What Is Meta Andromeda (and What It Means for Your Ads)

Meta quietly rebuilt the engine behind its ads. Here's what Andromeda is, what changed, and why it rewards presence and creative over micro-targeting.
Niels Stoelinga
Creative Director
·
Quick Take Aways
  • Andromeda is Meta's machine-learning retrieval layer: it scans tens of millions of ads and shortlists the most relevant before ranking picks the winners.
  • Meta reports roughly 10,000x the model capacity of the old system, a +6% lift in recall and +8% in ad quality. These are Meta's own figures.
  • As the matching gets this good, micro-targeting stops being the edge. The edge is strong creative and clean signal.
  • Andromeda rewards brands that show up with fresh creative continuously, not in bursts. That's the case for an Always-On System.

Meta quietly rebuilt the engine that decides which ads you see. Most advertisers haven't noticed. The ones who adapt first will win the next two years.

Meta Andromeda is a machine-learning retrieval system that sits at the front of Meta's ad-recommendation pipeline. It scans tens of millions of possible ads and shortlists the few thousand most relevant to each person, before the ranking stage picks the winners. It's the part of the system that decides which ads even get considered.

What changed

Retrieval used to be the dumb part of the funnel, a rough first filter. Andromeda makes it the smart part. Meta rebuilt it on deep neural networks running on NVIDIA Grace Hopper hardware, with roughly 10,000x the model capacity of what came before. Meta reports a +6% lift in recall (better ads surfaced) and +8% in ad quality. It's the engine underneath Advantage+, and it rolled out broadly through 2025.

Translation: Meta got dramatically better at finding the right ad for the right person, automatically, from a far bigger pool, without you hand-building the audience.

What it means for your ads

Here's the part the targeting-obsessed crowd is missing. When the machine gets this good at matching, micro-targeting stops being the edge. You're no longer winning by slicing audiences thinner than the next agency. You're winning by feeding the machine two things it can't generate itself: strong creative and clean signal.

Strong creative, because Andromeda can only choose between the ads you give it, and it rewards the ones people actually respond to. Clean signal, because the system learns from the events you send back (a well-implemented Conversions API, high Event Match Quality). Get those two right and the machine works for you. Get them wrong and no targeting trick saves you.

So the job changes. Less time in Ads Manager pulling levers. More time making more, better, on-brand creative, consistently, not in bursts.

Why this is the case for Always-On

Which is exactly the case for an Always-On System. Andromeda rewards brands that show up with fresh, strong creative all the time, not the ones who go dark between campaigns and flood the feed for two weeks a quarter. The algorithm now pays a premium for presence. We call that the Always-On System: continuous creative production, fed into the machine, compounding while everyone else restarts from zero.

The rules changed. They'll change again. The brands that build a system instead of chasing each update are the ones that stay ahead.

FAQ

What is Meta Andromeda?
A machine-learning retrieval system at the front of Meta's ad pipeline. It scans tens of millions of possible ads and shortlists the few thousand most relevant to each person, before the ranking stage picks the winners. It decides which ads even get considered.

What changed with Andromeda?
Retrieval used to be a rough first filter. Meta rebuilt it on deep neural networks running on NVIDIA Grace Hopper hardware, with roughly 10,000x the model capacity of the old system. Meta reports +6% recall and +8% ad quality. It's the engine under Advantage+ and rolled out broadly through 2025.

How should advertisers respond to Andromeda?
Stop competing on micro-targeting and start feeding the machine what it can't generate: strong, on-brand creative and clean signal (a well-implemented Conversions API, high Event Match Quality). More, better creative produced consistently beats slicing audiences thinner.

What does Andromeda have to do with Always-On marketing?
The system rewards brands that show up with fresh creative all the time, not ones that go dark between campaigns. The algorithm now pays a premium for presence, which is exactly what an Always-On System delivers: continuous creative production feeding the machine.

The bottom line

Meta rebuilt the engine that decides which ads get seen, and it now rewards presence and creative quality over targeting tricks. The advertisers who win the next two years won't be the ones slicing audiences thinnest. They'll be the ones feeding the machine strong, on-brand creative and clean signal, consistently. That's the case for an Always-On System.

Download The Always-On Playbook
Free for all brand s that inspire us.

The Sources used in this Article:

  • Engineering at Meta — Andromeda announcement, engineering.fb.com (December 2024). The figures (~10,000x model capacity, +6% recall, +8% ad quality, NVIDIA Grace Hopper) are Meta's own reported numbers, presented as "Meta reports", not independently verified.
  • Andromeda as the engine under Advantage+ and its broad 2025 rollout: Meta communications.