Performance Max: when it makes sense and how to keep it under control

The situation

A store with 4,000 products and a clean Merchant Center feed launches Performance Max and within two months sells at ROAS 6 without touching anything. A clinic with 12 leads a month launches the same campaign and receives 80 forms — 70 are spam, service pitches and people from other countries. Same campaign, opposite results. The difference is not luck. It is what you gave the system to learn from.

What Performance Max is

Performance Max (PMax) is a campaign type where Google decides where you appear — Search, Shopping, YouTube, Display, Discover, Gmail, Maps — and to whom, starting from your conversion goal and the assets you upload. You do not pick keywords, you do not pick placements, you do not set bids per segment. You pick the conversion, the budget, the target (CPA or ROAS) and the assets. The rest is the algorithm.

It is powerful precisely because it decides at every auction, on signals you cannot see. It is dangerous for the same reason: if the target is wrong, it optimises perfectly towards the wrong thing.

When it makes sense

• E-commerce with a feed — the best case. The Merchant Center feed gives the system products, prices, images, availability. The conversion (Purchase with value) is clear. Here PMax has largely replaced classic Shopping.
• Lead gen with solid data — it works if you already have dozens of conversions a month, measured correctly, and especially if you import qualified leads back (offline conversion import or enhanced conversions for leads). Without that, "conversion" means any form, spam included.
• When Search is saturated — you have covered the relevant keywords and want extra volume from other channels without building five separate campaigns.

When it does not

• A new account with no conversion history. The system has nothing to learn from and will learn from noise.
• Uncertain tracking. If you have not been through the conversions lesson, do not launch PMax.
• A budget too small for several channels. With a few tens of lei a day, a tight Search campaign is better.
• Services where every lead is expensive, quality matters more than volume, and you cannot close the loop with quality data.

How to build it correctly

1. One conversion goal, primary, clean. Purchase with value for e-commerce; qualified lead for services. Not "all conversions in the account".
2. Asset groups by theme, not one for everything. One asset group = one product category or one service, with headlines, descriptions, images and video that talk about that thing. Without an uploaded video, Google generates one — you usually do not want that.
3. Audience signals — customer lists, visitors, custom segments. They are a starting point, not a restriction: the system starts here and expands.
4. Search themes — the search topics you want covered. Still a signal, not a keyword, but it helps at launch.
5. Brand exclusions — the brand list you exclude, so PMax does not buy clicks on your own name that you would have received for free or from the brand campaign.
6. Exclusions — negative keywords (at account level and, where available, at campaign level), URL exclusions and turning off final URL expansion if you want traffic to land only on the pages you chose.
7. The target — start without a target CPA/ROAS or with a realistic one from history. An aggressive target on day one strangles delivery.

How to read the reporting honestly

PMax reports in aggregate. The easiest thing in the world is to see a high ROAS and not ask where it comes from.

• How many conversions are on brand? If you did not exclude the brand, part of the "performance" is traffic you had anyway. Compare with the brand campaign and with organic in Search Console.
• Search terms insights — which search categories brought conversions. This is where you see drift towards irrelevant themes.
• Channel and placement reporting — where impressions went. Cheap Display and video inflate impressions; what matters is conversions per channel.
• Asset groups — which one brings the conversions and which one just consumes.
• Incrementality — the hardest question: if you pause PMax, how many of these conversions disappear? A simple test: pause 2–3 weeks on one category and watch the account and site totals, not just the campaign.