Winning Variables

Best practices — what is working

Grounded in our own coverage data (Current Agenda §8), the July 2026 retro, and the Creative Strategy Best Practices playbook. Claims marked inference are patterns we are reading, not measured results.

• We have never opened most of our own taxonomy. Roughly 30 of 116 defined variable values have taken real spend — 5 of 26 subavatars, 9 of 21 angles, 6 of 22 hook styles, 6 of 24 video formats, ~2 of 20 statics. The ceiling we keep hitting is a range problem, not an ideas problem.
• Statics convert; video earns attention. In July the best CVR in the account sat in statics — Headline 6.30%, Feature-Benefit 6.08%, Lifestyle Image 5.58%. Do not treat statics as filler.
• Avatar and angle correlate with performance more than anything else we choose. Pick those two first, then build the hook and script around them.
• A better hook is not a better ad. July took hook rate 18.7% → 26.9% and ROAS fell 1.08 → 0.95. Judge on hold, click, landing-page fit and conversion together.

Subavatar

Working now

• No proven subavatar winner exists to copy. Five have taken real spend and none of the five clears 0.50.
• The signal is one level up, at audience: 55–64 M does 0.84 ROAS at $113 CPA (our most efficient, least funded), and 18–24 M does 0.71 at $143 CPA (our best large audience).
• Bulking drove our performance but is exclusionary — it caps out. Current direction concentrates ~80% of concepts on Fitness Curious: interested in fitness, not yet a committed hard-gainer.

Unopened, worth opening

• `SA:Sarcopenia Man` — 55–64 M is our most efficient audience and our least funded. This is that exact person, never tested.
• `SA:Gym Beginner` — maps to 18–24 M, our best large audience.
• `SA:GLP-1` — promised since June, still $0 spent.
• Stop defaulting to `SA:General` — 25 ads to nobody in particular, and our second-biggest bucket.

Briefing rules

• One subavatar per script. Multi-subavatar concepts make a win or a loss undiagnosable.
• Write the subavatar as a person, in their language, from reviews and Reddit — not as a demographic cell.
• Drop the old skinny-guy / "I forget to eat" framing. The problem is eating enough of the right things and staying consistent.

Format

Working now

• `FS:Progress` is the best-performing format in the account — 1.13 ROAS at an $87 CPA. Caveat: that is 2 ads on $609. A lead to fund, not a proven winner.
• Statics are the conversion engine — see the July CVRs above.
• Inference: progress/journey framing pre-qualifies the buyer, which is why it holds while generic formats do not.

Unopened, worth opening

• Proof formats are entirely absent. Nothing running in `FV:Testimonial`, `FV:Personal Story`, `FS:Review` or `FS:UGC Story`.
• Inference: that absence is the likeliest cause of the add-to-cart → purchase collapse. People arrive convinced and find nothing that proves it.
• 18 of 20 statics and 18 of 24 video formats have never been tested.

Briefing rules

• Every static tests a hypothesis. An image that tests nothing has little purpose.
• Match the format to the subavatar and the awareness stage, not to what looks novel.
• Inference: pair organic-looking hooks with longer VSL-style copy — organic hooks stop but don't convert, and long copy explains but has a weak hook and hold.

Angle

Working now

• Benefits beat features. Setting High Protein aside, feature angles do 0.27 and benefit angles do 0.54. `AN:Stay Satiated` (0.72) describes the customer's life; `AN:Complete Nutrition` (0.29) and `AN:Balanced Macros` (0.15) describe the powder.
• "High Protein" is a mechanism, not an angle — a spec, not a reason. It currently carries 58% of our spend.
• Inference: If-Then and Contrast framing converted ~2.4× better than pure Relatability in July. Relatability earns the stop; it gives no reason to act.

Unopened, worth opening

• 12 of 21 angles have never been tested. Open the benefit-led four first: `More Energy`, `Recover Better`, `Less Overwhelmed`, `Look Better`.
• Cut `Complete Nutrition` and `Balanced Macros` — 32 ads and $3,098 spent describing the product rather than the life.

Briefing rules

• One angle per concept, held all the way through the script.
• Keep angles concise — often a 3–5 word value proposition grounded in a real insight.
• Ask "how is ours uniquely convenient?" A generic benefit is not an angle.

Hook style

Working now

• We hook 23.5% and hold 12.4%. Our best hookers are our worst holders — POV 43.9% → 8.2%, Static-to-Video 44% → 7.2%.
• `HS:Meme` is 54% of spend, which is why the account reads as though we run one hook style.
• Inference: our hooks have attributes and emotion but no open loop. A meme lands its joke in two seconds and the loop shuts before anyone gets curious.

Unopened, worth opening

• 16 of 22 hook styles have never been tested. Open the structurally open-loop ones first: `HS:Curiosity Gap` (literally the open-loop hook, never run), `HS:Contrarian`, `HS:Mistake Callout`, `HS:Objection Callout`.
• Proof-led hooks are untested and pair with the missing proof formats: `HS:Social Proof`, `HS:Testimonial`, `HS:Authority`.

Briefing rules

• A hook is text + visual + audio, and all three get designed. The visual usually does the work of stopping the scroll.
• Build a specific unresolved question — an open loop — not just a louder opening.
• Three hook variants per concept, diversified across genuinely different emotional states, not reworded premises.
• Test every hook by cutting the first 3 seconds: is there still a reason to watch?
• Check valence. Ranking and tier-list formats create intrigue, but reframe negative frames toward a positive result.

⚠️ These tables are out of sync with the approved taxonomy. The Current Agenda's approved lists (§8) carry values this page does not — `SA:Breakfast`, `SA:ADHD`, `SA:General`, `HS:Meme`, `HS:Before & After`, `HS:Objection Callout`, and the static format `Progress` among them. Every coverage count above is drawn from the agenda's approved lists, so those are the numbers to trust. Reconciling the two lists is real work, not a formatting detail.

🧬 Subavatar List

🧬 Subavatar LP Pairing

🧬 Subavatar LP Pairing v2

🚝 Format List

🫠 Angle List

🏛️ Hook Style List