Meta Andromeda is Meta's machine learning system for ads retrieval, the first step of its ad recommendation process, where tens of millions of possible ads are narrowed to a few thousand relevant ones for each person. Meta's engineering team described it in December 2024 as a personalised retrieval engine running on NVIDIA Grace Hopper Superchip and Meta's own MTIA hardware, deployed across Facebook and Instagram.
Why advertisers care: Meta says it built Andromeda because the number of eligible ads is exploding, driven by Advantage+ automation and generative AI creative tools. Its own example is that more than a million advertisers used its generative AI tools to create more than 15 million ads in a month. Andromeda exists to sort through that volume, which tells you how Meta now expects ads to be made.
Where Andromeda sits in Meta's ad system
| Stage | What happens | Scale Meta describes |
|---|---|---|
| Retrieval (Andromeda) | Selects relevant ad candidates for a person | From tens of millions of candidates down to a few thousand |
| Ranking | Larger, more complex models predict value for the person and the advertiser | Works on the few thousand candidates from retrieval |
| Final selection | The ads shown to the person | The final set picked after ranking |
Meta says retrieval processes three orders of magnitude more ads than the stages after it, under tight time limits, which is why it's the hardest part to scale.
Meta names two pressures on this stage. The first is volume: features such as predictive targeting are expensive to compute, and Advantage+ keeps adding eligible ads by automating audiences, budgets, placements and creative. The second is speed: ads have to be picked fast enough not to disrupt what someone is viewing, while both delivery and each person's interests keep changing in near real time.
What changed with Andromeda
According to Meta's write-up, earlier retrieval systems used only limited personalisation, with isolated model stages and many rule-based shortcuts. Andromeda replaces that with:
- A much bigger model. A custom deep neural network that Meta says increases model capacity by 10,000 times for better personalisation.
- A hierarchical index. Ads are organised in layers so the system only checks the most relevant branches, trained together with the model.
- Model elasticity. Model complexity adjusts in real time to available computing power, which Meta says improves inference efficiency by 10 times.
- Faster feature work. Moving feature extraction onto GPUs gave more than 100 times better latency and throughput than the earlier CPU-based parts.
- Fewer rules. Less rule-based logic, so the whole system can be optimised end to end.
What Meta says it delivers
Meta reports that Andromeda improved recall in its retrieval system and ads quality on selected segments after it was deployed across Instagram and Facebook. It also links it directly to Advantage+, the suite that automates audience targeting, budget allocation, placements and creative. The post states that advertisers who turned on Advantage+ creative's AI-driven features saw higher return on ad spend, and Meta estimates that businesses using image generation are seeing more conversions.
Looking ahead, Meta says the model is expected to move to an approach that delivers a more diverse set of ad candidates, and it estimates another 1,000 times increase in model complexity as it runs on MTIA and future GPUs.
Targeting still shapes who sees an ad, and our Facebook ads targeting guide lists every option and its limits.
What it means for your Facebook and Instagram ads
Meta doesn't publish a checklist for Andromeda, so the points below are my reading of what its engineering post says, not Meta's instructions:
| What Meta says | What I'd do |
|---|---|
| Andromeda is designed to make use of rapid growth in the number of ads | Give the system several distinct creatives per ad set rather than one ad |
| Advantage+ automates audience, budget, placements and creative | Test broad audiences with Advantage+ audience, which Meta recommends A/B testing for almost all campaign types |
| Future versions aim for a more diverse set of ad candidates | Vary the angle, format and hook between ads, not just the colour of a button |
| Advantage+ audience isn't recommended for retargeting | Keep retargeting campaigns on your own custom audiences |
If you're producing more creative to match, our AI prompt pack for Meta ads speeds up drafting, and our breakdown of Facebook ad creatives that convert shows what the variations should differ on.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
What Andromeda isn't
- It isn't a setting in Ads Manager. You can't switch it on or off.
- It isn't a new campaign type. It works behind every ad that enters Meta's system.
- It doesn't replace ranking. Meta describes ranking as a separate, later stage with larger models.
My verdict
Andromeda is plumbing, but it's plumbing that tells you Meta's direction: more ads, more automation, more personalisation. The practical response isn't a new trick, it's a habit: keep a steady supply of clearly different creative in each campaign and let Advantage+ test it, while you hold back your retargeting audiences for campaigns you control.
Creative variety matters more as delivery changes, and our Facebook video ads guide covers upload limits and safe zones for video.
Frequently asked questions
What is Meta Andromeda?
It's Meta's machine learning system for ads retrieval, the first stage of its ad recommendation process. Meta says it narrows tens of millions of ad candidates to a few thousand relevant ones.
When did Meta announce Andromeda?
Meta's engineering team published its write-up on Andromeda in December 2024, describing it as deployed across Instagram and Facebook.
Can I turn Andromeda on in Ads Manager?
No. Andromeda is part of Meta's ad delivery system, not a setting or campaign type you choose.
Does Andromeda affect Advantage+ campaigns?
Yes. Meta's post links it directly to Advantage+, saying it's built to handle the growth in ads that Advantage+ automation and generative AI creative tools create.
Do I need more ad creatives because of Andromeda?
Meta doesn't say so directly, but its post says Andromeda is designed to make use of the growing number of eligible ads, which is why I'd run several distinct creatives per campaign.
