Creative velocity with AI video — A/B tests in a day
Creative tests used to drag for months. With AI a team ships 20 variants in a day. What that changes.
"Creative stagnation" was the core problem of media buying in the 2020s. Two clips swapped every three months, then surprise at the CTR drop. With an AI pipeline 20 creatives a day is reality.
Before
- Brief to agency — 1 week
- Casting + location — 2 weeks
- Shoot — 1-3 days
- Post — 1-2 weeks
- Approvals — 1 week
- Launch — 5-7 weeks total
- On a $30-100k budget — 4-6 final variants
After
- Brief — 30 min with marketer
- Script — 1 hour with GPT/Claude. 10 variants right away
- Generation — 2-4 hours for 10 clips
- VO/lip-sync — 1 hour
- Launch — next morning
- On a $3-10k budget — 20-50 variants
New A/B test rhythm
- Day 1: launch 20 variants on equal budget
- Day 3: stop bottom 70% by CTR
- Day 5: top 6 scale + variations of winners are produced in parallel
- Day 10: next batch of 20 creatives
- Loop never stops, marketing runs as a perpetual machine
Team changes
- Creative director → AI pipeline producer. Job: design prompt systems, manage templates, control iterations
- Designer → finishing artist. Polishes clips, fixes problem frames, ships to platforms
- Performance marketer makes winner/loser decisions on data, not taste
Pitfalls
- Audience fatigue. 50 variants per week into one audience — overdose, blur. Need rotation and segmentation
- Brand consistency. AI makes it easy to drift into noise. Need a strict style guide and QA filter pre-launch
- Quality metrics. CTR ≠ conversion. Test all the way down the funnel, not just the top