Performance Max in Google Ads: how it works and who it suits
Performance Max is the most contentious format in Google Ads. Some call it a black box that eats budget and never reports where it went. Others show campaigns with a ROAS above 1,000%.
Both sides are right. The difference lies not in the tool but in what it was given as input.
We run PMax in e-commerce and physical products daily, and we see the same pattern every time: PMax is neither «smart» nor «stupid» — it is exactly as effective as your conversion data and product feed are clean. In this article we break down how the format works from the inside, who it suits, who should avoid it, and show across two of our cases what correct configuration delivers.
What Performance Max is and how it differs from other campaigns
Advertisers used to allocate budget across campaign types themselves: search separately, shopping separately, display and YouTube separately. Each channel had its own bids, its own audiences, its own reporting.
Performance Max removes those partitions. A single campaign gets access to all of Google’s inventory at once:
- Search
- Google Shopping (shopping ads)
- YouTube
- The Display Network
- Gmail
- Discover
- Maps
You no longer choose where to appear. You give the system four things — a goal (a conversion), a budget, a product feed and a set of assets — and the algorithm decides in real time who to show the ads to, where and at what price.
This fundamentally changes the specialist’s role. In a search campaign the work is bids, keywords and negative keywords. In PMax the work is the quality of the input data. The algorithm will not fix a dirty feed, and it will not guess that your «conversion» is actually a view of the contacts page.
| Search campaign | Performance Max | |
|---|---|---|
| Query control | full (keywords + negatives) | limited (search themes, exclusions) |
| Channels | Search only | all Google inventory at once |
| What the specialist optimises | bids, keywords, ads | feed, assets, signals, target conversion |
| Data volume required | works at low volumes too | needs steady conversions to learn |
| Reporting transparency | high | partial (less channel-level detail) |
| Where it is strong | existing demand, narrow semantics | large catalogue, finding new demand, scale |
How PMax makes decisions: four levers that genuinely work
The most damaging myth about PMax is that «there’s nothing to configure». In reality you do steer the campaign — the levers have simply moved.
1. The target conversion. This is the most powerful lever of all. The algorithm optimises for whatever you have called success. If five events are active in the account simultaneously — purchase, add to cart, phone click, page view, sign-up — the system learns from a mixture of signals of differing value and loses focus. One clear event produces far cleaner learning.
2. The product feed. For e-commerce, the feed is your targeting. Names, categories, attributes and custom labels determine which queries your products appear for and to whom. An unsegmented feed of several thousand items means the budget spreads in an even layer across the whole catalogue, including products you have no interest in promoting.
3. Audience signals. These are not hard targeting but a hint to the algorithm about where to start looking. Remarketing lists, data from Google Analytics, custom segments based on interests and search behaviour — all of it speeds up the exit from the learning phase. Without signals the campaign learns more slowly and more expensively.
4. Assets. Headlines, descriptions, images, video, logos. The algorithm assembles combinations from them for each placement. Few assets means few combinations, which means narrower reach. An asset group with no video at all means the system will generate one itself from your images, and the result is usually worse than a proper clip.
Bidding strategy is a story of its own. «Maximise clicks» in PMax is almost always a mistake for a store: you will get traffic, not sales. Target ROAS (tROAS) or target cost per action (tCPA) tie the algorithm to economics rather than volume.
PLAY VINYL case: ROAS from 791% to 1091% after moving to PMax
PLAY VINYL is a premium audio brand: vinyl records, speaker systems and related equipment. Before us the account had been run by several agencies, and for a time it ran on its own, on previously launched campaigns. The client’s brief was specific: reduce cost per conversion.
What we found in the account: some campaigns had been spending budget for months with almost no conversions, and optimisation was running against several conversion events at once — in other words, the algorithm was being prevented from learning.
The statistics for the 6 months before our work:
- CTR 1% · CPC ₴8 · CPM ₴101
- Conversions: 846 · cost per conversion: ₴939.4
- ROAS: 791% · average budget: $3,200/month
What we did:
- Switched off the ballast — every inefficient campaign that was draining budget.
- Moved to Performance Max. On the available statistics, search campaigns in this niche were performing poorly, so we placed the main bet on PMax.
- Focused on a single conversion event and changed the bidding strategy from «maximise clicks» to target ROAS.
