osprex.ai · Custom Build campaign · paid funnel audit

Osprex Funnel Audit

Published Sep 14, 2026 · report #1

Meta data: Sep 1–12, 2026 (Ads Manager, reports through yesterday) CRM: GHL through Sep 14 Pixel 1572265074377416: Aug 17–Sep 13 VSL engine: since Sep 5

The bottleneck is upstream of the page: 74% of your landing-page views and 3 of your 4 "booked calls" came from Audience Network in-app placements for 14% of the spend. Those are game/app click-throughs and bot sign-ups (Colombian numbers, fake area codes, garbage names). Meta learned from them, so the Schedule optimization is now chasing more of the same.

The $200 cost-per-result cap is strangling delivery. Daily spend fell from $124 (Sep 5) to $22 (Sep 10) against a $200/day budget. With no real conversions to bid on, the cap tells Meta to stop buying.

Real traffic (Facebook + Instagram) is 216 landing-page views for $758, and it produced one bookable human who didn't show. That is 0.5% LPV→booking and ~$760 per booking. The page then loses 95% of visitors before they press play on the VSL.

The old Osprex Demo campaign (→ /rei, old pixel) is still running and spent more than the new one: $1,151 for 5 form leads, 0 shows. It is not part of this funnel but it is most of your September ad bill.

Campaign totals, Sep 1–12

Spend

$889

$200/day budget, ~$74/day actual

CPM

$34.90

FB $30 · IG $137 · AN $16

Link CTR

5.43%

1.61% without Audience Network

Landing page views

845

$1.05 each · $3.51 excl. AN

"Schedules" (Meta)

$222 each · 3 spam, 1 no-show

Real cost per booking

$889

1 human booked, 0 showed

Where the traffic actually came from

Platform breakdown of the one ad set ("Talking Head Ad Angles"). Placements are manual with Advantage+ placements off, and Audience Network is checked. Instagram is buying impressions at a $137 CPM, which means the creative has almost no engagement signal there.

Platform	Spend	Share	Impr.	CPM	Link clicks	CTR	CPC	LPV	$/LPV	Schedules	Read
Audience Network (in-app)	$127.93	14%	8,118	$15.76	1,103	13.59%	$0.12	626	$0.20	3	junk
Facebook (Reels $227, Feed/other ~$228)	$455.14	51%	14,962	$30.42	242	1.62%	$1.88	187	$2.43	0	real, weak
Instagram	$303.13	34%	2,209	$137.22	35	1.58%	$8.66	29	$10.45	1	overpriced
Threads / WhatsApp / uncategorized	$2.97	0%	189	—	4	—	—	3	—	0	noise
Total	$889.17	100%	25,478	$34.90	1,384	5.43%	$0.64	845	$1.05	4

Within Facebook: Reels took $227 for 26 clicks ($8.75 per LPV); "Ads on Facebook Reels" took $29 for 11 clicks at 0.11% CTR; Stories did 17% CTR on $11. The Instagram "Schedule" is Cris De Souza (cruise line, Sep 13, no answer).

Funnel, stage by stage

Red is the Audience Network share. Follow the blue: that is the funnel you are really paying for.

ImpressionsMeta

25,478

AN 32%

Link clicksMeta

1,384

5.4% CTR · AN 80%

Landing page viewsMeta · pixel PageView 2.3K since Aug 17

845

61% of clicks · AN 74%

VSL playspixel VSLPlay, all traffic

126

~5% of page views

Watched to 50%pixel VSL50

8% of plays

Booked (Meta Schedule)ad set, Sep 1–12

3 spam · 1 real

Showed

—

All paid trafficAudience Network share

Page views are not tracked server-side, so the play-rate denominator is the pixel's 2.3K PageViews (Aug 17–Sep 13, all sources incl. organic and bots). Even against paid LPVs only since Sep 5 (81), plays attributed to ads in the engine are 10 — a 12% play rate at best. Either way, roughly 9 in 10 people who reach the page never start the video.

VSL watch depth (vsl-0905, since Sep 5)

Milestone	Engine (deduped per browser)	Pixel events	% of starts (engine)
Play	30	126	100%
25% (~1:34)	14	—	47%
50% (~3:08)	7	10	23%
75% (~4:43)	4	6	13%
95% (~5:58)	2	2	7%

Half of the people who press play are gone before 1:34. Of the 30 engine-tracked plays only 10 carry an ad attribution (Angle 2 ×3, Angle 6 ×2, Angle 5 ×2, Angles 8/9/10 ×1) — the rest is organic and dialer traffic. Sample is tiny; the play-rate number is the one to act on, not the drop-off curve yet.

