Performance Max is the most profitable campaign type on Google... for Google. A single campaign type that buys on Search, Shopping, Display, YouTube, Gmail, Discover and Maps, with an official report that for years boiled down to one total figure and little else. This lesson teaches what the interface does not: where your money is really going and how to act when you don't like the split.
Why PMax is a black box (on purpose)
The implicit deal in PMax is: "give me budget, creatives and a goal, and don't ask how I do it". The problem is that inside that box live channels of radically different quality:
- A Search click on your own brand can be worth gold... or be traffic that was already yours for free (brand cannibalisation).
- The Display inventory ranges from serious news publishers to farms of websites built to collect impressions — the hunting ground of our Ninja Shield.
- YouTube's view-through conversions can inflate the results if your measurement counts them too generously.
With the same apparent ROAS, two PMax campaigns can be a gem and a haemorrhage. The difference only shows when you open the box.
How to estimate the split by channel
Google does not give you the breakdown directly, but it leaves cracks. Combining three of them, the split can be reconstructed accurately enough to make decisions.
1. The segments that do exist in the interface
PMax reporting has been improving: today the interface itself offers partial breakdowns (asset groups, PMax search terms, the placements where you showed). They do not answer "how much did you spend on each channel", but they narrow it down: search terms tell you what share responds to real intent, and placements give away the Display share.
2. The arithmetic of the formats
Every row of statistics leaves fingerprints: Shopping impressions carry an associated product, video reports views, Display has characteristic CTRs (very low) and Search has high CPCs with a high CTR. Cross-referencing metrics by date lets you estimate the weight of each format — this is the "forensic" approach: nobody tells you, you deduce it from the remains.
3. GAQL: the big crack
The API (and Google Ads scripts, which use it) exposes resources the interface does not show, or shows only halfway. Querying statistics by network type and the PMax placement listings gives you a breakdown the standard screen does not provide. It is the most reliable route, and the one you can automate every night.
💡 Ninja trick: this analysis is unworkable by hand every week — and it is exactly what we automate. Our scripts read via GAQL what the interface hides and paint it into a sheet: estimated split by channel, the list of Display placements where you appeared and their estimated cost. When Ninja Shield finds a farm of websites in that list, it excludes it at account level: the only exclusion lever PMax really respects.
The levers you do have in PMax
Once the split is known, this is what you can touch (and what you cannot):
| Lever | What it achieves | Real limit |
|---|---|---|
| Brand exclusions | Take your brand searches out of PMax | The most important lever: without it, PMax claims conversions that were already yours |
| ACCOUNT-level placement exclusions | Ban junk sites and apps on Display | The account list does apply to PMax; campaign-level exclusions do not always |
| Campaign-level conversion goals | Stop PMax chasing soft conversions | Requires a clean conversion architecture (the Measurement module) |
| Asset groups by intent | Separate messages and signals | It is not a spend breakdown, but it tidies up the learning |
| Conversion value by type | Make it optimise towards real money and not towards forms | Demands offline conversions or differentiated values |
The general rule: in PMax you do not steer the spend, you steer the incentives. Everything you can measure better (real values, clean conversions, exclusions of what never converts) is a well-placed incentive; everything you leave ambiguous, PMax will resolve in favour of spending more.
The typical case: brand inside, farms inside
The pattern that repeats most often when you open the box in real accounts:
- 20-40% of the spend on your own brand — conversions that organic search would have brought in for free. Fix: brand exclusions + measure the incremental, not the attributed.
- A tail of Display on irrelevant sites — pennies per click, thousands of clicks, zero sales. Fix: an account-level exclusion list, fed automatically.
- The rest, working fine. PMax is not the enemy: without supervision it is a partner that charges itself an excessive commission; with supervision, a volume machine.
What you should remember
- PMax's aggregate ROAS is not information, it is an average of channels that cannot be compared. Without a breakdown there is no diagnosis.
- The breakdown can be estimated: interface + format fingerprints + GAQL. And it can be automated.
- Your levers are incentives: brand exclusions, account exclusions, clean conversions and values.
- Before judging a PMax campaign, ask it three things: how much of this is my brand?, where is it treading on Display?, and which conversion is it really optimising towards?