Sumendit · PMax

Sumendit was running a single PMax campaign with all products dumped into one asset group. PMax's algorithm couldn't distinguish hero products from long-tail SKUs, and ROAS was stuck at 4×.

They wanted to push cost per purchase down without losing volume — a classic PMax over-consolidation problem.

• Split the catalog into 3 asset groups by margin tier — high, mid, long-tail.
• Built dedicated creative sets per asset group: 5 headlines, 5 descriptions, 4 image sizes, 1 video each.
• Used Search themes to feed PMax keyword signals it couldn't pick up from the feed alone.
• Set tROAS goals tier-by-tier — 800% on hero SKUs, 400% on long-tail.
• Excluded brand search to keep PMax honest on incremental conversions.

Inside 30 days, ROAS climbed from 4× to 14.4×. Cost per purchase dropped 61% while purchase volume held steady — the asset-group split unlocked spend efficiency PMax couldn't find on its own.

The structure now serves as the template for every PMax account I run.