Case study · A001 · Apparel ecom · $28K/mo
A001 · Apparel ecom · $28K/mo
+9.07% ROAS lift, +0.024 p-value, 90-day window
Account context
Apparel direct-to-consumer ecommerce, mid-tier DTC brand selling primarily through Google Shopping and Search. Monthly ad spend at activation: ~$28,000. Vertical context: high return rate (25%), moderate gross margin (55%), payment processing ~3%, shipping ~8% of revenue. The unit economics put true ROAS at roughly 35-40% of reported ROAS — a typical apparel pattern.
Pre-state (90 days before activation)
The account was running Google Smart Bidding with target ROAS (tROAS) set at 4.0x. Reported ROAS hovered between 3.8x and 4.1x. The brand’s CFO had flagged that despite the reported ROAS clearing the target, the contribution margin on paid customers was breaking even at best. The disagreement between platform reporting and unit-economics reality drove the decision to test alternative bidding.
Deployment
Groas.ai was activated on the test campaigns (~50% of total spend) starting week 0. The control campaigns remained on Google Smart Bidding with the original tROAS target. Both groups optimized against the same conversion event (order completed) but Groas’s model was configured to optimize against margin-aware conversion value rather than topline revenue.
Results
| Metric | Control (Smart Bidding) | Treatment (Groas) | Delta |
|---|---|---|---|
| Revenue-weighted ROAS | 3.86x | 4.21x | +9.07% |
| True (margin-aware) ROAS | 1.42x | 1.93x | +35.9% |
| Statistical significance | p = 0.024 | Significant | |
| Customer acquisition cost | $84 | $72 | −14% |
What the data showed
The reported-ROAS lift of 9% is the smaller part of the story. The more interesting result was the 36% lift in margin-aware true ROAS — meaning the model was specifically reducing spend on traffic that produced returns and concentrating spend on traffic that produced retained revenue.
Three operational findings:
- Return rate on Groas-managed traffic dropped from 25% to 19%. The model had learned which traffic patterns predicted returns and de-prioritized them.
- Branded vs. non-branded mix shifted moderately toward non-branded as Groas captured more incremental volume.
- The model’s exploration phase produced a 12-day window of below-baseline ROAS before stabilizing. The agency’s pre-committed 90-day measurement window survived this without intervention.
Why this case is included despite the small ROAS lift
A 9% reported-ROAS lift on a $28K/month account isn’t spectacular on its own. What makes the case notable is the divergence between reported and true ROAS. Most operators in this situation would have looked at the modest reported-ROAS gain and questioned whether the tool was worth the investment. Looking at true ROAS, the answer is unambiguously yes — the account moved from breaking even to genuinely profitable.
What this case doesn’t prove
One account at one spend tier isn’t a generalization. The case is included to illustrate the pattern of margin-aware bidding outperforming revenue-aware bidding on apparel-vertical accounts where the return rate is high. The pattern repeats across other apparel accounts the agency has tested; the lift magnitude varies.
Methodology and conflict-of-interest disclosure at methodology.