Paid · Meta · 2026-08-13 07:49
Making retargeting work
The plan, the creative evidence, and what the paid-ads seats say to actually do — in one place. Every figure is read live from the creative pull, never typed.
The one thing to put in front of you
Rung 2 is 50% off — and 50% off is the one retargeting offer this account has tried and had fail. Everything that worked sold a deep Lifetime discount that no longer exists. And the people you are retargeting declined a free trial, so "pay half, now" is a harder ask than what they already turned down, not an easier one.
That does not mean don't run it. It means run it as a test with the $89 cost-per-sale kill condition live from day one, frame the money, not the percentage — "$78.50 for the year" or the proven framing from our own upgrade emails, "6 months free" (they already sell it as "keep $142.64 = ~6 months free") — never a bare "50% OFF", and don't let anyone forecast 1.72 onto it. The budget number is genuinely your call — the two seats split, and their reasoning is set out below rather than blended into one recommendation.
The verdict — what to actually do
Retargeting is the only thing this account has ever done well. It returned 1.72 ROAS at $43 CPA across $6,807, while everything else returned 0.32 at $178 across $93,011. The plan is right to run it. Three things in the plan will stop it working.
- Verify the purchase event BEFORE un-pausing anything. Nic · 10 min · blocking Check what event the three paused campaigns optimise for, and Event Match Quality in Events Manager. The last three lead-gen campaigns spent $354 in nine days for zero purchases and 190 leads — that is what optimising toward the wrong event looks like. If they are set to leads, they need restructuring, not un-pausing.
- Consolidate to ONE campaign, ONE ad set. Do not geo-split. Nic · 20 min $7/day (£5) across three geo campaigns is barely $2.26/day (£1.67) each. Nothing exits learning at that size, the 30-day kill condition fires, and we wrongly conclude retargeting is dead. Pool T1+T2 (checkout abandoners + email-only) into one ad set — ~677 people, the only pool with real intent.
- Kill the lead-gen spend and fund retargeting from it. Nic, on Aga's approval · inside the envelope No new budget. Move the wasted lead-gen spend across. The two seats disagree on the number ($14/day (£10) vs $27/day (£20)–$34/day (£25)) — see the split below.
- Run two creative concepts. Not five. Fleet builds · Nic uploads At this budget more than two means neither gets enough impressions to say anything. The two are named below, and we already own both.
- Read Stripe at day 14, not Meta. Aga · 15 min, once One number: verified cost per Stripe purchase. Meta-reported ROAS is what made the lead-gen campaigns look alive while they sold nothing.
01What the account has actually done
Measured live from the Graph API: 302 distinct visuals regrouped from 485 spending ad rows, covering 99.1% of the $100,763 this account has ever spent. Only 92 of 302 visuals ever produced a single sale.
The account's lifetime ROAS is 0.42. That headline hides the only thing that matters here — the money and the results went to opposite places.
VIDEOQuiz — OLD Video 1
The most-funded creative in the account's history
Took 40.8% of all account spend and returned 14p on the pound. Four variants, three years.
VIDEONew Sales Ad
The best-returning creative in the account's history
A twelfth of the budget, almost the same number of sales, at an eighth of the cost each. Ran in remarketing.
02Retargeting is a different business
Split every visual by whether it ran in a remarketing/retargeting campaign. The gap is not subtle, and it holds across three years and $99,818.
| Lane | Visuals | Spend | Sales | ROAS | CPA |
|---|---|---|---|---|---|
| Retargeting / remarketing | 50 | $6,807 | 159 | 1.72 | $43 |
| Everything else | 252 | $93,011 | 522 | 0.32 | $178 |
Retargeting got 7% of the money and produced 23% of the sales. That is the case for the playbook's Meta test. It is not a case for expecting 1.72 again — read the next section before anyone forecasts on it.
