8fe1f3-8d · Meta act_1169431637682740 · FOCL and Palisade excludedBoth kill criteria are met. Across the 12 mature cohorts with credible attribution (Feb 2025 – Jan 2026, n = 18,831 customers), 6-month gross LTV rose $177.39 → $230.90 (+30.2%), OLS slope +$4.66/month, p < 0.0001, Spearman ρ = +0.92. 6-month margin LTV:CAC rose 0.91 → 1.18, peaking at 1.37 in Dec 2025. A Welch test on the first three versus last three cohorts gives +$40.79, t = 9.15, p < 1e-15. Nothing declined. The prior strategy's cohorts improved, materially and monotonically.
The uncomfortable half: the 3:1 target has never been met in any cohort, ever. The best month reached 1.37. And the risk H1 was reaching for is real — it just sits entirely in the censored window, where CAC has climbed from $103.79 (Dec 2025) to $255.00 (Aug 1–20) with no measurable LTV yet.
Supported for destination lander (26.2% spread in margin-LTV:CPP, same ordering in two independent windows) and, more weakly, for video vs static. Refuted for talent: founder content, creator whitelisting and external creators land at 0.82 / 0.82 / 0.83 once destination lander is held constant — inside overlapping confidence intervals. The thing the creative debate is actually about turns out not to matter to unit economics.
There is no decline to explain. And mix does not explain the rise either: shift-share attributes 105.3% of the LTV gain to within-segment improvement and −5.3% to offer mix. Welcome-offer share went from 12.3% to 79.1% of first orders — a shift toward the weaker segment. Offer architecture was a headwind that within-segment gains overcame.
Censored cohorts carry no 6-month LTV bar by design — blank, never zero. CAC uses Meta Ads API spend divided by Meta-attributed new customers. Meta CPP is shown separately and is never presented as CAC.
Why CAC exceeds CPP in every month: Meta's attributed conversions include returning purchasers that the CAC denominator excludes, the 7-day-click / 1-day-view window credits purchases the Shopify UTM join does not see, and Meta counts purchases where CAC counts distinct new customers. Whole-account spend also puts retargeting budget in the CAC numerator.
utm_source coverage, so those four cohorts' CAC ($271–$288) is an artifact of under-attribution rather than a real cost. They are excluded from every trend test. Including them would have made the "decline" look even more refuted, not less.Offer-band segmentation: mix −$2.82, within-segment +$56.33. Subscription segmentation: mix +$10.16 (19%), within +$43.35 (81%).
First-product mix held at 92–98% Longevity in every mature cohort, so product mix cannot explain anything. NAD+ Complete does not appear until Jun 2026, in the censored window.
Meta-acquired customer quality did not deteriorate — it improved by 30% on a 6-month basis, and the improvement is real rather than compositional: 105% of it is within-segment, offer mix contributed nothing, and first-product mix did not move at all. Full-price, welcome-offer, subscription and one-time segments all rose independently and significantly. What has deteriorated is cost, and only since February 2026: CAC has gone from $104 to $255, of which roughly two-thirds traces to landing-page conversion falling from 5.26% to 3.23% and the rest to media price and a halving of click-through rate. This is an acquisition-efficiency problem in the post-click experience, not an acquisition-quality or offer-architecture problem.
CPP = (CPM/1000) ÷ [CTR × landing-page-views-per-click × purchases-per-landing-page-view]. Log contributions; positive means the factor pushed CPP up.
Clean window (Feb–Sep 2025 cohorts), where warehouse spend ties exactly to the Meta API so no rescaling is applied. Bars are 6-month LTV; the line is Meta CPP; labels give margin-LTV:CPP.
Your account deep dive found the destination lander explains roughly 8× more cost-per-purchase variance than the hook. That holds: across landers CPP spans 38.3% while creative families span far less. But on a unit-economics basis the effect is much smaller, because the cheap landers acquire measurably worse customers — Pearson r(lander CPP, lander 6-month LTV) = +0.86. LTV6 spread is only 16.3%, so margin-LTV:CPP spread is 26.2%, not 127%.
