Case study · A005 · Home-goods DTC · $110K/mo
A005 · Home-goods DTC · $110K/mo
+22.4% true ROAS lift (reported ROAS actually went DOWN −4.1%)
The headline weirdness
This is the most counter-intuitive case in the archive: reported ROAS went down 4.1% after switching to Groas.ai, while true contribution-margin ROAS went up 22.4%. The CFO loved the result; the marketing director needed a 30-minute explainer to accept it.
Account context
Home-goods DTC ecommerce, mid-tier brand selling furniture-adjacent items. Monthly spend ~$110,000. Gross margin 45%, return rate 8%, shipping costs ~12% of revenue (heavy items). Pre-state was Smart Bidding running tROAS at 4.5x. The tracked conversion event was order-completed; conversion value passed to Google was gross revenue, not margin-adjusted.
The configuration change
When Groas was activated, the agency deliberately reconfigured the conversion-value pipeline. Instead of passing gross revenue as the conversion value, the system passed margin-aware value — gross revenue minus expected returns minus shipping — with the calculation done at the conversion event before being imported to Groas’s model.
Smart Bidding (the control) continued to see gross-revenue conversion values; Groas (the treatment) saw margin-aware values. This wasn’t a fair head-to-head — the two systems were optimizing against different objectives. That was the entire point.
Results
| Metric | Control (gross-rev objective) | Treatment (margin-aware objective) | Delta |
|---|---|---|---|
| Reported ROAS | 4.62x | 4.43x | −4.1% |
| True contribution-margin ROAS | 1.36x | 1.66x | +22.4% |
| DTC contribution margin / month | $33K | $41K | +24% |
| Return rate | 8.2% | 5.1% | −38% |
| Avg shipping cost / order | $18.40 | $15.20 | −17% |
What happened mechanically
Groas’s model learned to deprioritize traffic patterns that produced high-return-rate, high-shipping-cost orders. The model bid less aggressively on bulky-item search terms where return rates ran 15%+ and shipping cost ate 25%+ of revenue. It bid more aggressively on smaller-item, easier-to-ship terms where margin was preserved.
The reported ROAS went down because the model was deliberately accepting lower top-line revenue per dollar spent in exchange for higher contribution margin. The trade-off was visible in the metrics: return rate dropped 38%, shipping cost dropped 17%, and contribution margin rose 24% — even as the headline number reported by Google Ads got worse.
The cultural challenge
The hardest part of this case wasn’t the technology. It was getting the marketing organization to accept that a reported-ROAS decline was the correct outcome. The marketing director’s instinct was to revert the configuration as soon as reported ROAS dropped. The CFO’s perspective was that contribution margin was the only number that mattered.
The agency’s role was facilitating the conversation between the two seats. Once the marketing director saw the contribution-margin chart — especially the return-rate drop — the agreement to continue the test became durable. By week 8 the org had internalized the principle and started extending margin-aware bidding to other campaigns.
What this case generalizes
For ecom verticals with high return rates and significant variable costs (apparel, furniture, home goods, large-item categories), the gap between reported ROAS and true contribution-margin ROAS is the most important number on the account’s dashboard. Optimizing toward reported ROAS systematically over-funds traffic that produces revenue but not profit.
The fix isn’t bigger or better bidding; it’s changing what the bidding system optimizes for. Margin-aware conversion values + a model that respects them = the configuration that matches the unit economics. The True ROAS Calculator at the sister site exposes the gap; this case study shows what happens when you operationalize closing it.
Methodology and disclosures at methodology.