Most accounts running PMax ads are not really running them. They built one campaign, uploaded whatever creative was lying around, set a target ROAS that someone picked in a meeting, and now check a single number once a week. When the number is good, nobody asks why. When it drops, nobody can explain it either, because Performance Max does not hand over the reporting you would need to answer the question.
That is the actual problem with Performance Max. It is not that the automation is bad. In a lot of accounts it genuinely outperforms the manual structures it replaced. The problem is that it removes almost every lever advertisers used to pull, so the few levers left have to be pulled deliberately and well. This guide covers which ones still matter.
What PMax actually automates, and what it does not
Performance Max is one campaign type that buys across Search, Shopping, Display, YouTube, Discover, Gmail and Maps from a single budget. Google decides the split, the placement, the bid and which asset combination to serve for each auction.
What you still control is narrower than most people assume, but it is not nothing:
- The budget and the bid strategy target (target ROAS or target CPA)
- The assets: headlines, descriptions, images, video, logos
- The product feed and which products are eligible
- Asset group segmentation
- Search themes and audience signals (inputs, not targeting)
- Negative keywords at the account level, and brand exclusions
- Conversion actions and their values
What you do not control is the placement mix, the device split, the audience that actually gets served, or which of the six surfaces takes your budget on any given day. If a campaign structure depends on controlling those, Performance Max is the wrong campaign type, and a standard Shopping or Search campaign is still perfectly legal to run.
Your conversion data is the real bid strategy
This is the part people skip, and it undoes everything downstream. Performance Max optimises toward whatever you tell it a conversion is worth. If that signal is wrong, the automation will very efficiently buy the wrong thing.
Three failures show up constantly:
Counting every conversion action as primary. Newsletter signups, PDF downloads and "contact page visits" sitting alongside purchases in the primary column teaches the algorithm that a $0 action is worth chasing. Move soft actions to secondary so they still report but do not bid.
Static conversion values. If every sale is logged at the same value, PMax optimises for order count, not revenue, and will happily fill the account with low-margin orders. Pass dynamic revenue values, and if margins vary a lot by category, pass margin rather than revenue.
Broken or duplicated tracking. Double-firing tags inflate performance, which raises the effective target, which suppresses volume. Before you touch anything else, audit the measurement layer. Our conversion tracking setup guide walks through the tag and value checks worth doing first.
Fix this before optimising the campaign. Everything below assumes the numbers going in are real.
Setting a target ROAS that is not a guess
Target ROAS is the single biggest input you give Performance Max, and it is routinely set by vibes.
Start from break-even, not from ambition. Break-even ROAS is simply 1 divided by your contribution margin. A 40% margin means you break even at 2.5x. Anything you set below that is a decision to buy revenue at a loss, which is sometimes correct for new customer acquisition but should be a deliberate choice rather than an accident.
Work out both numbers before you open the campaign settings. Our ROAS calculator gives you the ratio your current spend and revenue actually produce, and the profit margin calculator gives you the margin figure the break-even number depends on. If you are acquiring subscription or repeat customers, the LTV calculator is the more honest input, because a first-order ROAS of 1.8 can be excellent when the second order arrives reliably.
Then set the target conservatively and move it slowly. Performance Max re-enters a learning phase on large target changes, and a campaign that spends three days relearning has lost more than the adjustment gained. Change targets by 10 to 15 percent at a time and wait for a full conversion cycle before judging the result.
A target set too high is the most common self-inflicted wound. The campaign simply stops spending, the advertiser concludes PMax "does not work for us", and the budget moves back to a campaign type with worse economics.
Brand traffic is the biggest source of fake performance
If your Performance Max campaign is allowed to serve on brand searches, it will, because those are the cheapest conversions available. Your reported ROAS will look excellent. You are mostly paying for traffic that would have arrived anyway.
There are two ways to deal with it:
- Account-level negative keywords for your brand terms, applied to the PMax campaign, with a separate brand Search campaign handling that traffic at a sensible CPC.
- Brand exclusions in the campaign settings, which handle brand and misspellings without the maintenance burden of a keyword list.
Do one of them. Then compare the before and after honestly: reported ROAS will drop, and that drop is not a performance regression, it is the number becoming true. Judge the change on incremental revenue and blended new customer cost instead. If you have never separated the two, the customer acquisition cost breakdown is a useful frame for what you should actually be measuring.
Search themes and audience signals do less than the name suggests
Both of these are inputs, not targeting. Google treats them as a starting hint and then expands well beyond them once it has conversion data.
Search themes are most useful on a brand new campaign or a new product line where the feed and landing pages do not yet give Google enough to work with. Add up to 25, keep them descriptive rather than exact-match style, and accept that they decay in influence as the campaign matures.
Audience signals are worth building properly from first-party data: purchasers, high-value customers, cart abandoners, site visitors by depth. These behave much the same way remarketing lists do in other campaign types, and the same list-building discipline applies. If your lists are thin or stale, the remarketing lists guide covers how to segment them so the signal is worth sending.
What neither of these will do is confine your spend. If you need genuine control over who sees the ad, you need a different campaign type.
Asset groups: the one structural lever left
An asset group is PMax's equivalent of an ad group, and how you split them is close to the only structural decision you get to make.
