Performance Max for online shops: what it does and what it hides

PMax is judged on attributed value

Performance Max buys all of Google Ads' inventory from one campaign and steers towards your conversion goals while doing so. The figure you use to judge success is the conversion value attributed to it. That is a different quantity from the revenue you would not have had without the campaign.

My view: almost every argument about PMax comes down to this gap. As long as a purchase that would have happened anyway counts just as much as an additional one, the figure also rewards demand for your brand and purchases by existing customers. That does not make the campaign bad or the report wrong. It does change the question you should ask of every setting: does it change what counts as success, or who is not allowed to count at all, or is it merely a hint to the system?

How much baseline revenue a campaign can carry

A high ROAS proves no profit. It only buys you headroom for revenue that would have arrived without the ad. You can calculate how much headroom:

Tolerable baseline share = 1 − (break-even ROAS ÷ reported ROAS)

A worked example with invented, round numbers: suppose a shop makes a 40% margin before ad spend, so its break-even ROAS is 2.5. The campaign reports €50,000 in conversion value on €10,000 of spend, a ROAS of 5. On paper that leaves a contribution of €10,000. If €30,000 of it would have come anyway, €20,000 of incremental revenue remains, which is €8,000 of margin against €10,000 of cost: a loss of €2,000 at a reported ROAS of 5.

This turns "Is the ROAS good enough?" into a question you can actually discuss. At a ROAS of 5 and a break-even of 2.5 it reads: is it plausible that more than half of this revenue would be there without the campaign? The calculation looks only at the single purchase. If you win new customers who buy again often, you can be more generous. No account report shows you the real share. Only a test does.

Where the ads run, and how little of it you control separately

According to Google, the campaign reaches YouTube, Display, Search, Discover, Gmail and Maps. For online sales goals your feed is the starting point, and it produces, among other formats, Shopping ads on Search, YouTube and Maps as well as dynamic remarketing on Display. One campaign therefore covers both prospecting and remarketing. That is its strength for small teams, and also the reason the total figure blends so many different things.

You can steer little of it separately. Channel prioritisation keeps Search, YouTube, Maps, Gmail and Discover bundled together. Separate priority tiers and optional daily budgets exist only for search partners and the Display Network. You cannot use this control to set a fixed YouTube budget.

Relative to Search campaigns, there is a clear rule: for an identical query, exact match takes priority over PMax, while search themes rank alongside phrase and broad match. A query you want to control yourself therefore belongs in a Search campaign as an exact match keyword.

If a channel shows no delivery and no fault, Google says this usually means the campaign is bidding on channels with a higher expected return at that moment. I read that as follows: the channel mix also follows your success metric. A large share of Shopping on brand searches then tells you something about where conversions are easy to get, and little about where they are additional.

Which inputs change the incentive and which are only hints

The list of settings keeps growing, but it works on four levels. The grouping is my own.

Hints are not worthless. They just do not correct a wrong measure of success. The customer list shows this most clearly. Google calls Customer Match the most valuable audience signal, and at the same time signals are not hard targeting, so ads can also run outside them. The same list can also serve as an exclusion. As a signal it says: look for people like these. As an exclusion it says: these people get no ad. If you only set the signal and hope for more new customers, you have not changed the incentive.

Search themes work the same way. Google shows an indicator of whether they bring additional traffic beyond your URLs and assets. That means traffic the campaign would not otherwise have found. It says nothing about additional revenue for your shop, even though Google uses the same word, "incremental".

Four levers I would check in this order

Conversion values. They come first because every other input builds on them. Suppose two products each cost €100, one with a 50% margin and one with 15%. Reported as revenue, both are worth €100, and the campaign has no reason to prefer the first. Google recommends that retailers use transaction-specific values and points to conversion value rules for different margins and customer value. How to do this is covered in margin as conversion value. If you report margin instead of revenue, break-even is 1 by definition and your previous target ROAS no longer fits. The limit: if your margin data is unreliable, you train the system on errors, and revenue remains the more honest measure.

New customer goal. It acts directly on the success metric. Start with the new customer value mode, which Google recommends for online sales: it favours new customers and keeps reaching existing ones. According to Google, new customer only makes sense only in some situations, for example with a dedicated new customer budget, and can cost you purchases from existing customers. Derive the estimated average revenue from new customers that you enter during setup from your repeat purchase data, because it shifts the incentive directly. And a new customer may also have bought without the ad: the goal narrows the gap, it does not close it.