But look at which offer each campaign carried
A visual reused across several campaigns has its spend and sales split evenly between them here, so treat these rows as indicative, not measured — the lane totals above are exact. The pattern survives the caveat: it is visible in every row.
| Retargeting campaign | Spend | Sales | ROAS | CPA |
|---|---|---|---|---|
| Retargeting | July 4th 2026 Sale | WW | API | $312 | 7 | 5.16 | $47 |
| Retargeting | July 4th 2026 Sale | NA-EU-AU-NZ | EXU | $280 | 5 | 4.49 | $56 |
| Retargeting | July 4th 2026 Sale | WW | EXU | $316 | 6 | 3.51 | $56 |
| New Users Remarketing | $3,851 | 117 | 1.66 | $33 |
| Retargeting | July 4th 2026 Sale | WW | LLU | Hot | $284 | 2 | 0.94 | $170 |
| Remarketing 50% Off | May 2024 | $745 | 18 | 0.87 | $41 |
| Retargeting | July 4th 2026 Sale | WW | LLU | Cold | $136 | 1 | 0.77 | $205 |
| Retargeting | July 4th 2026 Sale | WW | LLU | Warm | $289 | 1 | 0.36 | $433 |
| Retargeting | July 4th 2026 Sale | WW | QUP | $297 | 1 | 0.35 | $445 |
| Remarketing | New Users | $298 | 3 | 0.34 | $99 |
The finding that should change the plan
The promo-window winners (Memorial Day 2.47 · July 4th 2.38) sold a deep Lifetime discount that no longer exists. But the dominant winner is different, and earlier drafts of this page mis-stated it: New Users Remarketing — 120 sales at ROAS 1.58 — sold SUBSCRIPTIONS at ~$54 AOV, not Lifetime. There IS a subscription-era retargeting precedent. What remains true: the only explicit 50%-off retargeting run — the exact offer Rung 2 proposes — returned ROAS 0.87, the worst of the set.
Read it honestly, in both directions. That campaign ran May 2024, outside any promo window and without an email arc behind it, on 18 purchases — so it is confounded and small, not a verdict on 50% off. But it is the closest thing we have to the proposed offer, and nothing in the account supports forecasting 1.72 onto it. Those Lifetime offers no longer exist, so none of the winners is a like-for-like precedent.
03The creative that should work
Ranked by money, aggregated into concept families rather than single ad rows — the same concept is re-uploaded under near-identical names, so judging one row understates it.
VIDEOWhy You Started
Returning-member UGC — talking head, captions, app on screen
The most efficient concept in the account: ROAS {F['Why You Started']['roas']:.2f} at ${F['Why You Started']['cpa']:,.0f} CPA on only ${F['Why You Started']['spend']:,.0f}. Retargeting-native — it speaks to someone who already knows us.
VIDEOOld Man Doing Push-ups
The same footage as New Sales Ad, without the discount overlay
A man in his fifties doing a push-up with a small child on his back. This footage is behind both the #1 revenue ad and a top-ROAS 2026 ad.
IMAGEHandstand — 50% off static
A static that already carried exactly the Rung 2 offer
ROAS {F['Handstand Image']['roas']:.2f} at ${F['Handstand Image']['cpa']:,.0f} CPA. The only 50%-off creative in the account that worked — and it is an image, not a video.
IMAGEPDF Ad — the trap
Cheapest leads in the account, worst return in the account
1,324 leads at ${usd(1.04):,.2f} each and ROAS {F['PDF Ad']['roas']:.2f}. A cheap lead is not a buying signal. This is what the lead-gen campaigns are still doing.
Why the push-up footage is the pick
It is not a fitness ad. Nobody in it is impressive — a man in his fifties, on his living-room floor, with a child climbing on his back. It sells capability, not aesthetics, and it is Jesse's own origin story: being able to pick up his kids mattered more than any skill. It is also the only asset in 302 that won twice, under two different offers, in two different years. Everything else won once.
The contrast is PDF Ad — a shredded physique model, the cheapest leads we have ever bought, and effectively no revenue. The account has already run the A/B between aspiration and capability, and capability won.
Format: statics deserve a real slot
Classified on the asset actually served, not Meta's object_type — 53 of our
statics report as SHARE because they are page-post shares, which is why an earlier
read concluded no static had ever sold.
| Format | Visuals | Spend | Sales | CPA | ROAS |
|---|---|---|---|---|---|
| VIDEO | 247 | $87,286 | 495 | $176 | 0.34 |
| IMAGE | 53 | $12,421 | 186 | $67 | 0.97 |
Statics out-returned video in every single year (2023 2.45 v 0.19 · 2024 0.54 v 0.34 · 2025 0.93 v 0.40 · 2026 0.37 v 0.28) on a seventh of the budget. That makes them under-tested, not proven — they mostly ran inside promo windows where the offer deserves credit too. It is enough to earn one of the two slots, not enough to go statics-only.
04The offer conflict — a hard constraint
50off/ charges immediately and has no free trial. The checkout page says so
itself: "the discount replaces the free trial." So a single ad can never promise both 50% off
and a 7-day trial, and no Rung 2 asset may contain the words free, trial or
7 days.
