PMax asset groups: split by product category or keep fewer, bigger groups?

A lot of the Performance Max advice says to split asset groups by product category, like shoes, bags and accessories, and I'm not sure that holds for every store. Take a shop with 300 SKUs and very different margins. Splitting by product line gives you cleaner reporting and lets you set a separate target ROAS per group, and that's the main argument for it. But each group needs enough conversions to learn from, so a small accessories group with 15 sales a month will mostly just sit there. I'd start with fewer, bigger groups and only split off a line once it has real volume and a clearly different margin. Has anyone seen a split pay off at low volume, or did the learning cost outweigh the cleaner reporting?

In PMax, target ROAS is set at campaign level, not per asset group. So if you want a separate target for each product line, you need separate campaigns, and each one needs its own conversion volume to learn. That makes your 15-sales accessories line a problem either way.

Whether the split is worth it comes down to how big the margin gap is. If it's wide enough to justify its own ROAS target, split into campaigns and live with the slower learning. If you just want cleaner reporting, asset groups give you that without dividing budget and learning across campaigns.

Agreed on the campaign-level target. One thing that gets missed is that all asset groups inside a campaign share that single tROAS, so you don't need separate campaigns just to report by margin. Label products by margin tier in your Merchant Center feed (custom_label_0 with values like high and low) and build your listing groups on those labels. You get clean reporting by tier, and everything still learns from the same pool of conversions.

The catch is that a shared target can starve the high-margin line. If accessories need 4x ROAS and shoes are fine at 2x, one target will push budget toward whichever line converts more easily, and that's usually not the one you want. So the margin gap matters, just for a different reason than volume. A wide gap is what justifies splitting into campaigns, while a narrow gap means a label is enough.

Agreed on the shared target, and I'd boil it down to a simple test. If the ROAS you'd want for a line is off from the blended target by 50% or more, and that line can get to around 30 conversions a month on its own, give it its own campaign. If either one fails, use a label. Small margin differences just get averaged out by the shared target, and every split resets some learning, so it only makes sense to pay that cost when the gap is big enough to change where the money goes.

I'd go with the test, but change one input: don't pick the target ROAS for each line by feel, work it out from margin. Break-even ROAS is 1 divided by gross margin. A line at 40% margin breaks even around 2.5x, while a 20% line needs 5x just to cover product cost. Compare those break-even numbers to each other rather than to the blended target, because the gap between them tells you whether the shared target will push budget the wrong way.

The 30-conversion check also depends on the window you use. Take the last 60 days of actual data for that line, not a forecast. A seasonal line can clear 30 in Q4 and fall well short the rest of the year, so a split that looks justified in November may just sit there in March.

1 divided by gross margin is the right formula, but for apparel and shoes it can flatter the line. If 25% of orders come back, the revenue you actually keep is lower, so a shoe line showing 45% gross margin might really be closer to 35% once returns are netted out. That moves break-even from about 2.2x to about 2.9x, which is a much smaller cushion against a 3x blended target. Shipping and payment fees push it higher too, though usually by less. Run the margin math on the number after returns, because that's the money you actually keep.

Break-even is also just a floor. Your target has to sit far enough above it to make scaling the line worth it, so the comparison that matters is target minus break-even for each line, not break-even alone.