How Andromeda works
Every time someone opens Facebook or Instagram, an advert has to be chosen almost instantly from an enormous pool of eligible ones. Meta does not score them all. The work is split into stages, and Andromeda is Meta’s name for the first of them: retrieval, the stage that narrows the pool down to a manageable shortlist of candidates worth examining properly.
Retrieval is a matching problem rather than a pricing one. It asks which adverts are even plausible for this person in this moment, based on what the system holds about them and what it understands about each advert. Only the survivors are passed on to be ranked in detail, and only then does anything resembling an auction between advertisers take place.
This is Meta’s own vocabulary for its internal engineering, published by Meta rather than discovered by advertisers. There is no Andromeda setting in Ads Manager, no report that shows how your adverts fared at this stage, and no lever anyone outside Meta can pull.
Why it matters
It explains a result that otherwise makes no sense: an advert that is approved, funded and technically eligible but barely delivers. If the advert never reaches the shortlist for the kind of person you want, its bid is irrelevant, because bids are compared later. Raising the budget will not fix a retrieval problem.
It also explains why supplying several genuinely different adverts tends to beat perfecting one. Different concepts, formats and framings give the system more distinct chances to be considered plausible for different people. A campaign built on one idea in five colours is one candidate wearing five shirts.
Where the term gets misused
The main risk is a sales pitch. Any agency or tool that offers to optimise for Andromeda, or claims to know how the retrieval stage weighs your account, is describing something they cannot see and cannot test. Meta publishes the shape of its systems, not the recipe, and the recipe changes. Ask what would be measured to prove such a claim, and the conversation usually ends.
The quieter misuse is diagnostic. Blaming a hidden engine is a comfortable way to avoid checking the things you can actually see: a rejected advert, a starved ad set, a broken pixel, a landing page that loads slowly on a mobile connection.
How to act on it
Treat retrieval as background knowledge, not a task. What it should change is where you put your effort: into having enough distinct creative in circulation, into keeping audiences wide enough that the system has room to find people, and into creative diversity rather than endless small edits of one advert.
When delivery is poor, work through what is visible first: approval status, budget, the optimisation event, whether conversions are being received at all. If a campaign genuinely will not spend despite all of that being sound, the honest answer is that the system is not finding people it thinks will respond, and the fix is new creative or a wider audience, not a cleverer setting. Structural questions like that are what Meta ads consulting exists to work through.
