Product-level tROAS analysis
How BidSmart works backwards from each product's history, price and margin to the target it can individually bear, and finds where one campaign-wide average is starving winners or overfeeding losers.
A single campaign-wide target ROAS treats every product the same, but a £9 accessory and a £600 hero product have nothing in common. BidSmart analyses each product on its own history, price and margin to work out the target it could actually sustain, then shows you where the campaign average is leaving money on the table or burning it. Every adjustment it suggests is queued for your approval; nothing changes on its own.
Why the campaign average lies
Target ROAS tells Google to hit an average return across everything in the campaign. That's fine when your products are alike. It's actively costly when they're not: and they're never all alike.
Under one 400% target, a high-margin £9 accessory that could profitably run at 250% is being told to bid as if it needed 400%: so it wins less traffic than it should. Meanwhile a low-margin £600 product that can't clear 400% keeps drawing spend on losing auctions. The average looks fine. The two products underneath it are both being mispriced, in opposite directions.
Working backwards to the target a product can bear
Instead of asking "what did this product return against the campaign target," product-level analysis asks "what target could this product sustain on its own." It works backwards from three things:
- Conversion history: the product's own spend, conversions and conversion value, which set the ROAS it actually achieves.
- Price point: a £9 product and a £600 product tolerate completely different targets for the same absolute profit.
- Margin signals: from your feed, so a product's target reflects the profit it carries, not just the revenue it books.
From those, BidSmart derives the target each product could individually carry: then compares it to the target the campaign is currently imposing.
The two things it finds
Deep dive Money left on the table vs money being burned
Every product falls into one of two gaps once you compare its own sustainable target to the campaign target it's actually running under.
Money left on the table: products throttled too hard. These clear the campaign target comfortably; their own economics could support a lower target, which means bidding more aggressively and winning more of the traffic they convert well on. The campaign average is holding them back. Lowering their target (in small, approved steps) lets them grow. These are the quiet wins: products doing well that could do much more if Google weren't told to bid so cautiously on them.
Money being burned: products that never clear the bar. These consistently return below the campaign target and drain budget that the throttled winners could use. Raising their target pulls spend back from auctions they lose money on. Occasionally the right answer isn't a target tweak at all but the incubator: if the product is thin on data rather than genuinely unprofitable.
The hard constraint underneath all of this is data thinness. A product with four conversions has no statistically usable ROAS: reading a "sustainable target" off it would be noise dressed as signal. BidSmart handles that two ways: it borrows a starting estimate from the product's category and price-band siblings until it has its own signal, and it routes genuinely new or thin products to the incubator, where the job is buying data cheaply rather than hitting a target. A product only earns a confident product-level target once it has the conversions to justify one: and until then, BidSmart says so rather than guessing loudly. See Target ROAS explained for the statistics behind why thin data breaks tROAS.
From analysis to action: always via approval
Product-level analysis produces recommendations, not changes. Each proposed target adjustment appears as a bid recommendation with the product's numbers attached, and waits in your approval queue. You approve the ones you agree with; they execute in small steps, record their before/after state, and can be reversed. Nothing about your bids changes until you sign it off.
Structure still matters
Product-level targeting isn't a substitute for sensible campaign structure: it's a layer on top of it. Group products with genuinely matching economics into their own campaigns first (margin bands, price bands, lifecycle stage), carried from your feed via custom labels. Good structure narrows the averaging problem; product-level analysis mops up what structure alone can't reach.
Frequently asked questions
Isn't segmenting my campaign by margin enough?
It helps a lot, grouping products with matching economics is the right first move. But even inside a well-segmented campaign, individual products diverge. Product-level analysis catches the winners and losers that a group average still hides.
What if a product has almost no conversions?
Then it has no reliable ROAS of its own. BidSmart leans on comparable products (same category and price band) as a starting point and usually routes genuinely thin-data products to the incubator to build their own signal cheaply.
Do these adjustments happen automatically?
No. Every target adjustment is a recommendation in your approval queue. It executes only after you sign it off, and it's reversible afterwards.