Performance Max takes a third or more of Google spend in most retail accounts we see, and in plenty of them it earns its keep. The complaint we hear isn’t about results. It’s that nobody can see inside. Where did the money go? Which queries? Which placements? Google’s answer is mostly a shrug.

Fair complaint. But “it’s a black box” has become a licence to stop managing the campaign at all, and that licence is expensive. The levers still exist. They’ve just moved.

The feed is the steering wheel

In the retail accounts we run, most PMax conversions still come through shopping placements. Which means the feed is doing more targeting work than anything in the campaign settings. Titles, GTINs, prices and categories pass straight through into what PMax shows and to whom. If you only pull one lever, pull this one. We wrote up what feed work is worth separately; all of it applies double inside PMax.

Structure by margin, not by convenience

One asset group with the whole catalogue in it is the default setup, and it quietly hands your pricing strategy to the algorithm. PMax optimises toward whatever converts most easily, which is usually your discounted or low-margin range.

Split campaigns or listing groups by margin tier and give each its own ROAS target. An electronics retailer we work with separated accessories (around 60% margin) from devices (around 12%) and set targets accordingly. Blended ROAS barely moved. Profit from the same spend rose 18% in two months.

Take your brand back

Left alone, PMax will buy your brand queries and report them as performance. Those clicks would have cost a tenth as much in a dedicated brand campaign, and many would have arrived free through organic. Apply brand exclusion lists to every PMax campaign, run brand search separately, and check the split monthly. On accounts where brand traffic was never carved out, we routinely find that 20-30% of reported PMax conversions were branded.

Nudge with themes, cut with negatives

Search themes let you point the campaign at query spaces it’s missing. Campaign-level negative keywords, finally available to everyone, let you cut the junk: competitor names you never win, support queries, the products you discontinued last season. Neither gives you keyword control back, but together they meaningfully shape where spend lands. Review the search insights panel weekly and feed what you learn back in.

Make it report to you

Publicly available scripts can break PMax spend down by channel: Shopping, Search, Display, YouTube. Install one; setup takes under an hour and the visibility is permanent. The pattern to watch is Display and YouTube share creeping upward while conversions stay flat. That’s the campaign buying cheap reach to satisfy its target, and it’s your cue to tighten the ROAS goal or prune weak assets. On one apparel account, video share climbed from 4% to 19% of spend over six quiet weeks before the script made it visible. Nobody had changed a thing.

The weekly five

Every retail PMax account we manage gets this check once a week:

  • Brand vs non-brand share of conversions
  • Channel split, and whether it drifted
  • Spend concentration across asset groups
  • New queries in search insights worth acting on
  • New vs returning customer ratio

Twenty minutes, most weeks. The point is catching drift while it’s still cheap to correct.

Treat PMax like a talented junior media buyer: give it clean inputs, set hard constraints, audit the output weekly. It does the volume work better than any human desk ever did. It just shouldn’t be left to mark its own homework.