Meta ads playbook for FiftyStrong

Lifetime: $1,780 spent since Aug 7, 6 sales, $79 collected. The purchase campaigns cost $152 per sale. Under the coupon paywall, a customer is worth about $25–42 in her first 12 weeks. So we need a cost per sale near $40 to break even, and near $28 to pay back 1.5 times.

Ours is the Oct 14–18 run unless the row says otherwise. "Lander" is a person who loads the quiz from an ad.

What each fix does to our cost per sale. Cost per sale = cost per click ÷ (lander-to-checkout rate × checkout-to-paid rate). The rows below are illustrative. They assume the screen-1 fix (from 38% to 50% answering question 1) lifts every later step by the same factor.

The paywall is worth the most, and it shipped today. CPM is the hardest gap. Our lead campaigns bought impressions at $24–30, so a $30–40 CPM is reachable once Meta has real purchases to learn from.

Our account against the playbook

Bidding: lowest cost (highest volume), no caps. Every small-budget voice agrees (@stacy_siz, @Peter_Quadrel, Marcus Burke). Add a cost cap only after about 50 purchases and a stable cost per sale. Marcus Burke starts caps at 4–5 times the historic cost and lowers them step by step.

Optimisation event: Purchase, but only when money moves, at the amount charged. The common advice for small budgets is to step down to a cheaper event such as Lead. Our own history says no. Lead goals spent $869 for 56 checkouts and 0 sales. Purchase goals spent $911 for 23 checkouts and 6 sales.

The coupon paywall charges on day one, so Purchase is now the honest event closest to revenue. If we see fewer than 5 purchases a week after two weeks, step down to InitiateCheckout. Never step down to Lead.

Fix the signal before we spend again:

1. Send Purchase only on a real charge, at the charged amount and currency. Today a $1.99 week start reports a $34.99 Purchase. One account cut its cost per trial by about 35% when it stopped sending trials as purchases.
2. Stop staging and localhost from firing the production pixel.
3. Turn on automatic advanced matching. Read event match quality in Events Manager, and aim for 6 or more.

Another session has uncommitted work on a delayed "qualified trial" Purchase. Merge with it. Do not build a second version.

Campaign and asset setup

• Keep one manual sales campaign. Neither X nor YouTube had app data that favours Advantage+ sales campaigns.
• Keep Advantage+ placements, but block Audience Network in the account-level placement controls. In small founders' tests it gave 84–89% of clicks, and most were junk.
• Give every asset a 4:5 version for feed and a 9:16 version for Reels and Stories. Use placement asset customisation.
• Keep all AI creative enhancements off. Before Oct 28, turn off "product browsing", which Meta switches on by default. Also turn off creative-feature tests and the AI landing-page test.

Placements: build for Facebook feed. Several operators say Facebook feed is the older placement, and that older people click more. Facebook feed took 48% of our purchase spend, at a CTR of 3.0%, against 5.1% on Facebook Reels. Every asset was 9:16, so feed cropped the hook text. Wall-of-text statics and long videos win there for fitness apps aimed at older people (Nathan Hudson, Marcus Burke).

Audience: broad. Target women 45–64, with no interests and no lookalikes. Let the creative call out the persona. Upload buyers as an exclusion.

Use five English-speaking markets: US, CA, GB, AU and IE. Local prices are live for GB, CA and AU, and IE pays in euros. Leave New Zealand, the other euro countries and the Nordics for a later, separate test.

Formats that win for women 45+, against what we already have

Hooks

• Use a root-cause reframe in her words: "I thought it was my age. It was my muscles." Take the words from the Reddit corpus and from reviews.
• Call out the persona in the first line, in the third person: "for women over 50 who haven't exercised in years".
• Our long cuts lose people early. Average watch is 4–8 seconds of 21–37. Put the promise and the quiz ask in the first 5 seconds.

Copy

• Primary text: write one per ad, to match that ad's hook. Today 54 of 62 ads share one text. Open with a line of 14 words or fewer that repeats the hook. Add 3–5 short lines in her words. End with "Take the 2-minute quiz."
• More text options: add 2–3 primary texts and 2–3 headlines to each ad. This is the cheapest copy test we have never run.
• Headline: keep it to 40 characters or fewer, and make it match quiz screen 1. Candidates: "Strength plan for women over 50", "12 weeks. At home. No equipment.", "Your 2-minute strength quiz".
• Description: keep "At home · No equipment · Built for women over 50". Few placements show it.
• CTA button: keep "Learn more" on cold ads. Sabri Suby and several agencies agree, and nobody posted a test that beat it. Do not spend tests on the button.
• Landing match: quiz screen 1 must repeat the ad's hook. Our biggest leak by volume is that screen, where 62–76% leave.

Meta sets budgets on the campaign (CBO) or on the ad set (ABO), never on one ad. To control one ad, pause it with a rule.

