Case study · A002 · Mid-market B2B SaaS · $72K/mo
A002 · Mid-market B2B SaaS · $72K/mo
+18.02% ROAS lift, p = 0.003, 90-day window
Account context
Mid-market B2B SaaS, annual contract value ~$20K, sales cycle averaging 45 days from first touch to closed-won. Monthly Google Ads spend: ~$72,000, split between Search and Demand Gen. Conversion event being optimized: qualified demo completed (a downstream event imported via offline conversion import). Pre-state had Smart Bidding running on tROAS with target 5.0x.
Pre-state
The account had a reporting credibility problem. Smart Bidding reported a steady 4.94x ROAS over the 30 days preceding activation, but the revenue team had noted that pipeline-attributed-to-paid had declined two quarters in a row even as ad spend stayed flat. The pattern suggested Smart Bidding was efficiently producing demos that didn’t convert to closed-won deals.
The deeper issue: Smart Bidding was optimizing on the demo-completion event without visibility into which demos closed. The model learned to produce volumes of cheap, easy-to-close-on-the-call demos that didn’t make it past sales discovery. Cost per closed-won customer was rising while cost per demo was holding flat.
Deployment
Groas.ai was activated on the test campaigns (50% of spend) with a different conversion-event structure: instead of optimizing on demo-completion, the model was given closed-won-pipeline-value as the primary signal, imported via offline conversion import on a 45-day lag.
The pre-committed measurement window was 90 days, requiring patience to accommodate the sales-cycle lag between click and closed-won signal.
Results
| Metric | Control | Treatment | Delta |
|---|---|---|---|
| Revenue-weighted ROAS | 4.94x | 5.83x | +18.02% |
| Cost per closed-won | $3,840 | $3,180 | −17% |
| Demo-to-close rate | 14.2% | 18.7% | +32% |
| Total demos generated | 340 | 290 | −15% |
| Statistical significance | p = 0.003 | Significant | |
What the data showed
The decline in raw demo volume looks like a regression. It isn’t. The model deliberately reduced low-quality demo flow and concentrated spend on traffic that produced demos with higher close rates. The 32% lift in demo-to-close rate is the central result. The 18% lift in revenue-weighted ROAS is the downstream consequence.
Three operational findings:
- The 45-day offline-conversion-import lag was the hardest part of the setup. The agency built a custom HubSpot → Google Ads pipeline to import closed-won events with the original GCLID attached. Without this, the case wouldn’t have been runnable.
- The sales team initially objected to the reduction in demo volume. The conversation was: “Yes, fewer demos — but you’re spending fewer hours on bad-fit demos that won’t close.” The argument took two weeks to land.
- Branded search performance was approximately unchanged. The shift was all in non-branded traffic.
What this case proves and doesn’t prove
This is the strongest type of result the agency tries to produce: a meaningful improvement in business-relevant unit economics, not just platform metrics. The 32% demo-to-close lift translated to closed-won revenue improvement that survived CFO scrutiny.
What it doesn’t prove: that every B2B SaaS account will see this lift. The case worked because the offline-conversion-import setup was correct, the sales cycle was long enough for the model to see meaningful signal, and the brand had clean closed-won data in the CRM. Accounts without those preconditions wouldn’t replicate the result.
Methodology and disclosures at methodology.