There is a subtler version of the same problem, and it is the biggest risk in this plan: the people we are retargeting declined a free trial. Rung 2 offers them an immediate charge — a harder ask, not an easier one. Half price is a smaller barrier than full price, but "pay now" is a bigger barrier than "pay nothing for 7 days". That is a real possibility for why the May 2024 50%-off retargeting returned 0.87, and it is testable.
Frame the money, not the percentage: "$78.50 for the year" beats "50% off", because
the discount only means something against a price they remember. Link:
https://app.themovementathlete.com/payments/checkouts-v3/50off/12month/
05What the paid-ads seats say
Meta paid-acquisition specialist
"$7/day (£5) is not a test — it is a rounding error." Meta needs roughly 50 optimisation events per ad set per week to exit learning. At $7/day (£5) and a $34–$90 CPA this account produces under 1.5 purchases a week per ad set. It never exits learning, delivery stays erratic, and you cannot tell "retargeting doesn't work at the new offer" from "retargeting never got enough signal." The fragmentation kills this, not the offer.
Do not split the four tiers into four ad sets — none would ever exit learning. Pool T1+T2 (~677 people, the ones with real intent) into one ad set; hold T3+T4 back until T1+T2 produces sales. Run Rung 1 (trial) first even for T1, and rotate to Rung 2 only on non-converters after 14 days — a time-based rotation, not an audience split. Change one variable at a time.
Missing from the playbook entirely: a CAPI dedup and Event Match Quality check (if the browser pixel and server CAPI both fire, Meta sees 2× the events); a frequency cap (~430 people will hit 3–5× frequency inside a week and burn out); and an exclusion list drawn from BOTH Stripe estates — retargeting a lifetime member with a 50%-off subscription would be a brand disaster.
Growth CMO · allocation and sequencing
"That 1.72 is real data. It is also from a different business." Every campaign that produced it ran offers that no longer exist. Treat it as directional signal, never a projection.
Meta should not go last. The playbook orders it after the landing-page and Google work, which is right for cold traffic — the page is the gate. It is wrong for quiz retargeting: these people already saw the page. Run it concurrently with the conversion-demotion job. Cold Meta still goes last.
Do not take money from Google. Google is the only channel with a verified recent CPA (6 purchases on $194). You do not cannibalise the control to fund the test. Take it from the lead-gen campaigns that produced zero purchases in nine days.
On whether this is worth Aga's time: at roughly +$279/mo against a $7,172 gap, this is a margin lever, not a growth plan. It is defensible only because it does not compete for her attention — Nic executes, she reads one number once. If it starts needing briefs and creative reviews from her, the return goes negative. Retention is still where the gap actually lives.
Where the two seats disagree — Aga's call
The specialist says $27/day (£20)–$34/day (£25). Below that the ad set never escapes learning, so a small budget doesn't buy a cheap test — it buys an unreadable one.
The CMO says $14/day (£10), scale at day 14 only on a verified Stripe purchase. The one comparable offer returned 0.87; spending more before proof compounds a possible mistake.
The honest position: they are trading a measurement risk against a money risk. $14/day (£10) likely produces too little signal to conclude anything in 30 days; $34/day (£25) risks ~$1,013 on an offer with no supporting precedent. A defensible middle is $20/day (£15) for 30 days, pre-registered as a signal-generating test rather than a profit test — with the $89 cost-per-sale kill condition live from day one. What neither seat disputes: $7/day (£5) across three campaigns answers nothing, and the conversion-event check comes first regardless.
06The creative brief — two concepts
Concept A · video · returning-member UGC
- Format: vertical 9:16, 30–45s, captions burned in (sound-off is the default)
- Hook, retargeting-aware: acknowledge they have been here before. Our own winners prove the register — "You already know about TMA" (24 sales) and "You've seen what's possible" (15 sales) both sold.
- Body: a real member on camera, app briefly on screen. Reuse the Why You Started master if the original is still available — ROAS 4.00 at $32 CPA.
- End card carries the offer, and only one rung.
Concept B · static · capability, with the price legible
- Hero: the push-up-with-child frame, or a real member from the asset book. Capability, not physique.
- One line, money-framed: "$78.50 for the year" — never a bare "50% OFF".
- No trial language anywhere near a
50off/link. - 1080×1080 and 1080×1350. Precedent: Handstand Image, ROAS 3.65 at $16 CPA.
