Video summary

+1M€/MOIS ADSPEND, CLAUDE CODE, META ADS, RECHERCHE AVATAR | MASTER #46

Main summary

Key takeaways

Business

Business Execution Summary (Ads + AI “Acceleration” Playbook)

Core Strategy: “AI as Accelerator” + Human Creative/Information Loops

Multiple speakers emphasize that AI is not the business—it accelerates execution, especially information gathering. Humans provide the feedback loop that turns outputs into performance improvements, including:

  • Learning what works
  • Refining prompts and copy
  • Validating angles and avatars

They frame “If you don’t do [AI], you’re dead” as table stakes, with the next level coming from:

  • Deep avatar understanding
  • A continuous learning system: inputs → outputs → human feedback → improvement

Competitive Ad Intel System (“Ad Radar” Before Winners Scale)

A concrete internal process is described to monitor competitor ad performance using a frequency/impression trigger.

What they built (playbook)

  • Track competitor ads and sort by impression count
  • Trigger an alert when a competitor ad “explodes” in impressions (via fast refresh/monitoring cadence)
  • When the winner signal appears:
    • Identify the angle
    • Reuse the avatar
    • Recreate variations quickly—potentially before the competitor hits full scaling

Operational output

  • “Radar on all competition”
  • “Alert”
  • “AI recreation” of creatives (sometimes analysis first, then recreate)

Tools mentioned (examples)

  • Claude Code / “Cloud Code”
  • Gemini / “Twislab” / other LLM vision-based analyzers
  • “Frame-by-frame” analysis tools for competitor creatives (analyze “all footage,” then frames)

Creative Reconstruction Pipeline (Competitor Ad → Adapted UGC → Finished Ad)

They outline an end-to-end workflow using multiple AI components.

Process steps

  1. Frame-by-frame reverse engineering of the competitor ad (analyze visuals second-by-second)
  2. Psychological/copy analysis agent: determine why it works (copy + creative strategy)
  3. Recreate frames using a script plus adapted brand content
  4. Animate frames using tools such as:
    • Kling
    • Sora
    • Remotion
  5. Produce UGC-style outputs:
    • V1 AI concept
    • V2 improvements via further analysis/tuning

Key idea

Don’t just “clone”—understand the reason the creative performs, then adapt it to your brand/footage.

“Copy Meaning” Process (Systematic Ad Cloning + Iteration)

A structured method for generating new winning creatives.

Copy process (playbook)

  • Start by copying successful ads in your niche using the same “approach”
  • Expand inspiration:
    • From other niches (more technical; harder to adapt)
  • Iterate in batches:
    • Improve hooks without changing the entire concept
    • Revise the script (if you have creative strategist expertise)
    • Keep the concept but change the angle (avatar-specific targeting)

MrBeast analogy

Keep a winning format, then improve on the “next level” execution.

Guardrail: Copy + innovate

You can copy and win, but to scale without CPM inflation, you must add innovation:

  • New concepts outside the industry
  • Diversify creative inputs and angles

KPI / Metric Focus: CPMR, Frequency, Retention Curves, Conversion Rate

CPMR (Predictive Scaling Metric)

They repeatedly reference:

  • CPMR = CPM × frequency

Why it matters:

  • Rising CPMR suggests you may face higher costs (CPM/CPA) and reduced scalability
  • It can indicate you’re not reaching enough new prospects

Retention Curve Analytics (Video Performance)

They use AI to analyze Meta retention/drop-off patterns, including an example where:

  • Content drops early, then flattens after ~30s–1min
  • They claim they identified correlations like: “too many faces in the first 20 seconds” → retention drop

They also argue that analyzing volume + AI helps avoid “losing” learning from ads that weren’t profitable immediately.

Conversion Rate Tracking by Role

They claim conversion depends on who writes:

  • Handwritten copywriter converts better than AI-assisted creative writing in their tests
  • Reason: the human copywriter pays more attention and learns the avatar deeper

Ad Spend + Testing System (Budget Allocation + Campaign Types)

Testing budget rule

  • Keep a stable base budget at about 80%+ of spend
  • Use the remainder for testing
  • Inside testing:
    • Allocate about 20% of the testing budget to creative tests allowed to “burn” (intentional waste for learning)

Campaign architecture

They describe several campaign types:

1) Creative testing campaigns

  • Created frequently (example cited: “4 campaigns per day” style)
  • Each CBO contains ~15–20 creatives
  • Run minimum 3 days, extend if very profitable (or assess after ~1 week)

2) Landing page testing

  • Duplicate creatives onto new landing pages
  • Use persona-specific angles (example: “test testosterone-fueled landing pages” with testosterone creatives)

3) Scale campaigns

  • “One Campaign army”: duplicate winners into scaling setups
  • Scaling tends to use manual bidding (“manual beeding”)

4) Zombie campaign

  • Store losing creatives
  • Keep them capped/managed by CPA threshold (example: CPA ~$50)
  • Redeploy later if needed

Bid/cap testing example

They mention duplicating the same campaign under:

  • bid cap or CPA/cost caps

Example outcome:

  • Conversion rate reportedly improved from ~5% to ~8% using a cost cap approach

Landing Page Offer + Positioning Must Come Before Creative Volume

They explicitly warn:

  • Creative volume alone won’t fix poor offer or positioning
  • First lock positioning + mechanisms, then scale creative variety

Positioning “Mechanisms” and Avatar Depth as Differentiator

“Mechanism” framing

  • A mechanism is the distinct value framing inside positioning:
    • “unique problem mechanization”
    • “unique solution mechanization”
  • The product stays fixed; angles change
  • You can generate many angles from one product by changing:
    • the problem framing
    • the unique angle/story
    • the messaging mechanism

Avatar as cornerstone

Their strongest repeated claim: avatar understanding is the differentiator.

