the purchase count too. With few purchases, a single order can materially change the figure. There is no universal data threshold for every account.

Keep production costs visible

Platform acquisition cost may exclude video production. Report media and production separately and combine them when assessing economics. Many inexpensive clicks cannot justify an expensive production on their own.

Attributed purchases do not establish sales that would not otherwise have occurred. Measuring additional sales requires a design that can answer that question. Timing alone is not causal evidence.

An illustrative calculation with limits

In a fictional example, EUR 3,000 media spend and 100 recorded purchases produce EUR 30 per purchase. Adding EUR 1,000 production gives EUR 40 in combined cost per recorded purchase. These are illustrative numbers, not client results or expected effects.

Profitability still requires order value, product costs, returns and other expenses. New-customer status and incremental purchases are separate questions.

Display spend and purchase counts beside averages when exposure differs. Do not imply equal certainty from unequal data.

End reporting with an observation and decision, such as stronger click response but unclear purchase effect, followed by a bounded next test.

Your own comparison period still needs scrutiny. Discounted sales to existing customers are not an equivalent baseline for a new product advertised to people who have never heard of the brand. Record shipping changes, promotions and availability before attributing a cost difference to the creative.

Purchase counts help readers assess an average. Where only a few orders are recorded, one extra order can change the figure substantially. That uncertainty should remain visible beside the result. A useful conclusion can identify a stronger click response while leaving purchase impact unresolved. The next production brief can then address a specific hypothesis without claiming a universal acquisition-cost improvement from AI.