- Do not re-run the PDF Ad physique/lead-magnet pattern into a sales campaign — cheapest leads, near-zero revenue.
- Do not reuse any asset carrying $297, $247, 70% or 75% off, or a 4th-of-July lockup. Those offers are gone; the ad would promise what the checkout will not honour.
- Do not leave the "GET 50%" overlay on the push-up footage if it points at a trial link — the overlay implies a discount that link does not give.
07The creative pack — paste-ready
Named files and finished copy, so nothing here needs re-deciding. The assets are already in the repo at the paths below; the copy is written against the structure our own best-performing retargeting ads actually used.
The structure that won — copy it, don't reinvent it
Our two highest-ROAS retargeting-voiced ads ("You already know about TMA", 24 sales at 2.48 · "You've seen what's possible.", 9 sales at 3.14) run the same six beats. That is the asset, more than any single line:
- Name what they already didbeat 1 — recognition, not introduction "Maybe you took the free assessment. Read the success stories. Thought about whether it's right for you."
- "Here's what changed:"beat 2 — the pivot, three words The whole ad turns on this line. It gives a reason for the ad to exist today.
- The offer, against an anchor they rememberbeat 3 Never a bare percentage — the old winner said "normally $997. This week: $247."
- Proof as outcomes, not adjectivesbeat 4 "First muscle-ups after 6 months. Back pain gone. More lean muscle."
- Dissolve the last objection in a member's voicebeat 5 "I would've paid full price after seeing the results."
- One CTA, then the safety linebeat 6 Guarantee + member count. Never two competing asks.
The precedent nobody had found
The Short static — 22 sales, ROAS 2.50, $16-ish CPA — already carried "12 Months: $157 → $78.50" on the image, alongside the lifetime deal. Our exact Rung 2 offer, framed exactly the way this dossier recommends, has already sold. It ran inside Black Friday so the window shares the credit — but "$157 → $78.50" is not an untested framing, and that materially softens the 0.87 worry above.
The assets — already in the repo
| Use | File | Sales | ROAS |
|---|---|---|---|
| Concept A · video | marketing/assets/meta-ad-creatives/video_24238830839127139.jpg "You already know about TMA" — the 0:52 UGC master | 24 | 2.48 |
| Concept A · alt | marketing/assets/meta-ad-creatives/video_856889726702257.jpg "You've seen what's possible." — the 0:28 cut | 9 | 3.14 |
| Concept B · static | marketing/assets/meta-ad-creatives/image_a2c564434375806723fd257cf424697b.jpg "Short" — the ex-lifter pain hook, 1000×1000 | 22 | 2.50 |
| Hero footage | marketing/assets/meta-ad-creatives/video_896918766133384.jpg push-up with child on back — re-edit, strip the "GET 50%" overlay | 14 | 2.48 |
🔴 These are the thumbnail frames pulled from Meta. The editable masters live with whoever cut them — the Meta API serves frames, not source files. If a master cannot be found, the frame is still enough to re-shoot or re-cut against.
Copy · Rung 1 — trial, days 0–7 · NO discount language anywhere
Primary text
Maybe you took the assessment. Maybe you got as far as the price and closed the tab.
Here's what changed: nothing about the offer — it's still your first 7 days free. What changed is that your plan is already built and sitting there.
Not a generic beginner program. Your exact level on every movement, from an assessment you already did. Push, pull, core and mobility are all at different levels in every body, and it trains them where they actually are.
Members tell us the same thing: first pull-up in their forties. Back pain gone. Shoulders that stopped clicking.
7 days free. Cancel in two taps. 60-day money-back guarantee after that.
Headline: Your plan is already built
Description: Start your 7 days free — 100,000+ members
CTA button: Start free trial
Copy · Rung 2 — 50% off, days 17–30 · NO trial language anywhere
Primary text
You already know what this is. You took the assessment, you saw the plan, you didn't start.
Here's what changed: $157 a year is now $78.50. Half price, and it stays half price for as long as you keep training.
That's $6.54 a month for a program that rebuilds itself around you every session — 100+ skills with full progressions, from your first proper push-up to a handstand.
What members say after: first muscle-up in six months. Chronic back pain gone. Lean muscle that didn't cost them their joints.
One of them put it better than we can: "I'd have paid full price if I'd known it would work this well."
$78.50 for the year. 60-day money-back guarantee — if it isn't for you, you get it all back.
