Video summary

Comment Analyser Et Itérer avec Andromeda (Ce que Personne ne Comprend)

Main summary

Key takeaways

Business

What Andromeda changes (Meta’s creative “fingerprint”)

  • The speaker explains that Meta updated Andromeda so performance depends less on simple volume and more on “strategic volume” via diverse creative variants.
  • Andromeda is described as a system that identifies similarity between creatives using a new “Creative ID” concept—so small code-level or superficial edits no longer fool the system.
  • Key consequence: ad “learning/eligibility” is limited when Meta considers two creatives the same, reducing spend scale and often hurting performance quickly.

Playbook: “Strategic volume” instead of repetitive volume

  • Build big, deliberate creative variation (not just more of the same):
    • Creative differences: new formats, personas, styles, hooks, vehicles, settings, and concepts.
    • Iteration philosophy: “always iterating / always creating new concepts.”
  • The speaker claims “small changes” (e.g., swapping one word like “love”) can now:
    • be classified as the same Creative ID
    • penalize the account by limiting new audience delivery / learning.

Andromeda Creative ID: what gets grouped as “the same”

Meta is described as comparing creative similarity beyond code. The speaker says Andromeda may treat creatives as the same if they share:

  • Same visual with different text
  • Same creator / pose / lighting / “PS” (production setup)
  • Same setting even with a different person
  • Same product + identical colors/setup + similar script
  • Same “hook type” (e.g., same emotional/promise structure), even if the delivery differs

Implication: If you test only by changing headlines, one phrase, or minor edits, Andromeda may not treat it as a new creative—so it won’t expand learning.

Concrete framework to create “new” creatives that Andromeda accepts

The speaker breaks creative iteration into components you must change enough to produce a new Creative ID:

  • Format (must change)
    • Example: turn a testimonial video into:
      • UGC
      • mashups
      • green screen
      • higher production version
  • Hook (must change)
    • Example tactic: keep the overall product proof but change the attention-grabbing opening (the “H” / hook).
    • Best practice described: when testing a new hook, also change supporting footage so the whole creative looks meaningfully different.
  • Persona / target framing (must change)
    • Change who “is being addressed” (the avatar’s identity), so Meta routes it differently.
  • Vehicle (the delivery method)
    • Example: replace the ad’s “delivery style” (e.g., landing-page style hook, UGC style hook).
  • Setting
    • Don’t film everything in the same environment (e.g., curtains/room).
    • Change location type: car, gym, store, different room.

Real-world example the speaker cites

  • They mention observing a competitor-like pattern (“Shapermind”):
    • Ads ran for ~178+ days
    • They used a same underlying script but changed footage + hook
    • Result: Meta treats these as different enough (different creative instances) because:
      • different faces/scenes
      • different hook framing
      • still maintains proof/credibility (“proof of …”)

Case impact: what happened when Andromeda hit their account

The speaker’s account allegedly had OK profitability and good results, then within one week:

  • Best scaling campaigns lost performance
  • Scaling campaigns spending ranges cited as:
    • $5K / $10K / $15K (depending on campaign)
  • ROS (return on spend) drop example:
    • ~2.15 ROS → ~2.09 ROS
  • They emphasize they’re still validating using information from Meta plus their own testing.

2026 measurement: “Andromeda isn’t a black box” (3 KPI signals)

The speaker recommends focusing on three metrics to diagnose whether an ad is being treated as top-funnel and/or is worth scaling.

1) Spend (Meta’s “signal”)

  • Spend is treated as the most important Meta signal—Meta allocates budget when it believes the ad creates a good user experience.
  • They stress: Meta optimizes for user experience + conversion behavior, not directly for your profitability.

2) Frequency (where you are in funnel / retargeting level)

Frequency is explained as repetition/intended exposure behavior (top vs bottom funnel).

  • Low frequency (~1–2) → Meta is mostly reaching new/unaware people (top-funnel discovery)

  • High frequency (>2) → more remarketing / retargeting, meaning the creative is less “new” in behavior and typically won’t sustain as well long-term

Example given:

  • A new product ad had ~$5K spend over last 5 days
  • Frequency ~25 → indicates it’s being supported by other ads and used more as mid/bottom funnel retargeting
  • ROS examples mentioned around ~1.75 (for another ad)

3) Cost per Result (CPR)

  • CPR = cost to acquire/produce a result (effectively “cost per sale/customer” in their context).
  • Nuance: a high CPR is not always bad—sometimes it’s educational or it works together with other creatives.
  • Decision example tied to a target:
    • “If CPR is above benchmark doesn’t mean it’s bad”
    • They reference an internal target cutoff, and mention scenarios such as:
      • CPR around $30 (sale example)
      • CPR ~33 as “below target cutoff” with profitability and ~20% margin stated

Decision rule described: Don’t cut solely by result cost—consider funnel contribution and other funnel metrics.

How to interpret funnel position (simple diagnostic rules)

The speaker provides decision logic based on Spend / Frequency / CPR / CPM trends.

Top-funnel (new people)

Look for:

  • Low frequency
  • CPM declining over time (not just day 1)
  • Spend increasing as results validate

If these conditions exist → keep the ad because it can “feed the entire account.”

Bottom-funnel / retargeting

Look for:

  • Frequency rising
  • CPM increasing
  • “Spin” (engagement metric implied) decreasing

You may reduce spend or repurpose the creative for later funnel stages.

When to cut

Cut if:

  • spin decreases
  • CPM / target price rises sharply
  • cost per result is unfavorable

Then: cut (Meta likely sees diminishing returns / limited incremental audience value).

Action recommendations: what to do when an ad shows problems

If CPR is high but CPM is low / frequency is low (top-funnel potential)

  • Keep the ad and create new hooks and/or improve sub-elements:
    • strengthen hook-to-avatar match
    • adjust pacing/script/sub-metrics
    • refine persona targeting (reduce “too many avatars” problem)

If CPR is acceptable but creative saturates (frequency high; spin down)

  • Iterate creative stage
    • move the creative into a different funnel stage
    • transform it into a solution/offer framing that supports later funnel
    • aim to lower frequency and restore performance

If you suspect missing objections / messaging gaps

  • Use ChatGPT as a prompt tool:
    • ask what objections remain / why it doesn’t convert directly
  • then inject additional ads (static or simple UGC) to fill gaps (social proof, objections handling, etc.)

“Perfect match persona” process (creative ecosystem approach)

A structured approach is recommended for 2026:

  • Define “perfect match” personas precisely:
    • pain, frustration, goals, fears
  • Use ChatGPT research to refine:
    • packaging of concepts for the persona
    • best format/vehicle style for that persona
  • Build a creative ecosystem:
    • multiple creatives per persona with different hooks/angles/POVs
  • Change creative POV:
    • e.g., father messages could be framed through child or spouse perception rather than the father’s own viewpoint

Presenter / source attribution

  • Presenter (speaker): “Matthéo” / “Mateo” (name appears inconsistently in subtitles; also referenced as the one who will share a document if viewers comment “Andromeda”).
  • Primary referenced source: Meta (the speaker says the guidance comes from communications/information Meta provides, plus account validation).

Original video