Last week the account spent more on its cheapest-lead ad than on its best-signup ad. S1_Night_search had the lowest cost per lead in the account at $2.67 and the second-worst cost per signup at $73.35 — and Meta’s budget optimiser kept feeding it, because it optimises to leads, which is not what we are buying. S1 is paused as of Sep 8; S2 has been off since Sep 3. Lead volume will fall next week on purpose. Two findings inside Novyra’s own product now sit ahead of anything we can do in the ad accounts: the matcher asks for date of birth and does not record it when given, burning a metered chat allowance; and roughly one in twenty-five Instant Form addresses is invalid. On Google, our earlier “renamed event” diagnosis was wrong — tracking is working correctly. Google’s problem is $760 per signup and a policy limit on cancer terms.
Velocity this week: Meta Sep 1–7 ran $2,355.88 on 735 leads — $3.21 cost per lead, against $3.38 on 722 leads the prior week. Attributed signups in the cohort join: 39, a blended $60.41 per signup. By ad: C_dark $34.58, the V3 videos $42.86 and $43.71 on small numbers, PreHealth $64.39, S1 $73.35, S2 $333.81. Budget moved the wrong way before we caught it — S1 took $513.44 (up $361.49 week on week) while C_dark, the best performer, took $449.59 (down $167.29) and PreHealth fell $636.73. Audience Network stayed at $0 and zero leads on every ad for a third week; delivery remains Facebook-majority. CRM recorded 71 real signups dated Sep 1–7 after removing 9 test addresses, 52 of them ad-attributed. Google spent $760.51 for one conversion.
Across all sources, about two thirds of people who start a match never reach a search. From Meta it is roughly three in four. From Google, closer to nine in ten.
We think we know why, and it is not targeting. On a clean first session, using Novyra’s own suggested prompt, the matcher asks for the patient’s date of birth, does not record it when supplied, and asks again. It happened three times in a row in one session — date supplied with gender, then alone, then in plain ISO format. Gender was captured first time; date of birth never was. A chat allows 10 messages and a visitor gets 3 free chats, so this burns roughly a third of the first chat’s allowance before the profile has a single usable field.
Three independent checks now point at the same place: a manual clean-session reproduction, a second reproduction run separately, and the aggregate drop above — which sits exactly where the date-of-birth step sits in the flow. This is downstream of every click we buy, which makes it the highest-leverage fix available: it costs no media budget and it is ahead of anything the ad accounts can do. It is with FiveRivers to confirm and fix.
On the funnel rates: these come from web analytics, where consent declines and ad blockers mean events go uncounted. Read them as directional floors, not exact counts — the real drop-off is at least this steep, not less.
The order of leverage changed this week. Fixing the date-of-birth loop is now first, because it is free and it sits in front of every click we buy. Then: keep budget on the ads with the cheapest signups rather than the cheapest leads, which is what the S1 pause enforces; give Meta a real signup event to optimise against instead of leads; clean the Instant Form address quality that is producing a 3.9% hard-bounce rate; and hold Google flat while the policy and cost questions are answered separately.