Headline A: $78.50 for the year
Headline B (run head-to-head): 6 months free — $78.50 today
Description: Half price · 60-day guarantee · 100,000+ members
CTA button: Get offer
Link: https://app.themovementathlete.com/payments/checkouts-v3/50off/12month/
🔴 On headline B: the playbook bans the word free in Rung 2 assets. That rule exists to stop free-trial promises against a checkout that charges immediately — Aga's framing gives away months, not a trial, and it is her own measured history (the monthly→annual emails sell the same arithmetic). The carve-out condition: "$78.50 today" must sit in the same breath as "6 months free", so the checkout never contradicts the ad. Never "free" standing alone.
- Rung 2 contains no free, trial, try or 7 days — deliberately.
50off/charges immediately, and trial copy against it is a promise the checkout breaks. - Rung 1 contains no discount — a discount there trains people to wait, and undercuts the welcome email that asks full price around day 10.
- The anchor is $157, not $997. The old ads anchored on a $997 lifetime that no longer exists; anchoring on a price they never saw is a fabrication.
- Never 4.9★. Published rating is 4.7★ or 4.8★.
08The strategy — the full fleet's verdict (13 Aug)
A /growth-fleet run: analyst, funnel/CRO lead, Meta specialist, Growth CMO, the Maverick challenger, and the Skeptic gate. Timeline-gated (the evidence window is DECOMPOSE-FIRST); two open decisions honoured by DEC-id. What survived adversarial review:
What we are optimising for
ONE event: a Stripe-verified purchase. Not leads (the account's cheapest leads returned
ROAS 0.04), not clicks, not Meta-reported conversions. The dominant real conversion is the
$97 quiz-window annual — and the analyst's LTV math says that is also the minimum
price at which recovery works: a $78.50 half-price recovery at historic CPA nets ≈ $0 in year one.
Paid traffic therefore lands on trial/full-price surfaces, never on 50off/.
(Scoped to PAID — the email ladder's day-17–30 rung 2 is a ~$0-cost channel and stands per
DEC-2026-08-10-remarketing-ladder; its 15 Sep review inherits this LTV evidence.)
Lead-gen stays — the two machines called "lead gen" are not the same thing (13 Aug)
An earlier draft said "kill the lead-gen — $354/9d for zero purchases." That conflated two different machines, and the verdict flips once they are separated:
- The QUIZ lead campaigns (QLG) stay. Their leads are quiz email captures on OUR site — they reach ActiveCampaign, at $1.74–2.21 CPL. "Zero purchases in nine days" was never zero return: until 10 Aug the email estate never asked those leads for money (the welcome arc was activation-only by doctrine). The first sales arc (993) armed 10 Aug — so the honest read of what a lead is worth lands ~10 Sep. Break-even at current CPL: roughly 2.2 of every 100 leads buying, ever. Judge then, not now. Trim the weakest variants (the $3–5 CPL tail) if the envelope needs room — never the lane.
- The in-Facebook LEAD-FORM campaigns stay OFF (paused 11 Aug, NIC-97) until the AC bridge is verified live. That bridge has been dead since 1 Dec 2025 — 5 of 1,784 leads ever reached the email system. A cheap lead that evaporates is not a cheap lead; it is a paid donation to Meta. Off is not a strategy verdict — it is "fix the pipe first".
The retargeting test funds from the paused lead-form spend plus envelope headroom — the whole Meta line stays inside the ~$3,175/mo cap.
The audience — who is worth paying for
| Tier | ~/mo | Verdict | Destination |
|---|---|---|---|
| T1 · opened checkout, didn't pay | 430 | FUND — pooled | /start-training/ |
| T2 · gave email, never saw price | 247 | FUND — pooled | /start-training/ |
| T3 · started quiz, no email | 740 | FUND — pooled (for volume) | quiz root |
| T4 · bounced before Q1 | 1,670 | EXCLUDE | — behaves like cold traffic |
One campaign, one ad set, built from pixel audiences (quiz + checkout URL visitors), not AC lists — the pool needs every real member it can get. Exclusions from both Stripe estates plus active trialists. Frequency-capped: this is a ~1,400/mo stock, not a flow, and it saturates in 2–4 weeks.
Why the non-buyer data changes the copy, not just the offer
5,252 non-buyer objections: program fit 32% · time 28% · technical 18% · price 15%. Only one objector in seven is a price objector — a discount answers the smallest objection on the board. The creative must answer fit ("your plan is already built, at your level") and time ("15 minutes counts") before it ever mentions money. That is what the 6-beat structure in §07 does, and it is why the belief-building lives inside a 60-second video ad rather than one click away on a blog.
