• 1PD Ops Club
  • Posts
  • Let's understand Meta’s Andromeda - Our lives depend on it!

Let's understand Meta’s Andromeda - Our lives depend on it!

Hey Marketers,

In this podcast, Meta’s VP Matt Steiner and Eric Seufert really gave new perspectives on how signal quality drives AI’s decisions. And I’m glad, it affirms my belief that your data is basically the language Meta’s AI speaks.

So… What was the big conversation about?

Matt Steiner basically went through Meta’s AI advertising playbook - the evolution from manual optimization to machine-led campaign learning.

He explained:

  • The old Meta was built on manual input: audience targeting, lookalikes, ad set experiments, etc.

  • The new Meta runs on structured data that teaches the system what success means for your business.

That’s where the idea of Signal Engineering becomes everything. 

The new perspectives from Steiner

This is where the geeky stuff got interesting. Steiner mentioned several internal systems powering Meta’s ad optimization pipeline:

  • Andromeda - This is where signals (purchase, add-to-cart, subscription) are evaluated in real time to decide which ad gets shown.

  • Lattice - It processes trillions of predictions daily, basically, the “brain” that learns what converts for each advertiser.

  • GEM - It uses prior campaign data to suggest or generate ad creative variations that might improve performance

  • Advantage+ Shopping - Automates bidding using model-learned signal strength.

What Meta’s AI Actually Learns From

Steiner explained that the ad delivery system takes in structured event data, meaning the server-side event data you send via Conversions API (CAPI). 

That’s how Meta finds patterns that humans can’t.

This is why the consistency and meaning of your signals matter way more than how many events you send.

Why It’s a Wake-Up Call for Performance Marketers

The old game of targeting hacks and audience lookalikes is gone.

When you combine that with first-party data and server-side pipelines, you move from:

  • guesswork - intelligence,

  • random testing - predictive learning,

  • vanity ROAS - consistent LTV growth.

Let’s talk.

Are your signals properly sent inside Meta? Do enroll in our 14-day free trial. For more clarity, book a call with us and get it resolved.

P.S. Leaving the podcast link: Check it out

Reply

or to participate.