Meta Andromeda: what it means for catalogue ads and creative
What Meta's Andromeda ad retrieval system actually is, what Meta has said about it, and what it changes for catalogue ads and the creative you feed them.
AI Shopping Feeds
“Andromeda” has become shorthand for a lot of things in Meta advertising: creative volume, broad targeting, consolidated campaigns. Some of that is supported by what Meta has said. Some of it is advice people have attached to the name. This post separates the two, then looks at what it means for catalogue ads in particular.
What Meta has actually said
Meta described Andromeda on its engineering blog on 2 December 2024. The main points, in Meta’s terms:
- It handles retrieval. Meta’s ad delivery works in stages. Retrieval comes first: it narrows tens of millions of ad candidates down to a few thousand for a given person. Later, larger ranking models choose the ads that are actually shown. Andromeda is the retrieval stage.
- It’s built for scale. Meta says it runs on NVIDIA Grace Hopper Superchips and its own MTIA chips, and that it supports a 10,000x increase in model capacity compared with what came before.
- Meta reported improvements. The post reports a 6% improvement in recall for the retrieval system and an 8% improvement in ads quality on selected segments.
- More ads are coming into the system. Meta links Andromeda to Advantage+ automation and generative AI creative, which add more ads for different audiences, and says Andromeda uses a hierarchical index to cope with that growing creative volume.
- Diversity is a stated goal. Looking ahead, Meta says it expects future changes to retrieve a more diverse set of ad candidates.
That’s the primary source. Anything beyond it, including specific rules about how many ads you need, comes from advertisers’ own experience and interpretation, not from Meta’s description of the system.
What it means in practice
Here’s our reading of the implications. Treat it as analysis, not as a rule Meta has published.
Retrieval decides what gets considered at all
If an ad isn’t retrieved for someone, ranking never sees it. The better the system understands your ads and products, the more likely the right one makes the shortlist for each person. That puts weight on two things you control: the quality of your creative, and the quality of the data behind it.
Distinct creative beats duplicated creative
Meta’s stated direction is towards a more diverse set of candidates. A dozen near-identical versions of the same image, with a different headline on each, are not very diverse. Different layouts, different product angles, different offers and different formats are. If you’re making variations, make them genuinely different.
Product data is creative data
For catalogue ads, each product is effectively its own piece of creative. The system can only match products to people well if it knows what the products are. Clear titles, accurate categories, complete attributes such as colour, size and material, and good images all help. Our AI Product Optimisation rewrites thin titles and fills missing attributes in bulk, and Feed Audit finds the gaps.
What it means for catalogue ads
Catalogue ads were already built on the idea Andromeda leans into: lots of distinct ads (one per product), chosen per person by the system. A few things follow.
- A big, clean catalogue is an asset. Every product with accurate data and a good image is another distinct candidate. Products with broken images, missing prices or vague titles are candidates the system can’t use well.
- Catalogue creative is worth testing as its own variable. Plain product shots, branded templates and lifestyle images give the system different kinds of candidate for the same products. A designed template doesn’t replace your other creative. It adds a different kind.
- Several formats help. A product image designed for square, portrait and full-screen story placements gives the system creative that fits each placement, rather than a square image cropped to fit.
- Keep the catalogue fresh. Out-of-date prices and stock make products worse candidates, and can get them rejected.
A practical checklist
- Product titles say what each product is, with brand and key attributes
- Every product has a working, good-quality image
- Categories and attributes are filled in, not left blank
- Pixel or Conversions API content IDs match catalogue IDs
- Catalogue ads run alongside other creative, not instead of it
- Creative variations are genuinely different, not small tweaks
- Product images exist for the placements you buy, including 9:16
Where branded catalogue creative fits
If your catalogue ads currently show plain product shots, a branded template is one of the most distinct variations you can add, and it covers every product at once. Our DPA Creative Editor renders a template for every product in your feed, in square, portrait or story format, and keeps the images in step with your prices. For the basics of the format, start with how dynamic product ads work.
Frequently asked questions
What is Meta Andromeda?
Andromeda is the system Meta uses for the retrieval stage of ad delivery: narrowing tens of millions of possible ads down to a few thousand candidates for each person, before ranking models pick the ads that are shown. Meta described it on its engineering blog in December 2024.
Does Andromeda change how catalogue ads work?
Catalogue ads still pick products from your catalogue using your events and product data. What Andromeda changes is how Meta finds candidate ads at scale, which puts more weight on giving the system distinct, good-quality creative and accurate data to choose from.
Do I need more creative because of Andromeda?
Meta has said Andromeda was built to handle the growing volume of ads, including AI-generated ones, and that it expects to retrieve a more diverse set of ad candidates. Variety that is genuinely different, such as different layouts, angles or offers, gives the system more to work with. Near-identical copies of one ad add little.