03Case study · AMA Holdings · The Loaded Teashop
The growth desk behind US$44M of creator ads.
AMA Holdings is the US owner of The Loaded Teashop and Goli Nutrition. The Loaded Teashop sells through a creator ad program: hundreds of creators uploading to briefs, media buyers scaling the winners on Meta and TikTok. Since October 2026 I run its growth and automation.
The scale, with honest labels
Every figure is from the program's own systems (creator platform, Meta, Triple Whale, Northbeam) as of 2 November 2026. Team and client scale: media buyers spend, creators film, I build the loop between them.
Weekly ad spend, 13 weeks: US$30.60M
Jul 22 to Oct 20, 2026 · US$ · Meta + TikTok- Jul 22US$1.79M1.69×
- Jul 29US$1.73M1.79×
- Aug 5US$1.42M1.83×
- Aug 12US$1.92M1.75×
- Aug 19US$1.03M1.73×
- Aug 26US$2.00M1.38×
- Sep 2US$1.72M1.40×
- Sep 9US$2.50M1.44×
- Sep 16US$3.29M1.39×
- Sep 23US$3.77M1.42×
- Sep 30US$3.30M1.48×
- Oct 7US$3.04M1.68×
- Oct 14US$3.10M1.18×
See the 13 weeks as a table
| Week (2026) | Ad spend | Blended ROAS |
|---|---|---|
| Jul 22 to Jul 28 | US$1.79M | 1.69× |
| Jul 29 to Aug 4 | US$1.73M | 1.79× |
| Aug 5 to Aug 11 | US$1.42M | 1.83× |
| Aug 12 to Aug 18 | US$1.92M | 1.75× |
| Aug 19 to Aug 25 | US$1.03M | 1.73× |
| Aug 26 to Sep 1 | US$2.00M | 1.38× |
| Sep 2 to Sep 8 | US$1.72M | 1.40× |
| Sep 9 to Sep 15 | US$2.50M | 1.44× |
| Sep 16 to Sep 22 | US$3.29M | 1.39× |
| Sep 23 to Sep 29 | US$3.77M | 1.42× |
| Sep 30 to Oct 6 | US$3.30M | 1.48× |
| Oct 7 to Oct 13 | US$3.04M | 1.68× |
| Oct 14 to Oct 20 | US$3.10M | 1.18× |
The loop
From upload to live ad, and the numbers back.
Creators upload. The team approves. Media buyers launch. The results decide the next brief. Before, each hand-off was a person copying files and rows. Now it looks like this.
- Creators uploadTo briefs, on the creator platform
- QA queueWithin a minute
- Drive and creative sheetEvery 15 minutes
- Media buyers launchMeta and TikTok
- Ad feeds read the numbersEvery 15 minutes: Meta, Triple Whale, Northbeam
- Dashboard, bot, weekly reportThe whole team, the same numbers
- Next briefs back to the creatorsWhat to make next
-
Every minute
New uploads land in the QA queue with flavor, funnel stage and whitelist handles filled in, and anything missing flagged on the card.
-
Every 15 minutes
Approved videos go to Drive and the creative sheet, and the ad feed refreshes spend, ROAS and pass or fail for every video.
-
Every hour, every day, every week
Creators sync, upload history and the submissions log run on the hour. Reports post to Telegram three times a day. The top 50 gets read every week.
What I run
Three jobs, one person.
Growth specialist is the title. In practice it splits into reading the data, running the creators and building the systems, and the point is that the same person does all three.
Read what wins, every week
The 50 videos with the most spend among the week's uploads, each downloaded, transcribed and sampled frame by frame, then sorted by angle, hook, buildup, body, CTA and format.
- Median thumbstop, ROAS and CPA per family of winners, so a brief can ask for the family, not a guess.
- Audits of creator videos against those findings, written so a creator can act on them.
- One rule for a winner, applied to every video by the dashboard: ROAS 1.4× and CPA $26 or better on $100+ of spend.
- Brief, product and funnel performance tables the team can quote in the Monday call.
833 creators, one database
Handles, whitelist status, tier and drop lists in one table the whole team edits, instead of five spreadsheets that disagreed.
- Hourly sync from the creator platform fills missing handles and marks who is set up. Team edits are never overwritten.
- A video whose creator is not whitelisted is held with the reason on the card, so nothing ships that can't run.
- Creator Scout: a chat command that searches TikTok and Instagram, scores each creator against the ideal profile and returns a shortlist.
- Roster reads: who is uploading, who went quiet, who got samples and never posted.
20+ workflows that report themselves
n8n on top of Supabase, Vercel and the Claude API. Every scheduled job writes a run line, and a failure turns a card red on the dashboard instead of going quiet.
- The video pipeline: upload to media buyers in 15 minutes, with QA, whitelist checks and daily reports.
- Ad feeds: spend and ROAS from Meta, Triple Whale and Northbeam, refreshed every 15 minutes.
- A knowledge bot on web and Telegram that answers from 900+ team notes and cites them.
- An AI editing app that turns creators' phone clips into finished 9:16 ads.
