Performance Max: one ROAS target or one per margin tier?

Picture a store with 400 SKUs and margins anywhere from about 12% to 60%. One ROAS target across the whole campaign starves the high-margin products, while the low-margin bestsellers eat budget at break-even. Splitting asset groups by margin tier looks like the fix, but each group then gets thin signals and may never leave learning. And a 300% ROAS on a 15% margin item still loses money once shipping and returns hit, so the target really has to be profit-based. Has anyone run tROAS per margin tier in PMax for a while? Did the groups stabilize, or did they just compete with each other for the same auctions?

Asset groups don't really separate products. Which SKUs get traffic comes down to the feed and product groups, and asset groups in the same campaign still bid into the same auctions. That's probably why yours competed. If you want different targets per margin tier, the split has to happen at campaign level, with products separated by custom labels in the feed.

My rule of thumb is that a tier only gets its own campaign and tROAS if it can produce around 30 conversions a month at that target. Below that, it learns slowly and the target swings on a handful of orders. For the thin tiers, merge them and send margin-adjusted conversion value instead, so one tROAS target already means profit. The bidder then optimizes for what you actually keep, and the high-margin products aren't stuck with a target that only suits the low-margin ones.

Margin-adjusted value also makes one number mean the same thing across tiers. If you send revenue times gross margin as the conversion value, a tROAS near 100% is break-even on every SKU, so most of your per-tier targets just go away. Which margin figure are you feeding in, gross or after returns?

Agreed on the 100% break-even point, and I'd send gross margin at purchase rather than margin net of returns. Returns land weeks later, so if you subtract them up front you're guessing, and if you send nothing for refunds the bidder keeps learning from orders that were later reversed. Send gross margin at purchase, then push a conversion adjustment when a refund posts so the bidder sees the correction.

What trips people up is that conversion value is per order, not per SKU. A basket with one 60% item and one 12% item gets a single number, so mixed baskets blur the tiers. Check what share of your orders are multi-item before you trust the per-tier split. In a store where most people buy two or three things, the tiers drift toward the middle, and the margin-adjusted target ends up looking cleaner in the dashboard than it is in the auction.

I'd push back on using gross margin at purchase. On a 12 to 15% item, payment fees and shipping can eat most of that gap, so even a tROAS near 100% still loses money on exactly the SKUs you're worried about. Feeding contribution margin after fulfilment costs is messier, but it matches the profit problem you started with.

Depends on whether your fulfilment costs are per parcel or per unit. A flat shipping charge hits single-item orders much harder than it hits bigger baskets, so a per-SKU contribution number misstates the order it's attached to.