Your Pmax ROAS looks strong. The dashboard says 6x, maybe 7x. But new customer numbers have been flat for two quarters, and when you try to trace where those conversions are actually coming from, the answer gets murky fast. You're spending more than you were twelve months ago, and the business isn't growing to match. That's not a reporting anomaly. It's usually the first signal that Performance Max is doing exactly what it's designed to do, just not what you actually needed it to do.
This is the conversation we have regularly. Not "does Pmax work?" but "why does it look like it's working when the business isn't growing?"
What Performance Max actually does
Performance Max is Google's fully automated campaign type. You provide the inputs: a product feed via Google Merchant Center, creative assets (headlines, descriptions, images, video), and audience signals. Google's system handles the rest: where your ads appear, how much to bid, which creative combinations to show, and to whom.
That covers a lot of ground. Pmax runs across Search, Shopping, YouTube, Display, Gmail, Discover, and Maps. No keyword lists, no manual placements, no manual bids. The algorithm assembles the campaign from what you feed it and optimises toward whatever conversion signal you've set.
The structural unit inside a Pmax campaign is the asset group: roughly the equivalent of an ad group in a Search campaign. Each asset group contains the creative assets and audience signals that Google uses as a starting point. "Starting point" matters here. Unlike traditional audience targeting, the signals you provide in Pmax are suggestions, not hard limits. Google will expand beyond them if it believes it can find conversions elsewhere.
When this runs properly, with a clean feed, strong creative, and enough conversion volume for the algorithm to learn, Pmax can deliver solid commercial results. The problem is that it can also look like it's delivering results when the underlying picture is quite different.
The slow-burn problem: when Pmax starts cannibalising what's already working
The most common issue we see with Pmax in ecommerce accounts isn't a sudden performance collapse. Instead, it's a gradual distortion that's difficult to catch because the headline numbers stay healthy.
Here's the pattern. Pmax optimises toward whatever converts most easily. In many ecommerce accounts, the easiest conversions are branded search (people already looking for you by name) and retargeting (people who've already visited the site). Both groups convert at a much higher rate than cold audiences. The algorithm learns this, and without guardrails in place, it shifts budget toward them.
The result: Pmax ROAS improves. But you're increasingly paying to acquire customers you would have won anyway through organic search or direct traffic. New customer acquisition stalls. The channel looks efficient on paper. The business isn't actually growing.
We covered an almost identical dynamic in our piece on last-click attribution. The mechanism is the same: a system takes credit for demand it didn't generate. With attribution, it's the last channel in the path claiming the conversion. With Pmax, it's the campaign absorbing budget toward conversions it didn't create, then reporting them as wins.
Neither fails obviously. Both require you to look past the headline number to understand what's actually happening.
Why your Pmax ROAS doesn't match your actual margin
Say your Pmax campaign is reporting 7x ROAS. You're spending £15,000 a month, and the platform is attributing £105,000 in conversion value. On paper, that looks strong.
Pull it apart: if 40% of those attributed conversions are branded search and another 20% are returning customers reached through retargeting, the actual incremental revenue Pmax generated is closer to £63,000. The real return on genuinely new demand is around 4.2x. That might still be acceptable for your business. But it's a very different number to the 7x you're reporting against, and decisions made on the 7x are likely to be wrong.
This gap gets wider when the product feed isn't in good shape. Feed quality is one of the most direct performance levers in Pmax, and it's often the last thing teams audit. If titles, descriptions, product type fields, and pricing signals aren't clean, the algorithm has less to work with when matching ads to search queries. It compensates by broadening its reach, frequently in directions that don't reflect the margins you care about.
Campaign-level negative keywords are another common gap. They became available to all Google Ads advertisers in 2025, with the limit later raised to 10,000 per campaign, now matching the Search campaign limit. More often than not, the accounts we review are running well below that threshold, which means Pmax is serving against queries nobody would have consciously chosen to target.