The reallocation is still the right call — SH Solo delivers 1.48 against SH+SD Duo's 1.18 — but the prize is about a quarter, not a doubling. And it points at Peyton's A/B programme rather than net-new creative production, exactly as the prior read suggested, for a partly different reason.
Adjusting lander LTV6 for format × talent by direct standardisation moves the spread from 18.1% to 15.9% — the effect survives. Running it the other way, format spread goes 6.6% → 9.7% and talent 14.9% → 9.8% with overlapping intervals. Lander is the covariate that matters; talent is the one that dissolves.
Fitted on 13 cohorts with realised 6-month LTV. Baseline intercept-only LOOCV RMSE = $18.10; a signal must beat that to be kept. All errors are leave-one-out cross-validated, not in-sample.
Best model: first-order AOV + subscription take-rate, R² = 0.916, LOOCV RMSE $6.58. Both are observable on day 0. Notably, day-30 repeat rate does not predict 6-month LTV at all (r = +0.21, p = 0.49) and Skio cycle-2 retention is negatively correlated and fails cross-validation. Both are discarded despite being the intuitive choices.
Ensemble projection: 6-month LTV $209.53 (range $205.76–$211.87), margin LTV $136.19. Against Aug 1–20 CAC of $255.00 that implies margin LTV:CAC of 0.53 — the worst in the series. The projected weakness is entirely on the cost side; the LTV signals are broadly healthy, with subscription take-rate actually up.
Four caveats, prominently. (1) n = 152 over 11 days; one atypical week moves this materially. (2) It is a projection from a relationship fitted on 13 cohorts, not a measurement — the real number is not knowable until 8 Feb 2027. (3) The model is blind to the biggest change in the cohort: NAD+ Complete is 42.8% of post-10-Aug first orders versus 15.5% before and ~0% in every fitted cohort, and Longevity share (57.9%) sits far outside the fitted range of 92.2–98.0%. NAD+ Complete has no measured 6-month retention at all. (4) Aug 1–20 CAC includes the 12 Aug campaign rebuild; Aug 12–20 CPP was $179.24 against $195.92 for Aug 1–11, so the rebuild has cut CPP 8.5% — but landing-page conversion went the other way, 3.34% → 3.12%. The CPP gain came from cheaper traffic, not better conversion.
Definitions, metrics and thresholds fixed now so the February 2027 comparison is apples-to-apples. Control is the Sep 2025 – Feb 2026 mature cohorts: weighted 6-month LTV $215.71, CAC $124.28, margin LTV:CAC 1.13.
H1 would be reinstated if Feb 2026 completes below $195 6-month LTV on 29 Aug 2026; or Mar/Apr 2026 mature below $195 in Sep/Oct 2026; or a corrected spend series raises mature-window CAC enough to flatten the ratio. H1 stays refuted if Feb–Apr 2026 mature at or above $210, which their already-recorded 3-month LTVs ($185 / $190 / $174) are consistent with.
H2-for-talent would be reinstated if founder and whitelisting margin-LTV:CPP separate by more than 20% with non-overlapping intervals in a window where the lander is randomised or fixed by design — the current data cannot do this because lander assignment correlates with creative type. H2-for-lander would be refuted if a deliberate A/B routing the same hooks to SH Solo and to a Duo lander produced margin-LTV:CPP within 10%. That test does not exist yet, and it is the single highest-value experiment this analysis implies.
utm_content coverage is 99.4–100% every month, so the tactic join itself is sound.focl-dev.myshopify.com are all connected. TikTok connectors are no longer listed.99:Theme. In the other variant field 10 is Partnership and LP is field 11.acquisition_channel='facebook-ads' returns 9,663 customers while 'instagram' returns 15,703 — and those are utm_source=ig, utm_medium=cpc, i.e. Meta paid on the Instagram placement. Keying on 'facebook-ads' undercounts Meta by ~62%.