The blog idea — right instinct, wrong tier, not yet
For T1/T2 (already past the content stage) a blog detour adds three steps before a checkout that converts 11.6% of openers. For T3/T4 it is genuinely plausible — but the one on-persona asset (the BJ ex-lifter story) is not verified live, and a content-vs-offer split test at our volumes needs 200+ clicks per arm ≈ 8–30 months to read. NO for now; publish the BJ article anyway (it costs nothing and unblocks the play), revisit after a 60-day baseline.
Pre-registered honesty — what this test can and cannot say
- At plausible rates the pooled audience produces ~9 purchase events/month — below Meta's learning floor. This is a signal test, not a profit test, and the most likely outcome is UNPROVEN. That word is pre-registered: fewer than ~5 events in 30 days = "unproven, retest or drop", never "retargeting failed" and never "it worked".
- Primary kill: cost per Stripe-verified sale > $89 at day 30 (day-14 read is directional only). The old "<3 purchases/30d" trigger sits inside Poisson noise and is demoted.
- The first sales-arc email read lands ~10 Sep — inside this test's window. Meta will view-through-claim email conversions. The clean read is the email non-opener segment; the paid-vs-email additionality question waits for the 10 Sep read.
- ROAS on this page is Meta-reported = MODELLED. The 1.72 lane split is CONFIRMED from the pull; its transfer to today's offer is UNVERIFIED — nothing here forecasts 1.72.
🃏 The challenger's counter-take — clearly attributed, not blended
The Maverick seat's dissent, preserved because parts of it are arithmetically right: the checkout that loses 88% of its openers is a ~18× bigger lever than this whole campaign (146 checkout-openers per 9 days convert 11.6%; doubling that ≈ +$4,900/mo vs this plan's ≤ +$279/mo), reads statistically in 6–10 weeks where the ad test may never read, and costs no media. Likewise churned payers — people who already paid us once — are warmer than any non-buyer tier, and retention is the company's own named constraint. The fleet's answer: do both — the ad test costs Aga ≈ zero attention precisely so the attention can go there. And the challenger's sharpest design note is adopted: without a holdout, every Meta-claimed sale in September is plausibly an email conversion wearing a Meta badge.
His down-sell question ships as an email experiment, not an ads one: the
$9.99/30-day intro (30trial/1month — the only live 30trial path) answers fit, time AND
price at once for people who declined a free trial. The email rail can read a three-cell offer test
in weeks at $0 CPM; ads never can at these volumes.
Who does what
- Nic — verify the purchase event + CAPI dedup + Event Match Quality 10 min · blocking · before anything is un-paused
- Nic — kill the three lead-gen campaigns; build ONE retargeting campaign, ONE ad set 30 min · pixel audiences T1+T2+T3 · exclude T4 + both-estate payers · frequency cap 3–4/wk
- Fleet — finish the two creatives from §07; Nic uploads the 0:52 UGC master + the capability static · 6-beat copy as written
- Fleet — publish the BJ ex-lifter article free · unblocks the content-retargeting play for later
- Aga — one read at day 14 (directional) and day 30 (decision) 15 min · cost per Stripe sale vs $89 · everything else is noise
Read these before quoting any number above
- ROAS is Meta-reported, not Stripe-traced — one action type (
purchase), never summed. Treat every ROAS here as MODELLED. - Per-concept n is small — single to low-double-digit purchases. This is a ranked set of hypotheses to test, not a proven playbook.
- Creative is not separable from its offer or its window. Nearly every winner ran inside a promo with an email arc behind it. The offer, the timing and the creative share the credit.
- The window is 2023-07-18 → 2026-08-13 (Meta's 37-month wall), covering 99.1% of lifetime spend — not literally all time.
- The winning push-up footage appears to carry a stock watermark, so it is likely licensed. Confirm rights before rebuilding a campaign on it.
- All money is shown in USD, converted from GBP at 1.3504 (live rate, Thu, 13 Aug 2026 00:02:31 +0000). The ad account bills in GBP — which is why daily budgets carry the GBP in brackets: that is the figure Nic types into Ads Manager. Three years of spend is converted at today's rate, so dollar amounts for older periods are indicative, not what was paid at the time. Ratios are unaffected — ROAS divides GBP by GBP.