Systems
The systems, one by one.
Eleven of them, all in production for The Loaded Teashop team at AMA Holdings. Names are plain on purpose: the team calls them what they do.
-
Video pipeline
Every minute · every 15 minutesNew creator uploads land in a QA queue within a minute, with flavor, funnel stage and whitelist handles filled in. Every 15 minutes the approved videos are downloaded, filed in Drive by day, flavor, creator and funnel stage, and written to the media buyers' creative sheet with the right ad copy. "Send now" skips the wait. A failure writes its own row with the reason.
- n8n
- Google Drive
- Google Sheets
- Supabase
-
QA tab
On demandThe team reviews each video in the dashboard: approve, revision, hold or save for later, and fix a wrong flavor or funnel stage. A fixed funnel stage writes back to the creator platform. A video whose creator has no whitelist set up is held, with the reason on the card.
- Vercel
- Supabase
- n8n
-
Ad feeds
Every 15 minutesSpend, blended ROAS, Triple Whale and Northbeam ROAS, CPA and a pass or fail against the team's targets for every video, read from the creator platform into the Ads tab, with time frames from today to all time and a campaign table per flavor drop.
- n8n
- Supabase
- Triple Whale
- Northbeam
-
Creator database and sync
Hourly833 creators with handles, whitelist status, tier and drop lists. Every hour the creator platform's Meta assets fill in missing handles and mark who is set up; the pipeline reads the same table, so a creator marked "not set up" holds their videos until it is fixed.
- Supabase
- n8n
-
Submissions log and daily reports
Every 15 minutes · 9 AM, 3 PM, 9 PM EasternEvery video that reached a creative sheet in the last 30 days, who sent it, its row and whether it is still in its folder, kept in a spreadsheet with QC boxes per day. Three times a day, and once at midnight Eastern, the team's Telegram gets the count by stage.
- n8n
- Google Sheets
- Telegram
-
Weekly Top 50
WeeklyThe week's 50 biggest videos by spend, transcribed and sampled frame by frame, broken down by angle, hook, buildup, body, CTA and format with median thumbstop, ROAS and CPA for each. The write-up ends with what to brief next, and the angle tags go back onto the dashboard.
- Whisper
- ffmpeg
- Claude
- Google Sheets
-
Team knowledge bot
On demand · web and TelegramA chat that answers from the team's second brain: 900+ notes of call transcripts, concepts, audits, playbooks and the program's own ad numbers. Claude searches and reads the notes, cites them, flags advice older than six weeks, saves new team inputs back into the vault and can hand over the data as a new Google Sheet.
- Next.js
- Supabase
- Claude
- Telegram
- n8n
-
Creator Scout
On demandType /scout and a keyword. Apify searches TikTok and Instagram, Claude scores each creator against the ideal creator profile, the good fits land in a sheet and the shortlist comes back in the chat. The same pattern finds hiring leads for the recruitment side.
- n8n
- Apify
- Claude
- Google Sheets
-
AI editing app for creators
On demandRaw phone clips in, a finished 9:16 ad out: a transcript from Deepgram, an edit plan and text cards from Claude, shot labels from Claude vision, a render from ffmpeg. Roles, invites and monthly limits per member. It runs on a Mac mini behind a tunnel, with a Vercel front door and an installable web app on creators' phones.
- Node.js
- Postgres
- Claude
- Deepgram
- ffmpeg
-
Video texts
ContinuousA worker that writes down what every ad video says and shows, transcript and on-screen text by video ID, biggest spenders first, so the bot can search what the winners actually said instead of what we remember them saying.
- Deepgram
- Claude
- Supabase
-
Creator call reminders
Monday to FridayEmail and SMS reminders and announcements for the creator pods' daily calls, so the people who should be on the call are on the call.
- GoHighLevel
Winners, plainly
What counts as a winning video in a creator ad program?
The team's rule, and the one the dashboard applies to every video: at least $100 of spend, a blended ROAS of 1.4× or better and a cost per purchase of $26 or less. In the 30 days to 2 November 2026, 2,013 videos were judged and 239 passed, about 12%. That ratio is normal. A few videos carry most of the results, which is why the weekly analysis exists: find what the 239 have in common and brief more of it.
What does a growth specialist do in a creator ad program?
Keep everyone looking at the same numbers and shorten the loop between them. In practice: a weekly read of the top 50 videos by spend (angle, hook, CTA and format, with ROAS and thumbstop per family), audits of creator videos against those findings, a creator database the whole team edits, and the automation that moves an approved video to the media buyers in 15 minutes instead of a day. Media buyers spend the money and creators make the videos. I make sure the data from one reaches the other.
The stack behind it
- n8n
- Supabase
- Vercel
- Next.js
- Node.js
- Postgres
- Claude API
- Deepgram
- Whisper
- ffmpeg
- Apify
- Google Drive & Sheets API
- Telegram Bot API
- Triple Whale
- Northbeam
- GoHighLevel
- Remotion
Explainer videos for the team's tools are made in Remotion, voiced by AI, so a new teammate learns the dashboard in three minutes.