A 33-SKU Ayurveda catalog with no published COA, a 19-day-old feed, 0% GTIN coverage and a 17-page Spanish presell funnel that had burned $754 over 90 days for a single conversion across 6,091 clicks.
The strategic frame every angle traces back to - the villain we name, the mechanism that beats it, and the proof no competitor can copy.
Losing presell angles burning $754/90d at zero ROAS - particularly a 17-page Spanish funnel
Monthly kill-list discipline + concentration on the 4 winning angle families
Gender-segmented Ayurveda (Her-Lajit, Alpha) - uncontested by every shilajit competitor except Better Alt
17-page Spanish presell funnel demoted from paid to SEO. Queries were TOFU / awareness ('que es shilajit', 'energia vitalidad mujer') with no transactional intent. Demotion freed $754 / 90d at the unit-economics level.
Six losing angles demoted, none patched - a shatavari cross-association angle (wrong product fit), a thin pure-shilajit advertorial (too short for cold traffic), a defensive 'safety for women' opener (wrong tonal posture for cold buyers), an over-narrow grandma-vitality narrative, and a 'best shilajit brand' editorial that leaked PDP-side.
Discipline over patching.
Budget reallocated to gender-segmentation (Her-Lajit, Alpha), couples-bundle positioning, kidney-support gummy (gummy-exclusive niche), and absorption-pairing copy - the four lines that earned their cost per acquisition.
Pre-audit: 19-day-old feed, 0% GTIN coverage on 81 products, no published COA, no 3rd-party testing claims. Post-audit: feed-side foundations restored for the Better-Alt-class consumer who buys on lab proof.
Cost-per-acquisition mathematically unsustainable. Funnel reassigned to SEO-only, paid budget reallocated.
Session-level analytics told the story - 151 page views, 35 CTA clicks, 2 conversions. The audience landed engaged but bounced on the PDP. Fix list shipped to the brand: guarantee surfacing, lab-proof block, tier-bundle math, mobile sticky CTA.
Every 90 days, paid-zero angles are demoted - the cull is the system, not a one-off.