The target, proposed.
Same rules as the Ledisa plan, sized to Cleantra: YouTube only, one calendar month, held at the NC ROAS bar and read in our own data.
≈ $6.5K/day for 31 days on Cleantra YouTube. That is +35% on Cleantra's $571K of monthly paid spend, a big step, but under the ~$700K a month AppLovin video alone ran at over the last 90 days (NC ROAS 0.91). January is the first full month after the test gates and the December ramp, and it keeps BFCM week out of the test read.
The winners exist. The channel is missing.
Picked from Triple Whale and the company's Creative Report, not a guess.
AppLovin video on Cleantra Lymphatic in 90 days, at NC ROAS 0.91 (Triple Whale, TA 7d). 14 videos carried $1.52M of it at ≥ 0.9.
AppLovin in the last 30 days, at 0.78: under half its 90-day monthly pace. The same winners need new inventory. YouTube is the closest thing to AppLovin that Google sells.
$200K on top of $571K of monthly paid spend (Triple Whale, last 30 days). AppLovin video alone ran at ~$700K a month over the last 90 days, so the store has absorbed cold video at this size before.
Cleantra has tried Demand Gen. It fell short.
$14K at 0.47 through August, and the native run of the last 30 days is at 0.06. The setup failed in three known ways, and each one has a fix.
View as table
Bid on purchases
The Sep–Oct Demand Gen run optimized to "advertorial click + purchase". Google found people who click: 239 clicks to the offer, 1 purchase.
Proven creative
The summer run used Meta videos. Meta is Cleantra's weakest cold channel: 0.07 on $18K in the last 30 days. AppLovin is the strongest.
The right door
Demand Gen traffic went to the Search top-5 advertorial: $2.9K, 4 new customers. AppLovin sends 70% of its Cleantra visits through a short quiz on the brand domain.
Their idea, our structure.
Since Oct 23, Dominuss has run Ledisa video with each video as four ads, one per page. We take the idea of testing the door. We keep the structure we set for Ledisa, and the split follows Gian's Oct 21 rule for lander tests.
Google decides how much each door gets
With one ad per page, the split is Google's choice, not an even test: the top door got 1.6× the bottom one in week one. Video only, Oct 20–26: $4.0K, 4 new customers, NC ROAS ~0.05. Too early to call any door.
Google Ads 7566607070 + 5541627821, DOM video campaigns · NC = Triple Whale TA 7d by ad_id · 15 of their 61 video ads disapproved, 16 more "approved (limited)"
- Test the door, not only the offer page. Round 2: the quiz joins once the router page exists; the product page after that.
- Graduation. A video that holds NC ≥ 0.9 for 7 days moves into its own campaign.
- Purchase objective, no tCPA at start. Same as ours.
- One ad per door. Replaced by an even split inside one Voluum campaign, read on revenue per visitor. One ad per video also keeps Triple Whale's per-video read.
- Excluding men and 18–34. On Cleantra, men are 34% of non-brand Search purchases at the same ROAS (0.94 vs 0.94).
- Shorts and in-stream in one campaign. We keep them apart, so one format can't starve the other.
- Kept from the Ledisa plan: one campaign per format, YouTube only, purchase bidding, kill rules, gates, native 16:9, Search Brand ready before launch. Remarketing and the geo holdout come with the scale.
One door now, the split next.
Each video runs as one ad. Its link goes through a new Voluum campaign (direct tracking) to the T-Nutra offer page. Direct tracking can't split visitors between two pages, and it has no click where California and New York could be redirected, so both states are excluded in Google instead. The 50/50 with the quiz needs a router page on get.trycleantra.com (Netz), built separately and with its own OK. Four doors were cut to two on Oct 29, then to one for launch the same day.
Offer page (1/3/6)
The old default. When Google sent equal traffic to both offers, it lost, and since Oct 22 the main US Search path runs 100% T-Nutra.
T-Nutra free-gifts offer
Buy 1, Get 1 Free plus free gifts, the structure Black Friday keeps. What Google Search runs today, so it is the fair control.
Product page, 3 bundles
One-time purchase, not a subscription, so a win would mix page and offer model. Its 3-bundle price has a bug to fix first. Comes back as a one-time vs subscription test.
Short quiz → offer
The door the AppLovin winners were trained on: 276K of 396K Cleantra AppLovin visits in 30 days. Joins at 50/50 once the router page is live.
How we read it
The advertorial door
Dominuss's fourth door. On Cleantra it waits, for three reasons:
- → It lives on a third-party domain, so it can't share the split without a destination mismatch. It would need its own ad, which brings back Google picking the door.
- → Cleantra's Demand Gen to the top-5 advertorial: $2.9K, 4 new customers, 0.06 in the last 30 days.
- → A top-5 comparison after a Cleantra-branded video sends a sold viewer back to shopping.
It comes in on Dec 1 as a single-product article, only if door D (quiz) beats the direct doors. That would show that content before the offer works on video.
One product: two campaigns to start, never mixed.
In-stream is split from Shorts, so each format gets its own budget and learning: fewer videos, separate campaigns. Account: Cleantra - DGEN (970-617-2184). Both built paused on Oct 30. Remarketing joins with the scale. Search and PMax are not touched.
All $2.10M of Cleantra's AppLovin video is this product. Oregano, ParaVanish, Astaxanthin and Soursop have no video history and stay on Search and PMax. US minus California and New York, English. One video per ad, so Triple Whale reads NC ROAS per video.
