Keep the body. Test the opening.
Based on the early results, I kept the original as the control and briefed three opening sequences: checking the company, feeling the fabric and showing the duvet going into the machine.
The brief keeps the speaker, body and offer fixed and gives each opening its own first eight seconds. The next test has a clear question to answer.
Dates, sources and follow-up results
How the brief maps to the ads
Original concept MES-TRUST-GEN-PROD is linked to ML14 in the follow-up brief MES-TRUST-GEN-PROD-IT1-HOOK (ML63). Frames are matched through the existing portfolio asset register. The two shown ads are IDs 120252717366440279 and 120252717413610279.
The decision timeline
On 8 September, the client reported CHF 1,500 spend at 2.8× ROAS for the original concept. This report was not independently reconciled to the attribution platform. The 21 September iteration brief used that early read; delivery notes record two opening sequences supplied by 29 September.
The headline results come from the later 9 November Meta export covering 11 August–8 November. They describe the original ML14 variants, not the ML63 follow-up. Purchases and ROAS are Meta-reported, not incremental lift or unique new customers. These ads are included in the 717-purchase total elsewhere on the site.
The retained control’s later results
The 21 September brief explicitly kept ML14-H1 running unchanged. The 12 November export showed it still active, with 30 purchases, CHF 1,498.87 spend and 2.66 ROAS for 13 October–11 November. This window overlaps the headline reporting period; these purchases must not be added to the 212 total.
A separate variant’s observed pause
The 13 October review recorded ML45-H2 at 1.38 ROAS on CHF 701.03 spend, below the brand’s recorded 1.73 reference. The review proposed a controlled opening test; it did not document an instruction to pause H2.
The 27 October export showed its ad set paused, with CHF 739.58 spend through 26 October. The 9 November export still showed it paused, with the same spend through 8 November. Both windows cover that ad’s launch onward. The client controlled delivery; the record does not identify who made the pause decision.
These follow-up records establish continued original-ad performance and a separate ad-set pause. They do not establish money saved, incremental lift or a successful ML63 iteration.