Performance Max reports beautifully. Low cost per conversion, strong return on ad spend, steady growth. Then you pause it for a week and total revenue barely moves.
That is the central problem with PMax as most accounts run it: without structural constraints, it preferentially buys the cheapest conversions available, and the cheapest conversions in any account are the people who were already going to buy.
1. Exclude brand, explicitly
By default, PMax will serve against searches for your own company name. Those people were arriving anyway. Account-level brand exclusions move that traffic back into a branded search campaign where you can see it, cap it and value it separately.
This single change usually makes PMax’s reported performance look worse. That is the point — it is now reporting on new demand.
2. Split by margin, not by convenience
One asset group containing the entire catalogue optimises toward whatever converts most easily, which is rarely what makes the most money. Split by gross margin band or product role — hero, volume, clearance — and set targets per group that reflect what each is actually worth.
3. Feed the signals deliberately
Audience signals are a starting hint, not a target. Give PMax your converter lists and high-intent custom segments, and withhold the broad “all site visitors” lists that just point it back at existing demand.
4. Measure with a holdout
The only honest way to know what PMax contributes is to stop it somewhere. A geographic holdout for two to four weeks, or a clean before-and-after window against a stated hypothesis, will tell you more than any amount of platform-reported attribution.
What good looks like afterwards
Reported cost per conversion goes up. New-customer rate goes up more. Total revenue rises while the account’s own numbers look less flattering than they did before.
That trade — worse-looking reports, better-performing business — is usually the right one, and it is worth agreeing on it with whoever reads the monthly summary before you make the change.