Shopping product groups: build a tree by brand and category, or keep one catch-all?

Most Standard Shopping threads split into two camps when people run it next to PMax on the same catalog. One camp builds a tree by brand, category and product type and sets bids per node. That gives you control, but the tree turns into a maintenance job every time the feed changes. The other camp keeps one catch-all product group per campaign and uses priority (high, medium, low) to decide which campaign wins the auction for the best sellers.

I think the second approach works better for catalogs under a few hundred SKUs, because a node with twelve products usually gets a bid that's a guess anyway. The exception is when margins vary a lot. One bid for everything hides the products you lose money on, and in that case the tree earns its upkeep.

Brand and category mostly reflect how the catalog is organized, not how the money behaves, so a tree built that way tends to end up with a dozen nodes that all get the same bid. I'd build it around margin tiers instead. Split a node only when you can finish this sentence: "These should bid differently because ___." If the blank is margin, stock position or price against the competitor, the split earns its upkeep. If you can't fill it in, the product belongs in the catch-all.

Agreed on the test. One thing I'd add is to check whether the split actually changes what the bid does. If two nodes have different bid targets but identical ROAS in the search terms data for the last 30 days, they're the same node in practice, so merge them before you start paying the maintenance cost on every feed change.

I'd put a number on that check. A ROAS gap between two nodes can be pure noise, so it only deserves trust when each node has around 30 conversions in the window. Below that you're comparing luck. A decent rule of thumb is to merge two nodes unless their ROAS gap is clearly bigger than the week-to-week swing on a single node. Most small catalogs won't clear that bar, which is why the catch-all wins more often than the tree people like to build.

I'd push back on the 30-conversion floor. Margin comes from your cost data before any sale happens, so a node with eight orders can still earn its own target if its margin is 15% while the other's is 60%. Demand 30 conversions per node and small catalogs never get that split.

Priya, you're right that margin exists before any sale, so a 30-conversion floor is the wrong test for that split. But it depends on how the node gets bid. If you set a manual CPC from the margin math yourself, eight orders is enough to act on, because you're applying a cost you already know. If the node runs on target ROAS, the algorithm has to learn from that node's own conversions to hit the target, and eight orders won't teach it much. So the margin split is cheap under manual bids and expensive under Smart Bidding.