Meta changed how it picks which ads a person sees, and most accounts are still built for the old system. Here's what Andromeda is, what it rewards, and how to plan creative around it.
Every account we audit comes with a familiar setup. A handful of ad sets, a few audiences, and a creative library that looks big until you read it. Twelve ads, one idea. Same promise, same person, same picture, with a new headline or a new button color. On paper that's a testing program. To Meta's delivery system, it's mostly one ad.
What Andromeda actually is
Andromeda is Meta's ad retrieval system. Retrieval is the first step in deciding which ads a person sees. Before any ranking or auction happens, the system has to narrow the field. Meta says Andromeda takes tens of millions of candidate ads and narrows them to a few thousand relevant ones, and that it was built to handle the growing volume of ads coming from Advantage+ automation and generative AI creative tools.
That's the official part. Here's the practical part. If your ad doesn't make it through retrieval for a given person, nothing downstream matters. Your bid, your audience settings and your campaign structure only get a vote after the system has decided your ad is in the running.
Why your creative is doing the targeting
Paired with broad targeting and Advantage+, Meta now makes most of the decisions a media buyer used to make: who sees what, where, and when. The main thing it has to go on is the ad itself. What it shows, who's in it, what it promises, how it opens. In practice, your creative is doing the targeting now.
That has a consequence nobody put in the release notes. If the platform makes the targeting decisions for everyone, targeting stops being where advertisers differ. The difference moved into the creative. In the accounts we audit, the gap between the best and worst ad inside one campaign is often wider than that campaign's swings from month to month. Same audience, same budget, same landing page. The difference is the ad.
Variations vs. diversification
A variation is the same concept with a surface change: a new color, headline, crop or button. What we see in practice is that variations tend to share delivery with the original. They don't buy you new reach, because to a system matching ads to people, they look like the same ad.
Diversification means running concepts that are genuinely different from each other. A different persona, a different reason to buy, a different format, a different hook. Each one is built to reach a group of people your other ads aren't reaching.
A quick test: describe the ad in one sentence without mentioning colors, fonts or exact wording. If two ads get the same sentence, they're one concept.
The Concept Matrix
We plan creative as a grid, not a mood board. Three axes:
- Persona. Who's buying, what situation they're in, and what they already believe.
- Angle. The desire or pain, the reason to act now, and the objection to beat.
- Format. UGC, static, demo, founder, comparison or meme, plus the hook in the first second.
Every new ad fills a specific cell. A new parent who wants to save time, shown in a 20-second demo, is a different cell from a skeptic who needs proof, shown in a comparison static. Before anything gets produced, we can point to the cell it fills and the people it's meant to reach. When a cell wins, we build around it. When a whole row loses, we stop making it.
The matrix also shows you what's missing. Most accounts we audit have one or two personas covered heavily and the rest of the grid empty.
How many concepts can your budget support?
There's no magic number, and anyone who gives you one without seeing your spend is guessing. Every concept needs enough spend to produce a readable answer, and what counts as readable depends on your cost per purchase and how much signal you need before you trust a result.
The logic is simple, even if the math is specific to your account. Start from your monthly spend and your typical cost per purchase. Decide how many conversions a concept needs before you'll make a call on it. That tells you how many concepts you can actually read in a month. A bigger budget can feed more concepts at once. A small budget spread across too many starves all of them, and you end up with a dozen results that don't mean anything.
So we size the matrix to your spend and your margin, and grow it as the account grows. Fewer, sharper concepts beat a long list you can't read.
What to measure: the creative, not the channel
Channel-level ROAS averages across exactly the dimension where the variance now lives. If one concept is carrying a campaign and three are dragging it down, the campaign number can look fine and still tell you nothing about what to make next.
So report at the creative level. For each concept: what it spent, what it earned at contribution margin (after COGS, shipping and fees), how fast it fatigues, and which cell of the matrix it came from. Keep channel reporting for pacing and the board deck. Make decisions from the concept view.
Two numbers are worth watching every week: how many genuinely distinct concepts are live right now, and how many of your recent winners came from new cells rather than new versions of old ones.
The part that costs you something
This approach is slower to show a win than pumping out variations. Distinct concepts take more thinking and more production, and creative-level results need enough volume per concept to be readable. We'd rather tell you a result isn't readable yet than hand you a number that feels good and falls apart under a real question.
The alternative is paying for twelve versions of one idea and wondering why your reach stalled.
Want the full picture of how we run this? See Creative-Led Performance, the HUNT Framework, and our earlier piece on Meta automating your ads.
How many real concepts are you running?
Book a free creative teardown. We'll look at your live ads together and count concepts versus variations.
Book a free teardown →