Performance Max is the campaign type Google would prefer you to run, which is exactly the reason most account audits start there. It's also the campaign type that's hardest to audit honestly, the asset group reporting is sparse, the placement breakout is sparser, and the algorithm makes most of the meaningful decisions for you. The standard PMax dashboard tells you that the campaign is working. It rarely tells you whether it's working well.
We've inherited enough PMax campaigns to have built a standard 48-hour diagnostic protocol, six queries we run before recommending any changes. They're not exhaustive. They are designed to surface the structural problems that, in our experience, account for the bulk of the lift available on most under-performing PMax accounts.
The six diagnostics
01Brand vs. non-brand spend split
Run a search-terms report (yes, PMax does have one now, even if it's hidden) and bucket every term into "brand" or "non-brand." Sum the spend by bucket. The most common pattern we find on under-performing accounts: 60-80% of total PMax spend is going to branded queries, where it's cannibalizing organic clicks the user would have made for free.
Healthy non-brand share for a mid-market eCom PMax campaign is usually 50-70%. If non-brand is below 30%, you have a structural cannibalization problem masquerading as great ROAS.
02Asset group concentration
How many asset groups in the campaign? Of those, how many have driven >20% of conversions in the past 30 days? We routinely find PMax campaigns with 6-8 asset groups where 90% of conversions come from one. The other asset groups are eating budget for the algorithm to "explore", sometimes legitimate, often wasted.
If one asset group is doing 90% of the work, the diagnostic question is: are the others contributing learning, or are they bleeding spend? You usually can only tell by pausing them and watching what happens to the dominant group's CPA.
03Audience signal performance
PMax lets you provide audience signals, first-party customer lists, custom segments, in-market audiences, to "guide" the algorithm. Most accounts we inherit have either no audience signals at all, or have everything thrown in indiscriminately.
Pull the audience signal performance breakdown (Insights → Audiences). Look for signals where the algorithm is leaning heavily on a small subset and ignoring others. The diagnostic question: are the signals you're providing telling Google useful information, or are they redundant with what Google already infers? Lean signals (a small list of high-value customers) usually outperform broad ones (every contact in HubSpot, ever).
04New customer acquisition value mix
Turn on "New customer acquisition value" if it isn't on. Look at the mix of new vs. returning customer revenue. Most accounts we audit have PMax skewing heavily toward returning customers, which is, again, structurally great-looking ROAS that's mostly capturing demand that already existed.
If returning customer share is >60% of PMax revenue, you have a customer-acquisition-cost problem hiding inside what looks like an efficient program. The fix is usually a New Customer Acquisition value goal, not a campaign rebuild.
05Conversion lag distribution
In the conversion paths report, look at the distribution of time-to-conversion for PMax-attributed conversions. Heavy left-skew (most conversions within 1-3 days) is typical for retail/eCom and means your tracking is catching the conversion close to the click. Heavy right-skew (long tail of 14+ day conversions) for a high-intent product is a flag, usually means your conversion attribution is over-counting because users were going to convert anyway.
This one's diagnostic for whether the campaign is driving incremental conversions or just claiming credit for converters who would have showed up direct. Pair it with a hold-out test if budget allows.
06Listing group / Shopping feed coverage (if applicable)
For accounts with a Shopping feed, pull the product-level performance report. What percentage of SKUs are getting impressions? What percentage are getting clicks? What percentage are getting conversions?
The pattern we find most often: 80/20 distributions all the way down, where 20% of SKUs are doing 80% of the work, and 50%+ of the catalog isn't getting served at all. Sometimes that's right (those are the products people actually buy). Often it's a feed-quality problem, missing GTINs, weak titles, no product attributes, that's preventing the algorithm from finding query/product fit.
The diagnostic question: of the products that aren't getting served, which ones should be? That's usually the highest-leverage feed work in the account.
Reading the results together
Each of the six diagnostics produces a number. The diagnostics individually are useful; the pattern across them is more useful. The most common combinations we see:
- "Looks great, isn't great" pattern, Strong ROAS reported. Brand share > 70%. Returning customer share > 60%. Conversion lag heavily right-skewed. Translation: the campaign is mostly capturing demand that would have converted anyway. Lift opportunity: shift spend away from PMax and into a campaign type that drives genuine new acquisition.
- "Concentrated and brittle" pattern, One asset group does everything. Audience signals are stale or generic. Catalog coverage is narrow. Translation: the campaign is dependent on a single learning loop and will collapse if you change anything significant. Lift opportunity: structurally rebuild with deliberate diversification before scaling spend.
- "Starved for signal" pattern, No audience signals, no New Customer goal, conversion tracking incomplete. Translation: the algorithm is doing its best with the equivalent of a blindfold. Lift opportunity: feed it better data, usually a 2-4× ROAS improvement available without changing anything else.
Most accounts fit one of these patterns. About 1 in 5 fits two simultaneously. We've yet to encounter a meaningfully under-performing PMax campaign that didn't show up clearly in this protocol within the first day or two.
None of these diagnostics involve raising a bid. They're structural reads on whether the campaign is set up to succeed in the first place. Bid strategy adjustments without these reads are guessing.
What we don't do in the first 48 hours
Equally important to the diagnostic protocol is what we deliberately leave alone. Three things we don't touch in week one:
- Conversion goals. Don't change the optimization target until you understand what's currently being optimized for. Account managers love changing this; it almost always sets the algorithm back two weeks of learning.
- Asset rotation. Even if the creative is mediocre, swapping it triggers a learning reset. Wait until you have a structured creative test plan, then swap deliberately.
- Budget caps. If the campaign is hitting its budget cap daily, that's information; if it's not, raising or lowering the cap teaches you nothing. Match budget to learning needs, not to "growth" goals, at least at first.
The instinct on inheriting an account is to "fix things" by changing knobs. PMax punishes that. The campaign type is essentially asking you to pay for learning; every change resets the learning. The right move on week one is to diagnose, and only intervene where the diagnostic is unambiguous.