What the engine delivered
The challenge
The client had a high-consideration B2C product category (garden houses) with €1M in prior-year revenue but zero paid-experiment history, no creative or messaging winners on record, and broken UTM tracking that made GA4 nearly blind to ad-attributed revenue. Three channels needed to be evaluated simultaneously with no prior data on which would perform. Last-click attribution was systematically undercounting ad impact across a multi-session customer journey.
Our approach
- Stretch designed a structured 9-experiment sprint over 13 weeks, starting with a UTM tracking fix to make GA4 usable, then running A/B tests across 5 messaging angles, creative formats (static, video, carousel), retargeting audiences, lookalike segments, and a new Pinterest channel. A quiz lead engine was built to capture emails and enable 1-to-1 revenue attribution outside the pixel by manually matching submissions against the weekly sales export. Experiments were reviewed every three weeks with a clear mandate to kill underperformers.
The result
Meta scaled to €803K in platform revenue at 101x ROAS; Pinterest delivered the highest ROAS of the sprint at 189x on just 6% of budget (€0.75 per purchase); Bing was killed at week 9 after returning only 0.72x ROAS. The quiz generated 2,182 leads at €1.18 CPL and €42,546 in verified revenue across 22 confirmed sales. Static creative won on ROAS (263x vs 111x for video), while the 'installation service' messaging angle delivered the lowest CPC at €0.26.
One product category, zero experiment history, and a baseline of €1M in prior-year revenue. The brief was simple: find out what works. The condition was uncomfortable: we had to start with nothing.
Want the full case study?
Download the complete version including methodology, experiment details and full results.
