The levers that still exist
| Lever | What it actually does | How much it moves |
|---|---|---|
| Conversion values | Redefines what the bidding is optimising towards | Most of the effect |
| Asset quality and coverage | Decides what can be assembled, and where it can serve | Large |
| Audience signals | Seeds the exploration. Not targeting | Moderate, early on |
| Campaign and budget structure | Separates what should not be averaged together | Moderate |
| Brand exclusions and negatives | Stops it buying traffic you already had | Large, often ignored |
| Feed quality, for retail | Determines what it can sell and how well | Large |
Nothing on that list is a setting you nudge. They are all inputs you supply, and the campaign decides what to do with them. That is the actual shape of the thing: you are not driving, you are writing the brief.
What Google took away
Placement-level control, search term visibility beyond a summarised view, device and audience bid adjustments, and the ability to see which channel inside Performance Max produced which result. Reporting has improved since launch and is still narrower than a set of separate campaigns would give you.
This is worth stating plainly because most "PMax is a black box" complaints are accurate about the reporting and then wrong about the conclusion. Limited visibility into what it did is not the same as no influence over what it does.
The lever almost nobody pulls
Performance Max optimises towards the conversion value you report. Most advertisers report revenue. Revenue is not what the business is trying to maximise.
Send margin instead of revenue, or a value that accounts for returns, or one weighted for new customers over repeat ones, and the campaign starts buying something different, without a single setting being touched. This is the largest available change in how Performance Max behaves and it is made outside the ad account entirely, in whatever sends the conversion.
It also explains the black box feeling. If you report revenue and the campaign maximises revenue, it is working exactly as instructed while making the business worse, and nothing in the interface will tell you.
The brand traffic problem
Performance Max will buy your own brand searches and report them as conversions, because they convert well. They would have converted anyway. Without brand exclusions applied at the account level, a chunk of the reported performance is traffic you already owned, and the campaign looks better than it is.
When to take something out of it
- When one product line has different economics. Averaging a high margin line with a low margin one hides both.
- When you need query control. A regulated or trademark-sensitive category is not a good fit for a campaign type that will not show you every query.
- When testing. Isolating a change inside Performance Max is close to impossible, and a standard campaign will answer the question faster.
How to tell whether it is working
Not by its reported ROAS, which is scoped to itself and includes the brand traffic. Judge it on whether total new customer acquisition moved while total spend held, over a period long enough to clear the learning phase. That is a business measurement, not a Google Ads one, and it needs ad data sitting beside order data. Which is the same reason Google Ads and GA4 will never agree: each platform can only report on itself.