Process, Testing & Scaling

Test Order (Trunk → Branch → Leaf)

Test the highest-impact variable first: angle > awareness level > Lead/Headline/Image > format > copy length.
Trunk = awareness × motivation; wrong here and no copy saves it. Lock this before anything else.
Sequence: Awareness×Motivation → dominant desire → angle type → voice → ad type → creative format → lander → HL/Lead/Body/Image/CTA.
Color-block winners are hypotheses, not winners; take top 3 straight into full ads to confirm.
Never change two big variables at once: hold awareness while testing motivation; type before format; lander type before voice.
Lander beat order (highest leverage first): Headline → Lead (first 200 words) → Body → Images → CTA.
Sky House default control lander = Top 5 / Listicle; start there, don't re-prove it.
Sophistication ≠ awareness: if ~50 competitors run your mechanism, shift to a novel mechanism or identity — a new hook won't fix it.

CBO + one broad ad set; broad targeting is mandatory — creative is the targeting, narrow fights the algo.
Each testing ad set locks ONE variable = one hypothesis; write the hypothesis before building ads.
Launch 15-20 genuinely diverse creatives; expect only 3-5 to get real delivery — that's normal.
Max 6 ads/ad set in testing (clean signal), up to 12 in Control (scale + diversity).
Never put new ads into a winning ad set (Pack System); build a fresh set instead.
Deploy winners across ad sets via 'existing post' to keep Entity ID + social proof; duplicating resets both.
3×3 framework: 3 angles × 4 formats = 12 unique-Entity-ID ads (NOT Sky House's 4x4, which is 4 ad texts × 4 headlines).
Budget split: 60-70% cold prospecting, 30-40% retargeting (retargeting only after 1,000+ audience).

Ads scoring >60% similar (visual+copy+audio) collapse to ONE Entity ID and split delivery — launch 100, only 10 deliver.
Headline tweaks and color swaps are NOT new ads; repurpose winning MESSAGE across genuinely new formats.
Score below 60%: differ on format, motivator, visual style, voice, and tone simultaneously — not one dimension.
Same-body hook variation is dead; first-3-second clones count as the same ad.
Run a portfolio across all 5 awareness levels; the broad audience contains all of them — let the algo match.
Ongoing cadence: 5 new concepts/week across 2-4 formats; budget ~5-6 losers per winner (15-18% real win rate).

Never kill before 72 hours untouched; day-of-week skews 1-day data.
PAUSE: CPC >$2.00 after $50 spend AND zero add-to-carts AND zero purchases.
Kill/refresh: frequency >3.0 AND CTR drops 30%+.
No meaningful delivery after 48-72h = Andromeda rejected it, kill.
Kill ad set after 1x target CPA with no purchase; kill ad at CTR <1% after 1,000 impressions.
Min-spend gate before judging (Buontempo): CPA<$100→3.5-4x · $100-149→3x · $150-249→2-2.5x · >$250→1.5-2x.
Winner = hit min-spend gate + at/under target CPA + sustained over 7-day window (kills spike false-positives).
Read aggregate POD/concept performance, not single-ad CPA — Meta sequences ads and one warms the converter.

Pre-scale gate: known LTV/AOV + backend bandwidth for 2x buyers + automations in place. Fail any = hold.
Scale trigger: CPA at/below target 3 days + frequency <1.8 + 50+ events.
Default rate: +20% every 3 days; go +25-30%/day only with full unit economics + 5 profitable days.
Never raise budget >30% in one step — algo needs to relearn.
Break-Even ROAS = 1/(1-COGS%-Overhead%); Scaling ROAS = Break-Even + 1.0. Between them = maintain, below = cut.
Affiliate: break-even EPC=CPC, scale at EPC>2x CPC; track via network not pixel (reporting delay).
Build a financial model BEFORE launch — without your true break-even CPA/CPC you'll kill winners early.
Frequency >2.0 = pull back, audience fatigued.

Rising CPM is the PRIMARY fatigue flag; then declining CTR, frequency >3, rising CPA, dropping hook rate.
Fatigue hits in 2-4 weeks; plan new creative every 1-3 weeks.
Refresh by changing archetype/emotional driver/format with the SAME winning message — not color/headline tweaks.
Don't kill winners for being 'old' — let the numbers decide.
CPCQ + volume stable but lead quality dropping = audience pocket gone bad, not fatigue; pause CAPI feed + duplicate campaign to reset Meta's ML.
Market saturation (CPMs rising niche-wide, competitor ad libraries converging) = change the angle/mass-desire entirely, not the hook.

Hook rate: 20-25% good, 30%+ great, <15% cut. Hold rate: 20%+ good, <15% cut.
Link CTR: 1.0-1.5% good, 2%+ great, <0.7% cut; big outbound-vs-link gap = clickbait.
Healthy CTR band 0.90-1.60%; CPA cut at >3x target.
Scale confirmation: ad set spent 5-10x AOV profitably.
Build custom Meta columns: Hook Rate, Hold Rate, cost-per-10k-reach; save as preset.
CAPI/Event Quality 8+ is mandatory in 2026 — pixel-only is broken (88% iOS opt-out).
Direct call-out hooks look bad on hook rate by design (non-targets scroll) — judge on downstream conversion.
Optimize for the deepest funnel event with volume (50/wk ideal, 20/wk minimum); wrong optimization event = most common audit quick-win.

No clear winner + low CTR → go back one level, retest awareness or motivation.
No winner but decent CTR, no conversions → move to lander testing early; it's a page/traffic problem.
Signal too noisy → double spend to 2,000+ impressions/test before deciding.
One ad hogs spend, one starves → force equal spend via separate campaigns, then compare.
New offer, don't know what works → Format test (Reel/Static/Founder/Carousel, equal budget) or Marksman (3 different ANGLES, equal budget).