Ecommerce
Performance Max, with the reporting put back in.
PMax is not a black box so much as a box nobody opened. Search terms, placements and product-level data are all reachable – and without them you are trusting a system optimising towards whatever conversion value you set months ago.
The three things that decide whether PMax works
What conversion value you feed it
PMax optimises towards the value you report. Report revenue and it will find revenue, including from products you lose money on. Report a value net of margin and returns and the same campaign starts buying profitable orders instead. Nothing else on this page matters as much as this.
Whether brand traffic is fenced off
Unfenced, PMax absorbs searches for your own brand – the cheapest, highest converting traffic you have – and reports the result as incremental performance. A dedicated exact-match brand Search campaign takes those queries back, and the PMax numbers drop to something honest.
How asset groups are divided
Split by something that changes what you would do: margin band, price tier, or genuinely different buyer intent. Splitting along your catalogue tree is the common default and mostly just fragments conversion signal across groups that behave identically.
Getting the data out
The reporting exists; it is just not in the default view. Search term insights, placement reports, asset-group breakdowns and product-level performance can all be pulled through scripts and the API into a Looker Studio view you can actually read. We set that up in the first two weeks, because until it exists every optimisation conversation is speculation.
What we do not do
We do not run PMax as the whole account. It sits alongside brand Search, high-intent non-brand Search, and usually a standard Shopping campaign for the products that deserve individual attention. PMax is a channel, not a strategy, and accounts that hand it everything lose the ability to explain their own results.
Questions people actually ask
- Does Performance Max cannibalise brand traffic?
- Left unstructured, yes. PMax will happily serve on searches for your own name, convert them at a high rate and report it as incremental revenue. The fix is a dedicated brand Search campaign with exact-match coverage, which outranks PMax for those queries, plus brand exclusions on the PMax campaign itself.
- How should asset groups be split?
- By something that changes the decision – margin band, price tier, or genuinely distinct buyer intent. Splitting by catalogue category is the default and it is usually wrong, because two categories with the same margin behave identically and only fragment your conversion signal.
- How much reporting can you actually get?
- More than the interface suggests. Search term visibility, placement data, asset-group and product-level performance are all reachable through scripts and the API. An agency telling you PMax is a black box is telling you which tools they have not built.
- Should we use audience signals?
- Yes, as a starting hint rather than a targeting rule. Good customer-match lists and site-behaviour segments meaningfully shorten the learning phase. They do not constrain who the campaign reaches, so treat them as a nudge, not a fence.
- How long before it settles?
- Two to three weeks to exit learning, six weeks before the numbers are stable enough to judge. Significant edits reset that, so we batch changes rather than trickling them in.
- When is PMax the wrong answer?
- Very small catalogues, thin conversion volume, and businesses where a single wrong lead is expensive. Below roughly 30 conversions a month the campaign never leaves learning in any meaningful sense, and standard Search gives you more control for the same money.
Free
Get the Performance Max audit
Related.
The wider ecommerce approach, including feed strategy and how we work back from contribution margin.