Optimising your Google Shopping feed purely for ROAS is one of the most expensive mistakes a performance team can make — it trains Google's algorithm to maximise revenue on the SKUs that eat your margin fastest. After auditing 50+ Shopify stores in 2025, the single biggest contributor to margin erosion we found wasn't ad spend, creative quality, or bid levels — it was feeds that had no margin signal at all, leaving Google to route budget toward whatever converted at the highest revenue rate regardless of what that revenue actually cost the business.
The ROAS Trap: How Revenue Optimisation Destroys Margin Mix
ROAS as a primary Shopping KPI made sense in the pre-PMax era when you could force exact product coverage via Standard Shopping campaigns. Performance Max changed the contract: you hand Google a feed and a signal, and the algorithm decides what to show, to whom, and when. If the only signal you provide is target ROAS, Google optimises for revenue-per-impression — full stop. There is no built-in mechanism that distinguishes a $300 sale at 15% margin from an $80 sale at 42% margin.
The mechanical problem is that high-revenue SKUs are rarely your best-margin SKUs. A $300 item converting at 4× ROAS generates $300 of revenue per $75 spent. A $80 item at 6× ROAS generates $80 of revenue per $13 spent. Google surfaces the $300 product because it wins on revenue. But if the $300 product carries 15% contribution margin and the $80 item carries 42% margin, the $300 product generated $45 gross contribution versus $33.60 for the $80 item — a meaningful difference per order, and catastrophic at scale once you account for customer acquisition cost, returns, and fulfilment.
We rebuilt feeds for 11 DTC brands this quarter and the pattern emerged without exception: the bottom 20% of SKUs by contribution margin were receiving between 31% and 47% of Shopping impression share. One $4M/yr Shopify apparel brand was spending 38% of its Google Shopping budget on its three worst-margin product lines.
The same margin compression pattern — acceptable ROAS masking an 18–22% gross margin decline — was the central finding in our PMax asset group audit across three eight-figure DTC brands.
Per McKinsey's growth-marketing and sales research, companies that align media investment to margin contribution rather than top-line revenue routinely unlock 15–25% incremental EBITDA without increasing total spend.
The fix is not a bid adjustment. It is a data architecture decision you make at the feed level before a single auction is touched. That distinction matters because bid adjustments are reactive — they chase outcomes after Google has already made routing decisions. Feed-level margin signals shape those decisions upstream.
Performance Max's auction weighting means that attribute priority — not just attribute presence — determines whether your margin-tier signal actually reaches Google's ML layer; our PMax attribute priority guide ranks the 2026 attributes by influence on machine-learning signal strength.
That budget-routing failure is structural: without margin-tier segmentation at the asset-group level, PMax treats a 12%-margin SKU and a 42%-margin SKU as identical inventory — a problem our PMax feed segmentation guide addresses with a tiered asset-group architecture specifically tested on Shopify catalogues.
Mapping Contribution Margin to Feed Attributes Without an ERP
The most common objection to margin-based feed segmentation is "we don't have an ERP that pipes margin data cleanly into our feed." That's fair — and it's not a blocker. Enterprise ERP integration is ideal, but an 80%-accurate margin signal built from Shopify's native cost field is enough to separate your 40%+ margin products from your 12% margin products, which is the only segmentation that changes budget routing materially.
Start by exporting three columns from Shopify: variant_id, price, and cost. If cost is missing for some SKUs, use your category-level average COGS as a proxy — imperfect, but directionally correct enough for tier assignment. Contribution margin at the feed level doesn't need to match your finance team's fully-loaded accounting figure; it needs to be consistent and accurate enough to rank products against each other.
Before building a margin-tier architecture on top of a feed with unresolved data quality issues, run the 23-point feed audit checklist — missing GTINs and incorrect product types in the base feed will corrupt tier assignment and undermine the whole segmentation model.
Shopify metafields provide the cleanest injection point. Per Shopify's official metafields documentation, you can create a product.metafields.custom.contribution_margin_tier field and populate it via bulk editor, a custom app, or a scheduled data export from your analytics stack. Once the field exists, your feed management tool (DataFeedWatch, Feedonomics, GoDataFeed, or a custom Google Sheets–to–Merchant Center pipeline) can map it directly to a custom_label attribute.
Custom Label Architecture for Margin-Tier Segmentation
Once contribution margin data is accessible in your feed pipeline, the goal is to translate raw margin percentages into discrete, actionable tiers that Google's campaign structure can act on. We recommend a three-tier model as a starting point:
If you're running the same SKU catalogue on Meta Advantage+ Shopping alongside Google, margin-tier custom labels translate directly — Meta's catalog segmentation rules that produced a 20–25% CPA reduction use the same high-margin / low-margin split logic.
If you're building on top of this three-tier model, the custom label scaling framework shows how three DTC brands extended margin-tier labels into bid multiplier logic that sustained 8-figure ROAS without flattening contribution margin.
- Tier A (High Margin): 35%+ contribution margin — These are your growth SKUs. Allocate aggressive impression share, lower your tROAS target relative to your account average, and let Google find volume.
- Tier B (Mid Margin): 18–34% contribution margin — Maintain current spend levels. These SKUs are healthy but not the primary growth lever. Use breakeven tROAS targets.
- Tier C (Low Margin): Below 18% contribution margin — Suppress or constrain. Either exclude these SKUs from PMax asset groups entirely, or create a separate campaign with a significantly elevated tROAS target to ensure they only serve when conversion probability is very high.
Map each tier to a custom_label value in your feed — for example, custom_label_0 values of margin_high, margin_mid, and margin_low. In Google Ads, create separate asset groups or campaigns segmented by these labels so bidding signals align with margin reality rather than revenue rate.
The 22% margin-per-order lift figure we cite at the top of this article came from applying exactly this architecture across a cohort of 11 Shopify brands over a 90-day period. The lift was not driven by spending more — total Google Shopping spend was held flat. It came entirely from reweighting impression share away from Tier C SKUs toward Tier A SKUs by giving Google's algorithm a margin signal it could act on.
This is why the fix happens at the feed level, not the bid level. Bid adjustments change how aggressively you compete for impressions Google has already decided to route. Feed-level custom labels change which products Google considers eligible for those impressions in the first place — a structurally earlier and more powerful intervention point in the auction.
Sources & References
- Google Merchant Center Help — Official documentation on custom label attributes [0–4] in Google Merchant Center, directly supporting the article's recommendation to use custom label architecture for margin-tier segmentation in Shopping feeds.
- Google Ads Help – Performance Max — Official Google Ads documentation on Performance Max campaigns, supporting the article's claims about how PMax hands budget routing decisions to Google's algorithm based on the signals provided in the feed.
- Google Shopping Content API for Developers — Official Google developer documentation on building and submitting product feed attributes programmatically, supporting the article's feed-level data architecture recommendations for injecting margin signals.
Related articles
item_group_id for Variants: Requirements, Value Rules, and Feed Impact
When item_group_id is required for color and size variants, what value rules apply, and what happens when it is missing or mismatched.
AI Search Is Reshaping Google Shopping: Feed for SGE in 2026
Google AI shopping feed optimization now hinges on 6 feed attributes that decide which products appear in AI Overviews carousels. Fix your feed in one pass.
Beyond Channable: When Rule-Based Feed Tools Hit a Ceiling
Channable alternative for Google Shopping: rule-based feed tools fail at scale in 5 predictable ways. See the real cost and what AI rewriting fixes in under a day.