Lumeo
29 ads over the ROAS goal in 60 days, on a $200K-a-month account
A DTC men's performance gummy spending roughly $200K a month on Meta, compliance-constrained and static-led, in a market where every competitor makes the same claim. We ran creative strategy end to end: the research, the diagnosis, every brief in the account, and the editor pipeline that built them.
- 1DiagnoseThe 60-day export read properly; a $23K mistake withdrawn.
- 2RebuildThe 216 video: 1.08 to 2.02.
- 3IterateDCT#115-H3: 1.47 to 2.06, generation by generation.
- 4Fix the leaksThe quiz step, the ad names, the disapprovals.
The 60-day export before we started: $395,366 spent at 1.295 ROAS and a $66.30 cost per purchase, against a 1.5 goal. 91% of the budget sat in one campaign, 79.9% of spend ran on ads named like designer files so nothing traced back to a brief, and the quiz funnel lost two-thirds of its traffic at question one.
HYLN's part, Santi on the account: creative strategy end to end. Figures are the account's own reporting for the last 60 days, pulled on 27 September and graded against the account's own 1.5 ROAS goal. Next on the plan: the Q4 promotion calendar with ban prevention and a fallback, one guarantee figure across all creative, and funding the starved winners before briefing more new work.
Diagnose before briefing. Fix the leaks that were not creative. Then fund what was already winning and rebuild what was losing, before making anything new.
The diagnosis. The account was reclassified by its spend winners rather than by impression, and a $23K budget move that would have broken the efficient campaigns was withdrawn before it happened.
The 216 turnaround. The account was paying 7x more for the worse cut of the same video: $18,233 at 1.08 ROAS, while the better cut sat starved at $2,496 and 1.46. The hook stopped the wrong people, and the offer landed at a point where 2.3% of viewers were still watching. Budget moved to the better cut, the mechanism and the guarantee moved to the front, and the copy was rewritten for the man watching. The rebuilt cuts run at 1.89 and 2.02.
Iterating the only family that kept improving. DCT#115-H3 was a video in a static-led account, capped at about $2K per child. Every generation beat its parent, 1.47, then 1.64, 1.85, 1.94, now 2.06, so the call was budget before new creative.
The leaks that were not ads. The quiz funnel's first step dropped 66% of traffic, about $9.7K a week at 0.79 ROAS: fix step one or pull the spend. Untraceable ad names carried 79.9% of spend: the brief number now sits in every ad name, so spend maps back to the brief that made it. Disapprovals were instance-level, not content: duplicate and relaunch.
The system underneath. 560 post-purchase calls mined and about 1,070 verbatim customer lines from calls, Reddit, Trustpilot and ad comments. An avatar system of 5 cores and 16 sub-avatars, each graded against customer evidence, which surfaced an unmodelled buyer: men with no problem who simply want more. An 8-brand competitor teardown and a market sophistication audit. Every numbered ad in the account briefed, statics and video, plus 6 ranked iteration briefs off the spend winners, a net-new concept with its static companion, and an 8-concept month plan of 24 image-led ads. 3 video editors brought on and run through a shared brief queue, and a live watchlist of 31 competitors.

Most of what moved this account was not a new ad. It was reading the account properly, moving budget off the worse cut of a video the brand already had, and fixing a quiz step and a naming convention so good creative stopped being judged in broken conditions. Diagnosis first, then briefs.
