https://www.graphed.com/ - AI data analyst to build dashboard and get insights
In this episode, Cody hosts Romain Torres, co-founder of ArcAds, to dive into the next evolution of paid advertising — using AI to generate and test thousands of ad creatives at scale. Romain shares the strategies behind ArcAds’ explosive success and how marketers can now use AI agents, automation, and avatars to unlock hyper-efficiency in ad performance. This episode is a masterclass in modern performance marketing for eCommerce, mobile apps, and agencies alike.
Timestamps:
00:00 – Why top e-commerce advertisers generate 1,000s of creatives
01:10 – The old UGC content model: $100K for 1,000 ad variations
01:57 – ArcAds and AI avatars explained
03:00 – Lessons from gaming companies and their ad testing obsession
06:30 – How to make ad testing 10x cheaper and easier with AI
10:20 – Meta’s only recommendation to big advertisers: more creatives
13:10 – Why creative volume is now the biggest growth lever
15:00 – Romain’s 6-week creative testing sprint process
20:05 – The “Notion board” system for organizing ad experiments
25:40 – Automating script generation via Facebook Ad Library + Whisper
29:20 – AI’s true strength: copying and remixing top-performing formats
35:15 – Real examples: language apps, e-com, and viral ad structures
40:05 – Why localization with AI avatars is a game-changer
44:40 – Using failure and emotion in ads (gaming tactics for e-com)
47:50 – “Ads that don’t feel like ads” – winning creative philosophy
49:10 – Final frameworks and where to start with AI ads
Key Points:
• AI is shifting ad creative from an expensive, human-led process to scalable, high-volume automation — unlocking massive performance gains
• Meta’s performance advice is now centered on one thing: creative iteration
• ArcAds enables users to generate hundreds of UGC-style video ads using AI avatars, voice synthesis, and automated scripting
• Creative success depends on three pillars: strong scripts, tested variations of actors, and good editing
• Winning ad strategies rely less on creative instinct and more on statistical volume — test everything, let the data decide
Creative Frameworks Discussed:
Weekly Iteration Loop
- Organize creative ideas in a Notion board
- Test 10+ variations per concept
- Review weekly results → double down on winners
- Commit to a 6-week testing cycle to uncover scalable concepts
AI Agent Automation Workflow
- Scrape competitors’ Facebook ads using the Ads Library API
- Transcribe videos with Whisper
- Analyze hooks and trends with GPT
- Generate new scripts, swap in AI avatars, and produce at scale
Best Performing Ad Formats
- UGC-style narration with product demo B-roll
- Split-screen “AI tutor” dialogue format for language apps
- Localized voiceovers for different geographies
- Street interview simulations using avatars for finance/dating apps
Growth Tactics:
• Use AI to localize ad content and reach global markets without extra production
• Automate creative inspiration by spying on competitors and remixing their winners
• Build feedback loops with performance data to fuel ongoing ad ideation
• Don’t try to guess the best creative — let scale + data reveal the winner
Notable Quotes:
“You just don’t know which actor is best for your ad until you test it.” – Romain Torres
“Meta figured out the targeting. Now you have to figure out the creative.”
“AI is bad at being creative — but it’s
amazing at copying what works.”
“Your edge is not doing one thing well. It’s doing everything at 100x volume.”
Guest:
https://fr.linkedin.com/in/romain-torres-arcads
https://x.com/rom1trs
https://www.arcads.ai/