Daily spend: the cap is choking delivery

$22

9/1

$90

9/2

$31

9/3

$111

9/4

$124

9/5

$119

9/6

$99

9/7

$99

9/8

$46

9/9

$22

9/10

$48

9/11

$77

9/12

Budget is $200/day; the ad set never got close and has been decelerating since Sep 9. Three of the four "Schedules" landed on Sep 2 (the Audience Network spam burst), the fourth on Sep 4. Since then: zero conversions, and the algorithm has nothing under $200 to bid toward. Sep 1 and Sep 4 also show 3.4K and 10.1K impressions at 0.5% CTR — cheap Reels/AN inventory being sprayed.

By ad

Angle 2 got 65% of the budget because it "converted" — all four Schedules are on it, three of them fake. The CTR column is polluted by AN (Angle 6's 16% CTR is $0.10 clicks). Angle 3 is the tell: 7,096 impressions at a $2.78 CPM and 0.10% CTR is Meta dumping it into Ads-on-Reels overlays nobody clicks.

Ad	Spend	Impr.	CPM	Link clicks	CTR	CPC	LPV	$/LPV	Sched.	GHL leads
Angle 2	$575.05	10,322	$55.71	588	5.70%	$0.98	330	$1.74	4	6
Angle 1	$71.87	2,300	$31.25	99	4.30%	$0.73	62	$1.16	0	0
Angle 9	$61.08	860	$71.02	84	9.77%	$0.73	55	$1.11	0	0
Angle 6	$41.53	2,467	$16.83	398	16.13%	$0.10	240	$0.17	0	2
Angle 7	$38.47	848	$45.37	82	9.67%	$0.47	55	$0.70	0	0
Angle 5	$34.10	390	$87.44	18	4.62%	$1.89	14	$2.44	0	0
Angle 8	$29.50	707	$41.73	85	12.02%	$0.35	61	$0.48	0	0
Angle 3	$19.72	7,096	$2.78	7	0.10%	$2.82	8	$2.46	0	0
Angle 10	$11.47	247	$46.44	14	5.67%	$0.82	10	$1.15	0	0
Angle 4	$6.38	241	$26.47	9	3.73%	$0.71	10	$0.64	0	0
10 ads	$889.17	25,478	$34.90	1,384	5.43%	$0.64	845	$1.05	4	8

No ad has enough clean traffic to be judged on creative yet. Frequency is 1.1–1.3 across the board, so fatigue is not a factor.

Every lead this campaign produced (GHL, tag `custom build - osprex` / first-touch campaign)

Date	Contact	Ad	What happened	Verdict
Sep 2 12:52	Johanna · +57 312… (Colombia)	Angle 6	Form fill, no booking	spam
Sep 2 14:06	Petra Marmol · +1 311… (invalid area code)	Angle 6	Dialer no-show	spam
Sep 2 15:16	Juan José León Toro · +57 304…	Angle 2	Booked (Meta Schedule)	spam
Sep 2 15:40	Paola Andrea Téllez · +1 312…	Angle 2	Booked · F – Invalid Number	spam
Sep 2 15:43	María Luisa Arevalo · +1 313…	Angle 2	Form fill, dialer no-show	spam
Sep 2 15:49	"Alexxx" · +57 300…	Angle 2	Booked (Meta Schedule)	spam
Sep 4 15:48	Elvaince Menelas · Done Deal Funding	Angle 2	Booked · No answer · no-show	real, no-show
Sep 7 06:45	"Ngwykbvf Hwwhvh" · Dqiddi LLC · +1 521…	none	Form fill, dialer no-show	bot (passed phone check)
Sep 13 02:46	Cris De Souza · Victory Cruise Lines	Angle 2 (IG)	Booked · No answer	real, off-ICP, no-show

Six Spanish-named sign-ups inside three hours on Sep 2, four with Colombian numbers, all from Audience Network in-app — that is an incentivized-install app farm, not a targeting leak (the ad set is US-only; Colombia was later added as an exclusion, which does nothing against this). The US phone check shipped mid-day Sep 2 and stopped the +57 numbers, but 311/312/313/521 all pass a NANP regex, and "Ngwykbvf Hwwhvh" walked straight through on Sep 7.

