- You are running Performance Max and seeing the same handful of products take up most of the budget, month after month
- You have heard the advice "consolidate everything into one PMax and let Google handle the rest," and you have started to doubt it
- You have products with real potential that never get exposure because the budget is already allocated
- You want to understand when it makes sense to split up a Performance Max campaign, and when it does more harm than good
Performance Max is built to aggregate data. Google is right that Smart Bidding learns faster with more conversions in the same campaign than with data spread across ten small campaigns.
The problem doesn't arise in theory. It arises when the entire catalog — hero products, niche items, and everything in between — is stuffed into the same campaign without any form of layering.
The algorithm does exactly what it is told: it finds the conversions that are easiest to get and pushes the budget there. In practice, this means that a small number of historical winners take most of the spend, while products with real — but as yet unproven — upside never get the exposure needed to prove themselves.
It is not a glitch in the algorithm. It is a consequence of giving it too little structure to work with.
At DVISIONMEDIA, we structure catalogs into three tiers before letting Performance Max run at full throttle:
Hero products
Products that have already proven themselves. Dedicated budget so they can scale without cannibalizing the rest of the account.
Potential products
Products with good margins, a strong feed, and real sales potential — but without the history required to win auctions against hero products in a shared campaign. They get their own budget so the algorithm can learn specifically from them.
Low-performing products ("Zombies")
Products with no upsell potential or strategic value. They are either excluded or run in isolation with a minimal budget so they don't steal exposure from the other two levels.
The point is not to split for the sake of splitting. The point is to give each level its own budget or its own ROAS/CPA target — the only two valid reasons for a separate campaign at the campaign level. If you cannot specify exactly which budget or which target a campaign isolates, it does not belong as its own campaign.
A tier structure is only as good as the feed it is built on. If product data is thin, incorrectly categorized, or inconsistent, Performance Max cannot match products to the right search intent — regardless of how well the campaigns are structured.
This means, at a minimum:
- Custom labels that reflect tier, margin, and hero status, so segmentation can be managed at the product level without building a new campaign for every change
- Titles and descriptions optimized for search relevance, not just to "look nice" in the feed
- Ongoing monitoring of Merchant Center for approval errors that quietly remove products from the auction
Tools like DataFeedWatch make it possible to manage this segmentation at scale — but the tool does not solve the problem alone. The structure and logic must be there first.
What about click costs?
Once the catalog is structured correctly, the next step is typically the auction costs themselves. Here, certified Comparison Shopping Service partnerships can help reduce cost-per-click in shopping auctions without compromising exposure — though that is a discipline of its own with its own mechanics.
Three signs that your catalog needs layering instead of one big PMax:
- A handful of products consistently take 80-90%+ of the budget, quarter after quarter
- Products with good margins and a good feed get virtually no exposure, no matter how much you adjust bids
- You cannot see which product is actually driving which conversion because everything runs in the same campaign without custom labels
None of these signs mean you should split everything into micro-campaigns. They mean you should ask: does this product level need its own budget or its own goal? If the answer is no, consolidation is still the right way to go.
We have managed over DKK 210 million in ad spend, and the pattern is consistent across accounts: brands that consolidate too much end up starving their best growth opportunities — and brands that split too much end up with data that is too thin for Smart Bidding to learn anything at all.
It is not an either-or situation. It is about knowing exactly when each level needs its own budget and when it should simply be allowed to be part of the whole. That is the approach behind our E-COM OS — where Paid Search works in tandem with feed, creative, and CRO in one cohesive system, rather than being optimized in isolation.
1. What is Performance Max, and how is it different from regular Google Shopping campaigns?
Performance Max is Google's AI-driven campaign type that combines Search, Shopping, Display, YouTube, and Discovery into a single campaign. Unlike traditional Shopping campaigns, you have less direct control over placements, which makes structure and feed quality even more important.
2. When should I split my Performance Max campaign?
Only when a product level requires its own budget or its own ROAS/CPA target. If you cannot articulate a concrete reason, consolidation remains the right choice.
3. How many conversions does a campaign need for Performance Max to work optimally?
Google recommends a minimum of 30 conversions over 30 days before enabling Maximize Conversions or launching a new PMax. With fewer than 15 conversions per month, Smart Bidding has too little data to function stably.
4. Is it dangerous if one campaign takes up most of the budget?
Not in itself. The algorithm allocates budget based on expected value. It only becomes a problem when products with real potential are starved because they never get the chance to prove themselves.
5. What is the "one SKU, one campaign" rule?
The principle that each product should only exist in one campaign. If the same product is in multiple campaigns simultaneously, they compete against each other internally, which drives up CPC without creating extra volume — and makes performance data unreliable.
.png)