- Segmented the product feed: all 5,910 products received labels by genre, brand and other attributes; ads were split by product group so we could identify the most effective sections of the catalogue.
- Audiences: remarketing to visitors who had not purchased, PMax audience signals based on interests from Google Analytics, custom audiences (people browsing competitors’ sites or searching for «vinyl records», «speaker systems»).
The result:
| Metric | Before | After |
|---|---|---|
| CTR | 1% | 2% |
| Conversions | 846 | 1,509 |
| Cost per conversion | ₴939.4 | ₴565.84 |
| ROAS | 791% | 1,091% |
Note what actually produced the effect. Not «PMax magic» — but one conversion instead of several, target ROAS instead of clicks, and a segmented feed. The same PMax launched on a dirty account with five conversions and a feed in one undifferentiated block would have produced the opposite result.
TOP TREND case: PMax + DemandGen in physical products with no analytics access
The second case involves a different niche and considerably harsher conditions. TOP TREND is a product business: homeware and garden goods, car accessories, small electronics. The task was to reach a cost per conversion within a KPI of ₴150–450 and scale lead volume.
The starting conditions were difficult: the client had previously worked with several agencies, there had been a series of moves between accounts (because of spending limits and a change of legal status), and at one point the account was temporarily blocked. Most significantly — there was no access to Google Merchant Center or Google Analytics, because the client is an intermediary.
What we did:
- Performance Max + DemandGen as the main sales channels. Search was used sparingly because of the high cost per click in these categories.
- A separate landing page and separate campaign for each product. This made it possible to compare performance accurately between products rather than guessing from an averaged figure.
- Three ad groups: broad audience, interest and demographic signals, and custom segments (similar sites, search queries, behaviour).
- Custom audiences built on site visitors and dynamic remarketing through DemandGen.
The result: 5,000+ leads, a realised budget of ₴1m+, average cost per lead of ₴297.86 — within the target KPI. The PMax + DemandGen combination significantly outperformed search campaigns on price-to-performance.
What to take from this case: PMax can be brought into KPI even without Merchant Center and Analytics — but at the cost of considerably more manual work on structure. Splitting by product and landing page replaced the transparency that analytics would normally provide.
Who Performance Max suits, and who it does not
The honest answer: PMax is not a universal tool, and selling it as «launch it and it works» is dishonest.
PMax usually delivers if:
- You have e-commerce with a real catalogue — from a few hundred items upwards, with a correct feed in Merchant Center.
- There is a steady volume of conversions — the algorithm has something to learn from, rather than 3 orders a month.
- Correct tracking is in place for purchases or leads, with conversion value passed through.
- You want to scale beyond already-exhausted search demand — PMax is good at finding demand you were not covering with keywords.
- Clicks in your niche are expensive in search, and you need cheaper channels (as at TOP TREND).
PMax is better not made your main channel if:
- Conversions are critically few — services with a long deal cycle, narrow B2B, a product with a handful of sales a month. The algorithm will not exit learning.
- There is no proper tracking — optimising for «viewed the Contacts page» will reliably produce junk.
- You need tight query control — legal, medical and narrow technical niches where an appearance on an adjacent query equals wasted money.
- The budget is tiny. Spreading $150/month across all of Google’s inventory is not a test, it is noise. At those sums, focused search is better.
- You expect full transparency. Channel-level detail in PMax is limited. If you make decisions purely on «how much went to YouTube», the format will frustrate you.
5 mistakes that make PMax burn budget
In our experience, almost every failing PMax campaign we have taken over broke on the same list.
1. Several target conversions at once. The most common mistake — and exactly what was happening in the PLAY VINYL account before us. The algorithm receives a mixture of signals of differing value and optimises for the cheapest rather than the most valuable. One event means cleaner learning.
2. An unsegmented product feed. Several thousand products as a single mass means you are steering nothing. Labels by brand, category, margin and availability are your real targeting in PMax.
3. Brand traffic not excluded. PMax readily takes the cheapest conversions — and the cheapest are usually branded ones that would have arrived without advertising anyway. The campaign reports an excellent ROAS while the business sees no increase in sales. Brand queries should be separated out.
4. Intervening too early. The campaign needs time to learn. Switching off groups, cutting budgets and rewriting assets on day three means resetting learning to zero each time. It is the most expensive form of «optimisation».
