Performance Max: When It Works and When It Burns Budget

Performance Max asks you to hand over placement, audience, creative combination and bid in exchange for reach across every Google surface at once. Framed that way, the question is simple: is what you give up worth less than what you get?

Sometimes clearly yes. Sometimes clearly no. The failure is not choosing it or refusing it — the failure is running it without knowing which case you are in.

What it actually decides for you

A Performance Max campaign chooses where the ad appears, who sees it, which combination of assets is shown, and what to bid. You supply the assets, the conversion goal, the budget and the exclusions. That is the whole of your control surface.

Everything that follows from that is predictable. If your conversion goal is wrong, the campaign will optimise energetically toward the wrong thing. If your assets are thin, it will show the thin ones more often because they are all it has. The system is not going to correct your inputs.

When the trade pays

• A product feed with real breadth. Retail with hundreds of items is where the automation has something to solve that a human cannot: which product to show which person. With eight products it has almost nothing to decide.
• Clean conversion data with enough volume. Thirty or more conversions a month per campaign, measuring something that matters. Below that the model is learning from noise.
• A genuine creative library. Fifteen images, several videos, and copy with actual variance. Assets are the only lever you keep, and an account that supplies four images has given up the last one.
• Demand you are not already capturing. If Search is taking most of the available intent, Performance Max will mostly re-buy it and report it as new.

When it burns budget

The pattern is consistent. A lead-generation account with a form-fill conversion, no exclusion list, no feed, and six creative assets. It reports a low cost per conversion within a fortnight and the sales team notices nothing changed. The campaign found the cheapest form fills available, which is exactly what it was told to do.

It also struggles where the purchase is decided offline, where the sales cycle is long enough that the feedback arrives after the model has moved on, and where brand and generic demand have not been separated — a problem we take apart in attribution across channels.

The exclusions that change the outcome

• Brand terms, unless you are deliberately buying them. Otherwise most of the reported return is traffic you already owned.
• Existing customers, uploaded as a list and excluded, on anything where repeat purchase is not the objective.
• Placement categories that have never produced anything, reviewed monthly rather than once at launch.

Reading it honestly

Judge the account, not the campaign. Performance Max frequently reports a superb cost per conversion while total account conversions stay flat, which means it moved credit rather than created volume. The only reliable read is the before-and-after at account level over a full cycle, which requires measurement built before the campaign launched rather than after.

Set the optimisation target to a later event than the form fill — a qualified enquiry, a booked call, a sale — and most of the argument about Performance Max resolves itself. The campaign was never the problem. The target was.

If you are running it now and cannot tell what it is buying, that is the first thing a Google Ads engagement establishes. It is usually a fortnight of measurement work, and it decides whether everything after it can be read.