Dissection — "A · Says Who · 15s" + age-variant plan (40s / 50s / 60s)

Written: 2026-08-21 · Ad: ads/before-after-strobe-cut/final.mp4 (14.79s, 1080×1920, 24fps) Lineage: clone of @kangmiratips21 "I changed my life at 62" (13.6M views, 10.3% like rate) — teardown in teardown-kangmira-7606799992237444383.md, build plan in genplan-kangmira-beforeafter.md.

---

1. Why this is the ad to multiply

FiftyStrong quiz-validation test, 2026-08-05 → 08-20, account act_1761459111652871:

A wins on every downstream metric. It is already cloned into the 50Queen launch campaign (ad 120252196210740131, paused, pointed at 50queen.app/quiz).

The age breakdown nobody looked at until now

Pulled fresh from Meta (same date range, ad A only):

Two findings that shape the variant plan:

1. 45–54 is the cheapest cohort on both leads and checkouts — a third of the cost per checkout — but Meta gave it only 9% of spend, because delivery chases cheap impressions and older impressions are cheaper. A 40s-facing variant in an age-capped ad set is the only way to force spend into the bracket that converts cheapest. (Caveat: 3 checkouts is a small sample — treat as a strong hint, not a proof.)
2. 65+ has the highest CTR (15.1%) despite the creative saying "you're 56". A 60s-facing variant isn't a reach — that audience is already the most reactive to the format; it's currently just watching an ad addressed to someone a decade younger.

---

2. Anatomy of the winner, second by second

Verified against the frames of the shipped file. Audio: 113.6 BPM track, drop at 6.82s, bar lines at 8.93 / 11.05 / 13.16. Every cut sits on the grid.

Captions (burned in, Avenir Black, white with 3px black stroke, ~72% frame height):

• "you're 56 / it's too late to change" — on screen 0.00 → 5.08, i.e. the entire before.
• says who? — appears at the resolve (6.83) and stays to the end.

Copy that ran with it (variant A1, locked in ad-copy-spec):

"It's too late to start." It isn't. Strength, energy, and moving through an ordinary day without thinking about it — you can still have all of it at 48, at 55, at 62. It just takes exercising smart. Two minutes of questions, then a workout plan customized to your goals and built around your limitations. At home, no equipment.

Headline Get your personalized workout plan now · description At home · No equipment · Built for women over 50 · CTA Learn More → quiz. Optimised on Lead (quiz complete).

---

3. The mechanics — why it pulls 14% CTR and $0.85 leads

1. The hook is a quoted insult, not a claim. "you're 56 / it's too late to change" is in quotation marks — attributed to an unnamed someone. It qualifies the viewer (every woman who has heard a version of this), manufactures a villain without naming one, and turns the whole video into a rebuttal instead of a brag.
2. 46% of runtime is the before, and the before is genuinely unflattering. Cold light, no makeup, slumped shoulders, pinching her own stomach. That investment is what makes the after mean anything. Most transformation ads spend 90% on the after and land flat.
3. The caption dies at the reveal. The insult is on screen for 100% of the before and 0% of the after. says who? is the rebuttal — three words, no argument. The structure does the arguing.
4. The edit is cut to the music, not the other way round. Cut rate accelerates (2.6s hold → jump cuts → 10Hz strobe), then the strobe resolves on the beat drop, then a 2.2s hold that feels like relief. The drop is the transformation.
5. The proof is stacked, not singular. Three after-contexts in 8 seconds — playful (bedroom), capable (trail run), out living (car). Three locations signals "this is her actual life", not "her one good angle".
6. She stays recognisably herself. This is where the shipped ad improved on its 13.6M-view source: the source goes grey→brunette; ours keeps the grey-blonde hair through the transformation. The delta is body, posture and energy — not identity. That is precisely the teardown's "smaller delta converts better" hypothesis, and the checkout numbers vindicate it.
7. The loop is a hard reset. Held grin in a car → cut to a grey woman in a bathroom. Maximum contrast at the loop point; the rewatch is nearly involuntary.
8. The ad self-selects the audience the targeting can't. "you're 56" in frame 1 acts as a qualifier — the click is pre-warmed to a specific decade. Which is exactly the dial the age variants turn.

What the winner proved you can drop

The shipped cut removed the source's couch shot and the entire 6-still life montage (hospital, cashier, kitchen…) — the segment the original teardown thought carried "most of the emotional load." Performance says otherwise: the load-bearing elements are insult → unflattering before → strobe-on-drop → caption kill. Variants only need those four.