The other campaign is still burning

Osprex Demo Campaign (account 1557381365880920 → osprex.ai/rei, old pixel, Lead-optimized) spent $1,150.77 Sep 1–12 for 5 "Website Leads" at $230 each: Angle 5 $444 (1 lead), Angle 9 $348 (1), Angle 3 $136 (2), Angle 8 $123 (1). The two that reached GHL with attribution (Joe Pinto, chris grant) are both dialer no-shows. Combined September spend across both accounts is ~$2,040 for zero demos held.

What to do, in order

Best · do today

Cut the junk supply before touching anything else

Placements: edit the ad set → manual placements → uncheck Audience Network entirely, and uncheck "Ads on Facebook Reels" (overlay). Keep Facebook Feed, Instagram Feed, Instagram Stories/Reels, Facebook Reels. Do not switch to Advantage+ placements — that re-enables AN.
Bid strategy: drop the $200 cost-per-result goal → Highest volume. A cap on a pixel with 1 real conversion only throttles; it cannot protect you.
Optimization event: keep Schedule only if you accept ~$100–150/day of exploration for 2–3 weeks. Cheaper path: optimize for a VSL50 custom conversion for one week — it is a quality signal humans produce and bots don't, it exists on the pixel already (10 events), and it steers delivery back to people who watch — then move to Schedule once you have 10+ clean bookings.
Reset the learning: duplicate the ad set rather than editing in place so the three fake Schedules stop shaping delivery.
Form hardening (engine + site): server-side phone validation in /app/lead (reject on the server, not just the modal), an area-code allow-list (311/312-ish sequential fakes and unassigned NPAs like 521 should fail), a hidden honeypot field, and Cloudflare Turnstile on step 1. Bot leads are currently fed to CAPI as conversions, which is how this loop started.

Second · this week

Fix the page's one real number: play rate

~95% of visitors never press play. Test: autoplay muted with captions and a "tap for sound" state; make the poster frame the first thing in the viewport on mobile (currently the pain-mirror hero copy sits above it); shorten the headline above the player to one line.
Half of starters leave before 1:34 — the first 60–90 seconds of the 6:17 cut need the payoff, not the setup. Cut a 2:30 version and split-test by slug (videoSlug already isolates stats per cut).
Add a "book without watching" CTA under the player — real leads are booking off the copy anyway (the one human booked on Sep 4 from Angle 2).
Instrument page views server-side (a pageview event next to vsl_progress) so play rate stops depending on a pixel count that includes bots and organic.
Decide on the old Osprex Demo campaign: pause it, or repoint its ads to the tracked root funnel. At $230 per unreachable form lead it is your biggest line item.

Avoid

Don't scale, don't rewrite creative, don't trust "Schedules" yet

Raising budget or launching new angles now buys more Audience Network sign-ups and teaches the pixel harder.
No ad has 100 clean clicks; ranking angles on CTR today ranks placements, not hooks. Re-read ads after one week of FB/IG-only delivery.
Stop reading Website Schedules / cost per result as the KPI until the form is hardened. Use GHL contacts with a dialer disposition as the source of truth.
Don't add Colombia-style geo exclusions to chase this; it's a placement problem, and the traffic is already US-only.

Data notes

Meta attribution window is 7-day click / 1-day view / 1-day engagement, statistical modeling on. Meta's date picker ends yesterday, so Sep 13 is not in the ad numbers.
Pixel event totals (PageView 2.3K, VSLPlay 126, Schedule 21, Lead 19, VSL50 10, VSL75 6, VSL95 2) cover Aug 17–Sep 13 and include organic traffic, dialer traffic, and pre-launch test events; Meta's 4 Schedules are the ad-attributed subset.
Engine VSL stats live in the container's /tmp and can reset on a container recreate; they may undercount plays before Sep 7's deploy. Per-contact watch depth in GHL ("VSL Watch %") is durable.
"GHL leads" per ad counts contacts by first-touch UTM Content; last-touch overwrites on re-click, so Angle 2's count may be slightly inflated by re-clickers.
Events Manager flags $633 of spend as "affected by low data quality" (match quality 6.1–7.7/10) — sending phone + email hashes on every browser event would help once the leads are real.