Example (weight loss):

  • Naive: “lose weight with this diet”
  • Avatar-aware: “keep eating McDonald’s without gaining fat”

They argue AI can accelerate research via sales data and pattern discovery, but founders/marketers must still manually validate avatar insights.

Team / Ops: SOP-Driven “School” for Onboarding + Continuous Training

Organizational framework

  • “Business is succession of SOPs.”
  • They run an internal “school” per role:
    • People receive tabs/classes tied to their position
    • Onboarding uses SOP libraries + feedback loops + trial periods

Hiring/onboarding approach

  • Recruit many candidates
  • Run structured short trials
  • Keep only best performers and onboard them into the SOP school

Founder involvement principle

Founders shouldn’t spend most time building everything at the start. Instead:

  • Hire someone to set up systems
  • Maintain enough oversight to understand execution 100%

AI Toolstack: Claude Code (Vibe Coding) + Gemini/GPT + Automation Architecture

Claude Code architecture (framework)

Components described:

  • Skills: task-specific knowledge/actions
    • e.g., statistics generation via Gemini
  • MCPs: connect Claude to tools (ClickUp, Slack, etc.) to sync data and automate workflows
  • APIs: access tools without MCPs
    • e.g., generate frames via Gemini/Kling; “create bowls/videos”

Deployment/security

Claude Code can be run:

  • On a local Mac/PC terminal
  • Or on a VPS for security and team sharing

They highlight security risks if granting full access to an agent on a personal machine.

Use of “Manus” (High-Level)

Manus is described as a simplified multi-step executive assistant:

  • Voice/audio input → outputs like:
    • emails
    • appointment booking
    • research tasks
    • presentations

Examples mentioned:

  • Creator whitelisting / influencer discovery
  • Email sequences tailored to individual answers
  • Competitor ad analysis (frame-by-frame via prompt)

Key limitation: price/cost.

Advertising “Human Trust” vs AI Personalization

They claim big brands still win with humanized trust, such as:

  • Human founder presence on landing pages
  • Influencer/authority in ads

They argue AI “human rendering” on landing pages may fool less, but authenticity still matters most in ads with authority.

Concrete Metrics / Targets Mentioned

Creative/ad scaling

  • Zombie campaign CPA target example: ~$50 CPA
  • Retention analysis: early drop related to faces in first ~20 seconds

Performance examples

  • Rank improvement: “Adlier” from rank 15 → rank 3 in a single day (as claimed)
  • Conversion lift example: 5% → 8% when switching to CPA/cost cap approach

Volume + cadence guidance

  • A question about ~300 creations/day; advice depends on:
    • diversity
    • offer fit
    • not a fixed number
  • Landing page testing: about 3 per week on big shops (example)
  • Creative testing: run minimum 3 days, extend if profitable beyond ~1 week

Spend level referenced

  • “+€1M/month ad spend” (scale emphasized; no additional hard KPI targets stated)

High-Level Closing Tactical Tip

  • Share daily learnings with the team to motivate and align on the learning loop
  • Continuously improve AI systems because models/LLMs update rapidly

Presenters / Sources Mentioned

Presenters / guests

  • Valer (Val)
  • Lucas
  • Nico
  • Matthéo

Other named sources/tools/brands/cases

  • David Foggy (post mentioned)
  • MrBeast (example)
  • Comfort (brand example)
  • IM8 (brand example)
  • MarkBll Brands (example)
  • Gruns (brand/funding example, “supplements”)
  • Meta (ads, Meta ad manager concepts)
  • Shopify (backend/resubscribe tracking mentioned)

AI / tools mentioned

  • Claude / Claude Code (including “Claude Code context” like Open (G))
  • Gemini, GPT/GPT Chat
  • Manus, Lovable
  • Claude/MCP concept stack
  • ClickUp, Slack, Figma
  • Remotion, Kling, Sora
  • Dev/Facebook developer app/API setup
  • Melio, Nano Banana
  • Paper Clip (open-source OS concept)

Rate this summary

Your feedback will help improve summaries.

Improve this summary

Reprocess with a stronger model when the summary feels incomplete or inaccurate.

Pro

Translate summary in another language

Pro

Ask questions to this video

Chat for follow-up questions, clarifications, and source-backed answers.

Coming soon

Share this summary

Share on X/Twitter Share on LinkedIn Share on Reddit

Original video