---

4. The variant grid — what's frozen, what turns

The clean experiment changes the age variables only and freezes everything else, so any performance delta is attributable to age-match.

Frozen across all variants (byte-identical or same-spec):

• The cut list and timings (0–5.08 before / strobe / 6.83 resolve / three afters)
• The audio track and its 6.82s drop
• Caption typography, position, and the on/off timing
• The insult → says who? mechanic
• The three after-context types: at-home playful · outdoor cardio · seated-in-car grin
• Copy shape (quoted dismissal → rebuttal → three ages → quiz offer), headline, description, CTA, Lead optimisation
• Same-person rule: hair colour and identity survive the transformation — only body, posture, light and energy change

The four dials that turn per variant:

Why these lines. The dismissal must be one that decade actually hears. Nobody tells a 45-year-old "it's too late" — she hears decline-fatalism: "it's all downhill from here." The 67-year-old hears role-confinement: "just be a grandma" — which is the original source's exact line, proven at 13.6M views, aimed at the bracket where our CTR is already highest. The 56 control stays untouched — it is the 50s variant, and it keeps its history as the benchmark.

Backup lines if a hook underperforms at 50 leads: 40s — "wait till 50 / it only gets worse"; 60s — "you're 67 / just take it easy"; 50s refresh (round 2) — "at your age just walk more" (the genplan's original pick, never tested, and the only line that names the thing the app replaces).

Character notes per decade

• 40s before (~45): dull dark-blonde hair with visible grey roots, ~20lb around the middle, dark circles, oversized tee — perimenopause tiredness, not old age. 40s after: same hair, upright, defined arms, bright-eyed. The delta reads as "got her energy back."
• 60s before (~67): fully silver, blunt cut, more pronounced slump, weight through middle and upper back, the current before pushed ~10 years. 60s after: silver hair stays silver (the shipped ad already proved a grey after works), lean but soft-armed, moving with ease — carrying, walking briskly, laughing in the car. Strong, not sculpted.

---

5. Targeting — the part that makes the variants worth building

The current ad set is one broad pool (women 48–65, Advantage+ audience off). The age breakdown shows Meta starves the youngest, cheapest bracket. So:

Same campaign, CBO at $50/day, hard age caps (Advantage+ audience OFF — same discipline as the current ad set spec), Lead on pixel 1055150507004620, placement customization 9:16 → stories/reels + 4:5 → feeds via asset_feed_spec (reuse the exact creative spec from the 50Queen launch build; captions sit at 72% height and survive the 4:5 centre crop — verified on the current pair).

Decision rule: run each cell to ~50 leads (~$45–60 at current CPLs, so roughly $150–180 total). Compare CPL first, checkout rate second. Kill any cell whose CPL exceeds the control by 50%+ at 50 leads; swap in the backup hook line for a killed cell before abandoning the decade.

---

6. Production — everything is reusable except the pixels

The pipeline that built the winner (per genplan-kangmira-beforeafter.md, as-built notes):

1. Character sheets — nano_banana_pro, after-first, then derive the before from the after ("make her heavier/older" anchors better than the reverse). 2cr each.
2. Start frames / storyboards — the shipped build used 8-panel storyboards (pose + transition alternating) fed to minimax_h3 as image_references. The strobe pair must match on head scale + vertical head position; alpha-blend the pair at 50% to verify before generating video.
3. Clips — 4 per variant via minimax_h3. Generate 6s, use 2–3s; before-clips drift flattering the longer they run.
4. Post — the existing ffmpeg cut list, the same audio, drawtext captions with the new line. Zero new editing decisions.

Cost: ≈122cr per complete pass; budget 2–3 passes per variant (the before is always the asset that needs re-rolls) → ~250–350cr per variant, ~500–700cr for both new variants. Account balance was 1,907cr as of 08-06 — comfortably enough.

Known traps carried over from the first build: the before comes out too flattering on pass 1 (name the failure modes: loose triceps, rounded belly, under-chin, overhead light with no fill); minimax_h3 drifts framing and grade from its start frame, so re-align the strobe pair after generation; never pass a character sheet and a scene photo containing a person in the same reference set.

---

7. Next step

Nothing generates until approved. The cheap de-risk, per variant, is the same as last time: 6 credits — after sheet → before sheet → strobe-pair frames → alpha-blend check. That answers "is the before convincing" and "will the pair interleave" for 5% of a full pass.

Proposed order: V60 first (highest-CTR bracket, proven hook line, smallest creative risk), then V40 (biggest economic upside, least proven). Control needs no work.