Phase 1 costs about $700. That fits the $741 left under the $1,000 spend cap.

Phase 1 ads: Beginner friendly, Chair, Floor rise and Stairs, with the old "Saturday 17 October" date removed, plus one new wall-of-text static. Use women 45–64, the five English-speaking markets, and Purchase as the event. Make no edits for 7 days. In this phase let Meta split the money, because the paywall is the question, not the creative.

Phase 2 structure

• Use ABO (ad-set budgets) while testing. Each ad set then gets its own money, so the test always spends. Inside one ad set, Meta gave 3 ads 59% of the October spend, and 7 ads got under $15 each.
• Adapt Marco's seed rule. Pausing each ad at 2,000–3,000 impressions filters out losers. It does not prove winners: "3k impressions tells you what's bad, never what's good" (@adriamatz). At our CPM, 2,500 impressions cost about $150 per ad. Use about 1,000 impressions (Sergio C's minimum) for the filter, then a read round on the survivors.
• Move a winner by post ID. It keeps its likes and comments in the Core ad set.

• Phase 1 read at 25 checkouts: 7 or more sales means go to phase 2. 3–6 sales means run on to 50 checkouts. 0–2 sales means stop and fix the paywall.
• Decide sales per account, not per ad. At $50–100 a day, no single ad gets enough sales to judge. Judge ads on the earlier steps, and judge the account on sales. Cal AI's COO: "$50/day on Meta tells you almost nothing" about cost per sale.
• Change things at most every 3–4 days. Never change audience, countries or budget in the middle of a read. Each edit restarts learning.
• Raise budget in small steps. Hold each step for days, not hours. Examples: +20% every 3 days (@nicktheriot_), €50 steps held 1–3 weeks (BPM dating app).

1. Mine the Ad Library every week. Check BetterMe (its chair yoga, wall Pilates and 40+ funnels), Reverse Health, JustFit, ChillFit, Lasta, Joi, and the menopause brands. Keep ads that are active for 60 days or more, or that run the same creative in many ads. Save them in .claude/competitor-teardown/.
2. Tag each one by persona, angle and format. Borrow the format. Write the message from her own words: the Reddit corpus, reviews and support mail. Copied messages do not scale, because bigger buyers outbid the same creative (Nathan Hudson).
3. Start from the angle data we already have. In 279 organic videos, chair or no equipment, minutes, and belly beat strength and identity 4 times per video. Belly is the door, and identity is the room.
4. Launch 3–4 new cells a week that differ in persona or format. Meta groups similar ads under one Entity ID. So a new hook on the same video counts as the same ad.
5. When one wins, iterate it. Make 80% of new ads variations of the winner. Use new hooks on the same body, 4:5 and 9:16 cuts, and persona word swaps. Make 20% new.
6. Track hit rate and cost per winner. Expect 5–8% of ads to win (Motion), so about 1 winner per 15–20 ads tested.

First four test cells for phase 2

Plan for the season. Dec 26 to mid-January has cheap CPMs and resolution demand, so it is the best window of the year for fitness apps. Get the account out of learning, with real purchases, before Dec 26. Have the creatives ready.

1. Change the server events: send Purchase only on a real charge, at the charged amount. Merge this with the in-progress qualified-trial work.
2. Stop staging and localhost from firing the production pixel.
3. Turn on automatic advanced matching in Events Manager. Read event match quality.
4. In Advertising settings, block Audience Network. Turn off product browsing, creative-feature tests and the AI landing-page test.
5. Make 4:5 versions of Beginner friendly, Chair, Floor rise and Stairs. Remove the old date. Make one new wall-of-text static.
6. Write one primary text per ad that matches its hook. Add two headline options per ad.
7. Build phase 1 as a new ad set in the Purchases campaign. Use women 45–64, the five markets, Purchase, lowest cost, $50 a day and 5 ads.
8. Run it for 14 days with no edits. Read the result at 25 checkouts.
9. During phase 1, make the four phase-2 test cells.

All research files are in deep-research/runs/meta-ads-playbook-20261021-113254/:

• our-account-audit.md: lifetime totals, settings, breakdowns by age, country and placement, the events we send, and unit economics.
• x-structure-bidding-budget.md: X on structure, bidding, budgets, targeting and kill rules (56 searches, about 40 thread pulls).
• x-creative-benchmarks.md: X on formats, hooks, volume, copy and benchmarks, checked against RevenueCat, FunnelFox and Motion reports.
• youtube-founders.md: your four videos plus 16 more, with timestamps. The strongest app sources are Marcus Burke, Fuelin, Built With Science and Hannah Parvaz on Sub Club.
• raw/: every API pull and transcript.

Most loud voices on X and YouTube are agencies and course sellers. App operators who post real numbers are fewer, and most spend under $10k a month. No source ran a fitness app for women over 50. We inferred the 45+ advice from BetterMe, menopause DTC brands and placement data.