The before-state: we were a "normal" video agency
18 months in. 12 DTC clients on retainer. A team of 7 — 2 producers, 3 editors, 1 motion designer, 1 PM. We delivered 4 to 8 videos per month per client, billed at $900–$1,700 a piece. Gross margin: 28-32%. Monthly revenue around $92k. We did good work. We were proud.
And yet, every month I'd look at the P&L and feel something was off. The team was burned out. Editors wrapping at 10pm. Clients complaining about turnaround ("3 weeks for 6 videos, really?"). And we'd complain about briefs. Classic spiral.
The tipping point: September 2025
In 4 weeks, 3 clients churned. Three. All at once. They all said the same sentence, almost word for word:
"You guys are good, but you don't deliver fast enough or in enough volume. We found something at 1/5 the price that ships 30 videos a week."
The "something"? An AI stack built in-house by a growth marketer. Not pretty, not premium — but volume + speed. And the Meta Ads market loves volume.
I closed my laptop, went for a walk, and made the call: we kill the classic agency. We pivot. 90 days.
The 90-day pivot, step by step
Days 1-15 — Audit & stack selection
- Internal audit: how many hours actually spent per deliverable? Answer: 11h on average for a single UGC video. Embarrassing.
- Tested 6 AI stacks on 30 test videos. Final pick: Claude (script + brief) + Hoox (video generation) + Meta Ads Manager (test & scale).
- 1 pilot client chosen: a DTC skincare brand we'd built trust with, ready to sign an NDA and accept beta status.
Days 15-45 — Building the orchestrated pipeline
- Architecture: one human orchestrator drives 5 AI agents in parallel (audience research → angles → scripts → video gen → QA + post).
- Trained the remaining 3 people on the orchestrator role — not executor. We let 4 people go. The hardest call of the pivot.
- First tests: 12 videos in 2 days for the pilot client. CTR x1.8 vs old produced videos.
Days 45-75 — Pilot on 3 clients, before/after metrics
- 3 clients moved to orchestrator-mode. Signed a contract amendment: same price, 10x volume delivery.
- Brutal measurement: Meta Ads CPA on AI videos vs human videos. Result: -34% CPA on average.
- 1 client refused the pivot. We let them go cleanly, no hard feelings.
Days 75-90 — Reprice & scale
- Repriced every client: shifted from "price-per-video" to "orchestrator subscription" ($4,290/month for 50-100 videos/week).
- Of 11 remaining clients: 9 accepted immediately, 2 negotiated. Zero left.
- Infrastructure scaled: added a 2nd human orchestrator to absorb 4 new clients signed in November.
Results at D+90
| Metric | Before | After | Delta |
|---|
| Gross margin | 30% | 80% | +170% |
| Volume delivered | 4-8 videos/mo | 50-100/wk | x40 |
| Time-to-deliver | 3 weeks | 1 day | -95% |
| Annual churn | 25% | <5% | -80% |
| Monthly revenue | $92k | $129k | x1.4 |
| Headcount | 7 | 3 | -57% |
3 sales scripts that changed everything
Script 1 — Pitching the pivot to an existing client
"We spent 6 months measuring one thing: why your videos perform or don't. Factor #1 isn't shoot quality. It's volume of angles tested. Starting next month, we ship 50 to 100 videos a week instead of 6 per month. Same budget. You test 10x more angles, you find your winners 10x faster. Want to start Monday?"
Script 2 — Qualifying an orchestrator-fit prospect
"3 quick questions: (1) How much do you spend on Meta Ads monthly? (2) How many distinct videos do you test per week today? (3) If you could test 10x more on the same budget, what changes for you?" — If answer to (1) > $22k/mo and (3) "everything," it's a fit. Otherwise, we say no.
Script 3 — Objection "the human quality?"
"I get it. Here's the deal: we ship 20 videos this week, free. You push them on Meta. If CPA is worse than your current videos, we walk. If it's better, we sign. Human quality isn't what feels right — it's what converts."
5 client objections + handling
1. "It's AI, people will spot it"
"Across 8,200 videos shipped in the last 6 months, the rate of 'this is AI' comments is 0.3%. The other 99.7% comment on products, humor, the angle. Viewers scroll. They don't forensic."
2. "We want to keep human sourcing"
"Perfect — keep 20% human-sourced for hero pieces (real customer testimonials, founder story). The other 80% in orchestrator mode for volume testing. You get the best of both."
3. "Too expensive"
"Compare cost/video, not total. At $4,290/month for 50 videos/week (~200/month), you're at $21 per video. A human UGC is $900. You divide by 40."
4. "Untested in our vertical"
"We've shipped across 14 verticals: skincare, supplements, fashion, electronics, fitness, food, pet care, kids, jewelry, beauty tools, home, gaming, edtech, B2B SaaS. If yours isn't there, first month at 50% — you test risk-free."
5. "Quality risk not under control"
"QA is our obsession point. Every video passes 3 checks: technical (duration, ratio, audio), brand (no banned claims, no celebrity-adjacent faces), performance (CTR prediction via internal model). Nothing ships without that gate."
What I'd want to tell you if you're hesitating
The pivot wasn't clean. I let go of 4 people I cared about. I slept 4 hours a night for 3 weeks. I was scared no client would stay. But at D+90, I'm running a company I no longer recognize — in the right way. Margin x2.6, volume x40, lean team, happy clients.
If you're where I was in September — churning clients, burnt team, ugly P&L — you have 90 days. Not 12 months. 90 days. The market moves faster than your annual strategic plan.