Performance Max is Google's most automated campaign type: the system, not you, decides audience, placement and bid. The price of that freedom is simple — the system is only as good as the signal you give it. When we took over Goldia, the US, UK, Canada and Australia were running in the same account, from the same budget pool, with largely the same assets. On paper it looked sensible: one setup, one maintenance routine.
A machine that optimises the average
The problem surfaced within weeks. Four currencies, four seasonal curves and four sets of shipping terms were feeding a single learning pool. The system read all of it as one "average conversion value". The result: the US pulled almost the entire budget and Australia got barely any impressions. Blended ROAS in the report looked fine; per market, two countries weren't really running at all.
We separated measurement first
The first step wasn't the campaign, it was measurement. We rebuilt GA4 e-commerce events and Google Ads conversions via GTM with currency normalisation. Without that, "is Australia profitable?" had no answer. Then we split the country feeds in Merchant Center: local price, local shipping rule, local currency. Most disapproved products came from country-level feed errors.
Then the architecture
A separate campaign set, budget and tROAS target per country. Asset groups were localised; the same product spoke differently in the UK and in Australia. We pulled brand searches out of PMax into a separate search campaign so PMax's "success" wouldn't be inflated by brand traffic.
And finally the season
Summer in Australia is in December; Black Friday exists in all four countries but Mother's Day dates differ. One season plan didn't fit four countries. We built a separate budget plan for each country's own calendar.
In short: if you run PMax across markets, feed the system four separate markets instead of one "average market". Don't reduce automation; ask it the right question.