POAS-driven bidding in Google Ads

Instead of sending revenue as the conversion value, we send the contribution margin per order. Via Profitmetrics (or an equivalent setup with per-SKU margin data), the real contribution is calculated for each order and written back to Google as the value bidding optimizes toward.

Now Performance Max and Shopping understand the difference between the two products from before — and bid accordingly. It requires no new campaigns and no extra budget; it's the same engine, suddenly chasing the right target. The result is a quiet but sweeping reallocation: budget migrates toward the products that actually make money.

Technically it assumes margin data is correct and kept up to date per product, and that the value is sent with the right currency and timing (including handling returns where possible). If the data foundation is skewed, Google simply optimizes quickly toward the wrong thing — which is why this is a setup to build carefully, not stand up in an afternoon.

In Shopping and Performance Max there are no keywords you bid on. The feed IS your keywords. Titles, product type, attributes and images decide which searches you even match on — and a weak feed puts a ceiling on the campaign, no matter how good the bid is.

Titles are the heaviest lever. Front-load the most important search terms: brand, product type and the attributes people actually search for (color, size, material, compatibility). "Men's running shoes size 43 black — [brand] [model]" beats "[model]" every time. Product type and Google product category give the campaign structure and let you segment.

Custom labels are the secret behind POAS control in practice. By putting a margin bucket (e.g. high/mid/low) in custom_label_0, you can set different tROAS targets per bucket — bidding harder on the profit-drivers and keeping low-margin items on a short leash. Other labels can carry bestseller status, stock level or season. We often maintain the feed via DataFeedWatch or Producthero, so the rules run automatically.

Performance Max is effective, but opaque. Without structure you hand the whole catalog — and the whole budget — to one algorithm that's happy to throw money at brand searches and bestsellers you'd have gotten anyway. So we build control in from the outside.

Split asset groups or campaigns by margin and role via custom labels: the profit-drivers get room to scale with a higher tROAS ceiling, while entry products and low-margin items are steered separately with tighter targets. Exclude brand searches from PMax (via brand exclusions or a dedicated brand campaign), so you can see the real, incremental value add instead of PMax taking credit for people who searched for you directly.

Use search term insights and asset reports to understand what PMax is actually running on, and feed it negative signals where it wastes. Feed-only PMax (without image and text assets) can give cleaner, more Shopping-like behavior when you want control over the catalog rather than broad placements.