The context
I’m Head of AI (Marketing Automation) at Reverse Tech. The paid channel needed more creatives than any team could produce at a reasonable cost. Instead of hiring more people, we built a factory.
The problem
Creative production was the bottleneck of the entire channel. Every concept went through briefing, design, review and manual upload to the platforms. What we learned about what worked stayed in each trafficker’s head. The external tools that promised to fix it cost tens of thousands of euros a year and didn’t let us control the engine.
The system
A closed loop that goes from data to published ad:
- Winner mining: the system analyzes the creatives performing in our own account and in competitors’, and pulls concrete patterns out of them.
- From pattern to brief: those patterns become concrete production briefs.
- Routing by complexity: each asset goes to a person or to the generative engine depending on how hard it is. The routing rules live in our platform, not a vendor’s.
- Shipping via API: finished assets enter the Meta and Google ad ecosystems directly through their APIs.
- Learning loop: each ad’s performance flows back into the system and feeds the next round of concepts.
The result
Production went from ~200 creatives per week to over 3,000. Output per person rose about tenfold, including on creating the assets, which was the hard part, not just on uploading them. Building it in-house also removed ~€20k/month in external platform costs, and we own the engine and the data.