Performance Max is a single Google Ads campaign that spends across every Google surface at once, chooses where each impression goes by itself, and reports back mostly in aggregate. You give it a budget, a conversion goal, a product feed and a pile of assets. It decides the rest.
That is genuinely useful when your feed is clean and your conversions are measured correctly, because the machine is better than a human at deciding which surface earns the next euro. It is actively harmful when either of those is broken, because the campaign will still spend the budget and still report a number, and the number will look fine.
What it will not tell you
The reporting is the trade. You do not get a clean search terms report, you cannot see spend split by surface the way you can in separate campaigns, and asset level rows are not what a beginner assumes: the same click gets attributed to every asset that was on screen, so adding up the asset rows produces a number far larger than the campaign actually delivered. Read the campaign total, never the sum of the parts.
The campaign will spend a broken budget as confidently as a working one. Nothing in the interface distinguishes the two.
The feed decides most of the outcome
If your feed is in the campaign, Shopping inventory usually takes the majority of the spend, because it is the cheapest conversion available and the machine is optimising for conversions. That has a consequence people miss: turning a feed on inside a campaign whose budget is already capped is a redistribution, not an expansion. The Shopping share goes up, something else goes down, and the blended return moves without a single new customer arriving.
So before judging a Performance Max campaign, check what fraction of its spend went to Shopping and whether that fraction changed. A return that fell after a feed went live is usually the mix moving, not the products failing.
When it is the wrong campaign
Three cases, and they are common. First, when conversion tracking is not verified end to end, because the campaign optimises to whatever it is told and a mismeasured signal trains it toward the wrong customer. Second, when the catalogue is small or margins vary wildly across it, because a single goal cannot know that one product is worth four times another. Third, when the account has no branded search protection, because a campaign told to find conversions will happily buy the ones that were going to happen anyway.
In all three the fix comes before the campaign, and it is usually a measurement job rather than a bidding job.
How to judge it after a month
Look at total account revenue and total account spend, not the campaign card. Performance Max is very good at claiming conversions that other campaigns or channels would have produced, so a campaign level return of four means nothing if the account level return did not move. Then check how many products actually got impressions against how many are eligible, because approved in Merchant Center is not the same as being shown, which is its own problem.
What we do with it
We verify the tracking before touching the campaign, then the feed, then the structure. Most of the return in a Google account is sitting in those two, and the bidding argument that everyone wants to have first is usually the smallest of the three. How we run the channel is on the Google Ads page.