Segment by margin or product economics first, and by theme second. A single asset group containing a $19 accessory and a $400 flagship product will optimise toward whichever converts more easily, which is almost never the one you want more of. Splitting by margin band lets you set different targets for genuinely different economics.
Listing groups control which products are eligible within each asset group, so use them to enforce that split rather than relying on the creative to do it.
Then feed each group enough creative to work with. Performance Max assembles combinations, and a group with three headlines and one image has almost nothing to assemble. Aim for the full complement: multiple headlines at different lengths, several descriptions, images in all three aspect ratios, and at least one video. If you do not supply video, Google generates one from your assets, and the auto-generated version is usually worse than anything you would have approved.
Watch the asset performance labels, but treat "Low" as a prompt to investigate rather than an instruction. An asset can be rated low simply because it was served rarely.
Getting visibility back
The reporting gap is real, but it is smaller than it was. Three things are worth doing every month:
Run a placement report. It is buried under the campaign's insights, and it will show you where display and video impressions actually landed. Expect the same junk inventory that plagues display generally, and exclude accordingly. The Google Display Network guide covers which exclusion lists are worth maintaining, and they apply here too.
Pull the search terms. Performance Max now surfaces search category and term data. It is less complete than a Search campaign's report, but it is enough to spot irrelevant themes and add negatives.
Use a PMax script or the API for asset group level data that the interface does not show. Several well-maintained public scripts will dump asset group performance, listing group spend and placement data to a spreadsheet on a schedule.
None of this gives you full transparency. It gives you enough to catch the obvious waste, which is most of it.
Where PMax sits next to your other campaigns
Performance Max does not coexist neatly with Standard Shopping. When both target the same products, PMax generally wins the auction, so a Standard Shopping campaign running alongside it will quietly starve.
Search campaigns are different. Exact match Search keywords take priority over Performance Max, so a well-built Search campaign keeps its high-intent traffic. That makes the practical structure: Search for known high-intent terms and brand, Performance Max for everything else, and no Standard Shopping campaign fighting it for the same inventory.
For a store starting from a clean account, the sequencing that works is to get the account and tracking right first, launch Search on the terms you already know convert, then let Performance Max take the discovery and long tail. Our Google Ads account setup guide covers the foundation, and PPC for ecommerce covers how the campaign types are meant to divide the work.
A 30-day plan for a campaign that is underperforming
If you have inherited a PMax campaign that is not working, resist the urge to change six things at once. Performance Max punishes that more than most campaign types, because you lose the ability to attribute the result.
- Days 1 to 3. Audit conversion actions and values. Fix primary/secondary assignment. Verify tracking fires once. Change nothing else.
- Days 4 to 7. Apply brand exclusions or brand negatives. Record the new baseline ROAS and accept that it is lower and more honest.
- Days 8 to 14. Split asset groups by margin band. Fill every asset slot in each group. Add a real video.
- Days 15 to 21. Pull placement and search term reports. Add exclusions and negatives. Let it run.
- Days 22 to 30. Only now adjust the target ROAS, in one 10 to 15 percent step, in the direction the data supports.
One change per window, with a full conversion cycle in between. If your sales cycle is longer than a week, stretch each window accordingly.
Before the target change in the final week, check that the landing page is not the constraint. A campaign hitting a 1.2% site conversion rate does not have a bidding problem. Run the numbers through the conversion rate calculator and compare against your category norm, and if the gap is large, the fixes in our ecommerce CRO guide will move the account further than any bid adjustment.
Mistakes that cost the most
- Setting a target ROAS above break-even by a wide margin on day one, then concluding PMax cannot spend
- Leaving brand traffic in and reporting the inflated ROAS to a client or a board
- One asset group for the entire catalogue
- No video asset, so Google generates one
- Judging the campaign in week one, during learning
- Optimising toward a conversion action worth nothing
- Running Standard Shopping alongside it and wondering why one campaign stopped spending
FAQ
How long does Performance Max take to exit the learning phase? Usually one to two weeks, and it is driven by conversion volume rather than calendar time. A campaign generating 30 conversions a week stabilises much faster than one generating three. Any significant change to budget, target or asset groups restarts it, which is why the 30-day plan above spaces changes out.
Can I stop PMax ads from showing on Display and YouTube? Not directly. There is no channel opt-out within Performance Max. You can exclude specific placements and content categories, which trims the worst of it, but if you need to buy Search only, run a Search campaign.
Is target CPA or target ROAS better for PMax? Target ROAS when you have reliable revenue values and variable order sizes, which covers most ecommerce. Target CPA when order value is roughly fixed or you are generating leads rather than sales. Starting on maximise conversions or maximise conversion value without a target is reasonable for the first two weeks, then add the target once you have data to base it on.
How many asset groups should one campaign have? Enough to separate genuinely different product economics, and no more. Three to five is typical for a mid-size catalogue. Every extra group divides the conversion data further, and groups that never accumulate enough volume to learn will underperform the single group they were split from.
Does Performance Max cannibalise my Search campaigns? Exact match Search keywords take priority, so a well-structured Search campaign holds its high-intent traffic. Broad and phrase match terms are more contested. The clean fix is to keep Search focused on terms you know convert, and let Performance Max cover discovery.