Brand exclusions. First look at how much brand is in play. An agency's analysis of its own accounts found that 38% of 84 campaigns had more than 20% of clicks on search categories containing the brand name. It is not representative, but it is a reason to look at your own account. If you want to exclude brand, this is the clean way to do it: according to Google, unlike negative keywords it also blocks obvious misspellings, and with a feed you can limit it to Search text ads and keep brand traffic for Shopping. Brand exclusions, like negative keywords, only apply to Search and Shopping inventory. Excluding brand should not be a reflex; the test case below shows why. Details are in brand in Performance Max.

Customer list exclusion. This is the hardest way to direct budget towards new customers. It pays off when your existing customers come back without an ad. With a range that is bought again at short intervals and faces strong price competition, I would be more careful, because exactly these buyers may go to a competitor. List quality matters too: anyone who is not on the list, or is not matched, stays in play. According to Google, the exclusion only applies to Search, Display, Shopping, Video and Discover inventory, and for pure new customer acquisition Google itself recommends the new customer goal, because it offers more reporting and value settings.

Separate campaigns only come up after that. Google mentions separate campaigns where budgets or ROAS targets differ, and custom labels in the feed provide the interface. From a retailer's point of view this fits when margin classes genuinely need different targets and you cannot express them through values. More campaigns on their own do not change the success metric.

What the reports show and what stays invisible

Visibility is better than older articles claim. There is a channel report, search term insights showing whether a query came from search themes or keywordless serving, and reports on budget, audiences and placements. This is as of November 2026, and the features change quickly. These reports answer the question of where conversions were attributed. They do not answer which of them would have been lost without the ad.

Google advises judging performance at campaign level, because PMax optimises for marginal ROI and the averages of individual segments do not tell the whole story. I find the argument against judging individual channels convincing. It does not solve the actual problem, though, because the campaign ROAS also rests on attributed conversions. A larger sum of the same metric is no proof of incremental revenue.

Two checks in the account are still worth doing:

• In the channel report, set the filter to "Conversions", not "Results". "Conversions" shows what the campaign optimises for, "Results" also includes your other goals. Only the first view tells you anything about the incentive.
• In the search term insights, work out the share of clicks containing your brand name and compare it with the calculation above. If the reported ROAS is close to break-even and the brand share is high, the headroom is probably already used up. This remains a rough approximation, because some brand purchases are additional too.

Why only a holdout answers the question

Only a comparison with regions or groups without the campaign gives you the share the formula needs. One publicly documented case argues against simple answers: Caraway, a US cookware brand, ran a three-week geo holdout measured on total revenue in its data warehouse. According to Haus, Google overstated PMax with brand by 33%. At the same time, PMax with brand delivered 1.95 times as much total revenue as PMax without brand. So the platform exaggerated, and the obvious conclusion, "take brand out", would still have been wrong in this case.

This is an older single case, published by the test provider. Across several tests, Haus reports, according to ppc.io, an average of 24% more incremental revenue from the exclusion, which points the other way. Only the method transfers: measure a correction factor, apply it to the reported ROAS, and hold the result against break-even. The instructions are in the geo test guide for retailers in the DACH region (Germany, Austria and Switzerland). If your revenue is too small for a meaningful result, use the table above as a stress test: up to which baseline share does the campaign pay off, and how likely is that share for your brand?

When you can spare yourself the scepticism

The gap is not equally large in every shop. That is my assessment, not a measured rule. A young brand with little brand search and infrequent repeat purchases has little demand that PMax could claim for itself. There, from around 60 conversions a month, the campaign is often a sensible default, because it covers a lot of inventory with few staff and the reported value is likely to be closer to incremental revenue.

It is different with a well-known brand, a high share of existing customers and a ROAS that barely moves when you change the budget. There I would no longer justify any budget increase with the campaign ROAS until a test has been run. If the conversion values themselves are still unreliable, that also takes priority over any debate about channels and signals. What remains for you to decide is the budget level at which an unmeasured baseline share becomes too expensive.