Layer in broad audience signals, blended reporting across all placements, and limited visibility into where budget is actually going, and you end up with a campaign that's genuinely difficult to interrogate. The trade-off for Pmax's reach is that it's a relatively opaque system. You know it's running. Working out precisely what it's running against takes deliberate effort.
Demand Gen vs Performance Max: what each is actually for
Demand Gen and Pmax increasingly overlap, and running both without a clear structure between them is one of the more reliable ways to drive up costs without growing the business.
Demand Gen runs on YouTube, Discover, Gmail, and Display (it expanded to include Display in 2025). It's built for upper and mid-funnel activity: reaching people who don't know your brand yet, or warming audiences that are in-market but haven't committed. The key difference is targeting control. Demand Gen lets you specify who you're reaching rather than delegating that decision to the algorithm.
Pmax runs across every Google surface, including Search and Shopping. It's optimised for conversion volume at the bottom of the funnel, where purchase intent is already present. Less targeting control, broader reach, more automation.
Neither is universally the right choice. The question is which one fits your objective. If you're trying to build demand with a cold audience, Demand Gen is the right tool. If you're capturing existing demand at scale, Pmax can do that well. The issue starts when both campaigns run simultaneously, targeting overlapping audiences, with no exclusions or budget logic between them. At that point, you're not running a considered funnel strategy. You're running two campaigns competing for the same pool of users and paying for it twice.
When Performance Max creative requirements make or break it
Pmax is only as capable as the inputs you give it. On the creative side, that means headlines, long headlines, descriptions, short descriptions, landscape images at 1.91:1, square images at 1:1, logos, and optionally video. More assets, in more combinations, means more placements the algorithm can access. If you under-supply creative, you limit where the campaign can appear.
Where brands typically go wrong is either providing too few assets or loading everything into one broad asset group, which gives the algorithm very little to work with in terms of signal.
Asset groups should be themed around your product range. If you sell across three categories, each should have its own asset group with creative that speaks specifically to it. A single catch-all group tells Google almost nothing about what you're selling or who you're trying to reach. The algorithm does its best with what's available, but "its best" in that situation usually means broadly matched, poorly targeted spend.
There's also a specific tactic worth knowing for ecommerce accounts: feed-only asset groups. These strip out all creative and limit the campaign to Shopping and Dynamic Remarketing placements only. It's a way of directing budget toward the highest-converting surfaces without the algorithm distributing it across display or YouTube inventory where intent is lower. It won't suit every account, but in accounts where Shopping significantly outperforms other placements, it prevents a lot of misallocated spend.
When it's time to bring in Performance Max management
There's a version of Pmax that in-house teams can manage effectively, particularly when the account is clean, the feed is well-maintained, and there's enough conversion volume for the algorithm to work from. The commonly cited threshold is 30 to 50 conversions per month, though the real number varies by account and bid strategy. Below that, the system is learning from too little data and making poor optimisation decisions as a result.
The point at which external Performance Max management starts to pay for itself is usually when one or more of the following applies.
You're spending at a scale where inefficiency shows up in real money. At £10,000 a month, a 20% waste rate costs £2,000. At £50,000 a month, it's £10,000. At that level, the cost of fixing the account is typically a fraction of what poor setup is costing each month.
Campaign structure has grown without clear logic. Multiple overlapping campaigns, asset groups with no thematic consistency, no documented rationale for how budget is allocated.
You've lost visibility into what's actually converting. If you can't answer which placements are delivering which results, or whether Pmax is absorbing budget from your brand search, you're optimising against incomplete data.
At Gravytrain, we typically start by separating branded and non-branded performance before touching anything else, because until you understand what Pmax is genuinely responsible for, any optimisation decision is being made against the wrong baseline.
See how we approach Paid Search
Getting it right
Performance Max is a capable system. It can deliver strong results with the right inputs and the right oversight in place. The difference between "looking efficient" and "actually driving growth" almost always comes down to how much active management is sitting behind the automation.
If your Pmax account is overdue for a proper review, or you'd like a second opinion on whether your current setup is working as well as it should be, get in touch.