5. A thin asset set. Three headlines, two images, no video — and the system simply has nothing from which to assemble combinations for different placements. Reach narrows, and auto-generated video performs worse than a proper clip.
How we launch Performance Max at VELAR
The sequence we work through on every new account, regardless of niche.
Step 1. Tracking audit — before any launch. We verify that a purchase is genuinely recorded as a purchase, that conversion value is passed through, and that there is no event duplication. If tracking is broken we fix it, and only then talk about campaigns. Launching PMax on broken analytics means paying Google to learn from falsehoods.
Step 2. One target conversion. We choose a single event to optimise for and remove the rest from the campaign goals.
Step 3. Clean and segment the feed. Labels by category, brand and margin. Separate product groups mean separate asset groups. That is how we see which sections of the catalogue actually earn.
Step 4. Remove the ballast from the account. We switch off campaigns that have spent budget for months with no conversions. This almost always produces the fastest gain in efficiency, simply through reallocating money.
Step 5. Give the algorithm signals. Remarketing, Analytics data, custom segments based on competitors and search behaviour. Not for control — to start learning faster.
Step 6. Bidding strategy tied to economics. Target ROAS for a store, target cost per action for lead generation. Not «maximise clicks».
Step 7. Leave it alone for 2–3 weeks, then scale the winners. Let it exit learning, analyse by product group, and raise budget where the economics work.
VELAR’s position: PMax is not an autopilot, it is an amplifier
Our position is simple and slightly unwelcome for anyone looking for a «make it good» button.
Performance Max does not make advertising smarter — it multiplies whatever you put into it. Clean tracking, one conversion and a segmented feed multiply into a ROAS of 1,091%, as at PLAY VINYL. Dirty analytics, five goals and a feed in one undifferentiated block multiply into a burned budget — with the same interface and the same buttons.
So when we are asked «should we launch PMax», we almost never answer straight away. First we look at the tracking and the feed. Very often it turns out that the correct first action is not launching a new campaign but two weeks of putting the data in order. It is less impressive, but it is exactly what later produces a difference of several hundred percentage points of ROAS.
And one more thing: PMax does not replace search. In most e-commerce, a healthy account is PMax across the catalogue plus a separate, controlled search campaign for brand and high-volume commercial queries. Betting everything on one format is not a strategy, it is a simplification.
Want to understand whether your account is ready for Performance Max — and what exactly is burning budget right now? We examine this in a free Google Ads audit: we look at tracking, campaign structure and the feed, and show you exactly where money is being lost.
Read also:
- «Meta Ads or Google Ads: which to choose for your business»
- «What ROAS is and what counts as a good figure»
Frequently asked questions
What is Performance Max in simple terms?
Performance Max (PMax) is a Google Ads campaign type that simultaneously shows your advertising across Search, the Display Network, YouTube, Gmail, Discover and Maps. You do not manually choose placements, bids or audiences: you give the system a feed, a set of assets (copy, images, video), audience signals and a goal — and the algorithm distributes the budget across channels itself.
Performance Max or a search campaign — which is better?
They are not replacements for one another. Search wins where demand already exists and you need tight control over queries. PMax wins where there are many products, a quality feed and enough conversion data. For most e-commerce the right answer is both: brand and high-volume search separately, PMax across the whole catalogue. At PLAY VINYL we did the opposite and dropped search, precisely because the statistics in that niche showed poor performance.
How many conversions does PMax need to work?
The algorithm needs a volume of data to learn from. As a guideline, a steady few dozen conversions a month on ONE target event. If conversions are in single figures, or spread across five different events, PMax never exits the learning phase and behaves unpredictably. In that case it is better to start with search and accumulate data.
Why does Performance Max burn budget?
The most common causes: optimising for several conversion events at once, a dirty or unsegmented product feed, no exclusion of brand traffic, too few assets in the group, and intervening in the campaign before it has exited learning. Almost always the problem is not PMax itself but the data it was fed.
Can I control where PMax spends the money?
There is no direct channel switch — and that is the main complaint about the format. But there are levers: segmenting the feed into product groups, excluding brand terms, account-level negative keywords, separate campaigns for different margin groups and the choice of target conversion. Those are how we steer PMax, not sliders.