Prospecting · Shorts
- Campaign ID
- 24410015997 · $100/day
- Places
- Shorts only
- Videos
- #1909 FluidPools and #10897 SockMarks, 9:16, one ad each
- Ad group
- One: "Shorts - Broad 25+". Interest ad groups rarely serve for health products, so Google's optimized targeting finds the audience
- Audience
- Broad, all genders, 25+, optimized targeting ON
- Bid
- Max conversions (Purchase) → tROAS at 50
Prospecting · In-Stream
- Campaign ID
- 24415194152 · $100/day
- Places
- In-stream + in-feed
- Videos
- #4590 BackedUp and #488 Sludge, two-column 16:9, one ad each. Native AI reframes come next and run as their own ads, so we see whether native pays for its credits. Never a pillarboxed vertical
- Ad group
- One: "In-Stream + In-Feed - Broad 25+"
- Bid
- Max conversions (Purchase) → tROAS at 50
Remarketing · later
- Lists
- Visitors & checkout starts 30d, buyers excluded. They live in the Cleantra Search account and get shared through the MCC
- Starts
- With the scale, not at launch
- Door
- T-Nutra offer page, then the winning door
- Budget
- ≤ 10%. Health policy can block lists. Ledisa's offer-page RMK did serve ($316, Oct 20–26)
How it reaches $6.5K/day: more campaigns, not bigger steps
A video that holds ≥ 0.9 for 7 days graduates into its own campaign, on the winning door, at ~$700–900/day. January runs on about eight campaigns plus RMK, and that needs the weekly pipeline of new videos (see Creative and the benchmark). Long-form videos run in their own ad group so they are read apart. Each budget moves +20% per step at most, and the bar never moves.
We start from what already sells, then add volume.
Six of the 14 AppLovin winners (Creative Report, Jul 28 – Oct 26, Triple Whale 7-day pixel). We read each transcript against Google's rules. Every card shows what we keep and what has to change before Google. The benchmark adds a weekly pipeline of new videos on top.
What the benchmark changes
Read the benchmark →Resilia, Lymphoria and Serene Herbs launched 1,049 new videos in October (run as 2,145 ads) and let Google find the few that sell: about 15 videos carry half of each brand's views. Our 14 winners are a better start than theirs, but they are not enough volume.
We edit them ourselves, with AI
We don't send small edit requests to the creative team: the fixes, the native 16:9 versions and the new variants are made on our side, scripted. Video runs in the company's Higgsfield workspace (CLI, plan credits); voice, music and lip-sync in ElevenLabs (company account). Both connected Oct 28. Cuts run free on our Mac. Already done on Oct 27: cut-only fixes of #488, #434, #1909 and #4590, each verified by re-transcribing (the burned-in captions are cut along with the audio). #1909 and #4590 still need a new hook.
Measured on Oct 28 with Higgsfield's free cost estimate: a 16:9 reframe of a 30 s video ≈ 363 credits; a new 5 s shot 7.5–35 credits depending on the model (Kling 3.0 Turbo → Seedance 2.5). The pilot batch (week of Nov 2: the fixed winners, two new hooks, two long-form stories) logs credits and hours per kind. Our own cap is set: Higgsfield 2K credits in October and 10K in November, ElevenLabs 5K and 75K, checked every Monday. Rules: no voice or lip-sync changes on a real actor without a release, our own credit cap (the shared pool has none, and it fell ~3,000 credits in a few hours on Oct 28 without us), files stay off GitHub.
The 14 winners: AppLovin originals
Cleantra Lymphatic videos with ≥ $30K on AppLovin and NC ROAS ≥ 0.9, Jul 28 – Oct 26 (Triple Whale 7-day pixel, Creative Report). Original 9:16 files hosted by AppLovin. The status column is the Google read.
| Concept | Ad ID | Format | Spend | NC ROAS | File |
|---|
Know by Dec 1. Run it in January.
Each phase has a gate. Money only scales on Triple Whale and Voluum numbers, never on Google's conversion column.
The ramp, day by day.
Spend follows the gates. The NC revenue line is the floor case (0.9); at 1.0 it sits on the spend line.
Max exposure before the scale decision: ~$5.6K (2 campaigns × $100 × 28 days), less if a kill rule fires. The smaller test makes the December ramp steeper: ×32 in 20 days. A single budget can't get there at +20% per step, so the ramp has to come from graduated campaigns and new videos, and only starts if Dec 1 says ≥ 0.85. If the pipeline can't feed it, January lands below $200K rather than below the bar.
YouTube creates the search. Brand must catch it.
Cleantra's brand Search campaigns show on about half of brand searches. The Lymphatic brand campaign, the product the videos sell, shows on one in five. We fix that before launch.
So the lever is keyword bids, not budget. Seven brand Search campaigns, last 30 days: 14.1K impressions, $4.4K.
Viewers hear the name. They don't read it.
Our speech-to-text heard the winners' voiceovers as "Clentra", "Klentra", "Clientra" and "Clienture". People already search the misspellings: 414 impressions in 30 days, 3.7% of brand search terms. Video will multiply them.
impressions, Sep 27 – Oct 26 · Cleantra - Search 257-883-8909, brand (ONB) campaigns
Three reads. Our data only.
The YouTube tab in Tepui, built for Ledisa, gets a Cleantra view. The door table comes with the second door; the geo holdout, with the scale (at $200/day it can't show anything).
Tepui · YouTube
Month-to-target, pacing, TA7 / TA14 / LC7 per campaign and per video (one video per ad, Triple Whale carries the Google ad_id), policy status per video, kill-rule flags.
Door table
Visits, share of split, click-out, CVR and revenue per visitor for each door, from the Voluum campaign. Gates on day 14 and day 28 read this table.
Is it new, or AppLovin's?
If YouTube only takes credit from AppLovin or Meta, the store's total new customers don't move. We read them in YouTube states against the 20% holdout states.
Where things stand, Nov 2.
What is ready, what needs a yes, and what comes next.