Meta Ads reporting
Meta Andromeda: what changed and how many creatives to run
Andromeda is the system that picks which ads compete for each person. Here is why creative now does the targeting, and how to check your variety is working.
Also available in: Português · Español
You launch eight ads in one ad set. A week later, one of them has eaten nearly the whole budget and the other seven have a few hundred impressions between them. That is not a bug. It is Meta's delivery system making choices, and the first stage of that choice has a name. Meta Andromeda is the ad retrieval system that decides which ads even get into the running for each person before the auction. Because of it, creative now does much of the work targeting used to do. The practical takeaway: variety beats volume, and the proof that it works lives in your ad-level report.
What Andromeda actually is
Every time someone opens Facebook or Instagram, Meta has to pick one ad out of millions. It does that in stages. The first stage, retrieval, takes tens of millions of eligible ads and pulls out a few thousand candidates. Ranking models then decide what actually shows.
Andromeda runs that first stage. Meta introduced it on its engineering blog in December 2024, and in March 2025 tied it directly to creative diversification on Meta for Business: as advertisers upload more creatives, the system can run more complex models to match the right creative to each person. The figure Meta shared, an 8% improvement in ads quality, is Meta's own number, not an independent measurement.
What that means for a media buyer: if an ad does not survive retrieval, it never reaches the auction. And what helps it survive is looking relevant to someone, which depends far more on the message and format than on an interest you ticked in the ad set.
Why creative became the targeting
With a broad audience and Advantage+ on, the ad set says very little about who will see the ad. The ad itself does. A founder-to-camera video pulls one kind of person. A price-led carousel pulls another. A product demo pulls a third. The system learns who responds to each and finds more people like them.
So "variety" does not mean the same image in four colors. It means different angles:
- A different problem: the pain each ad speaks to.
- A different person: who appears or who is talking.
- A different format: short video, static, carousel, testimonial.
- A different promise: price, speed, safety, social proof.
Four ads that each change one of those gives the system more real choices than twelve headline tweaks.
How many ads per ad set
Meta does not publish an ideal number of ads per ad set, and any exact number you read online is the author's opinion. What the Help Center does say points the other way from "more is better": in its article on Learning limited, it lists running too many ads at the same time among the reasons an ad set fails to reach roughly 50 optimization events in a week. Every new ad splits the same budget and the same conversion volume.
A rule that holds up in practice:
| Situation | What to do |
|---|---|
| Small budget, few conversions a week | A few ads, very different from each other |
| Healthy budget, ad set out of learning | Add one new angle at a time |
| One ad has taken most of the spend for weeks | Test a new angle against it, not a variation of it |
| Ads with zero spend for 14 days | Pause them; they only muddy the read |
Mind the learning phase too: adding an ad is an edit that can send the ad set back into learning. Launch in batches, not one a day. The details are in the Meta Ads learning phase.
How to read whether variety is working
The ad set view hides the answer. Switch to the ad level, same date range for every ad, and look at four numbers.
1. Share of spend
What slice of the ad set budget went to each ad? If one ad takes almost everything and the rest sit still, the system picked a winner. That is normal. The problem is when the others never get enough delivery for you to know whether they would have worked.
2. Frequency per ad
Frequency climbing on the winning ad while cost per result climbs with it is the classic sign of creative fatigue. With real variety, the system has somewhere to move spend when the winner tires. Without it, the whole ad set gets worse together.
3. CPM per ad
Different ads reach different people, and that shows up in CPM. If every ad has nearly the same CPM, they may be talking to the same crowd, which means they are less different than they look. A much higher CPM on one ad can mean it is reaching a more contested audience.
4. Cost per result per ad
This is the deciding number. But only compare ads with enough spend to have a few results; an ad with US$ 15 spent and one conversion tells you nothing. When you can join it to your CRM, check revenue by creative too, because the ad with the cheapest lead is not always the one that closes. Revenue by creative walks through that read.
How to measure before and after a creative strategy change
If you are moving from "one ad set per interest" to "a few broad ad sets with varied creative," measure it so the comparison is fair:
- Write down the baseline. Cost per result, CPM, frequency and result volume for the last 4 weeks, at the campaign level.
- Change one thing. Either the structure or the creative. Both at once hides the cause.
- Wait for learning to finish. A three-day read after a big change is noise.
- Compare like with like. Four weeks against four weeks, ideally with no holiday or sale in either window.
- Look at the whole funnel. If cost per lead dropped and sales did not follow, variety bought you volume, not customers.
If your question is when to replace the ad that has worn out, creative fatigue: when to swap lists the signals in order.
What Andromeda does not change
- Broken tracking stays broken. If your pixel and Conversions API send the wrong event, the system learns the wrong thing faster.
- A weak offer stays weak. Creative variety does not rescue a price the market will not pay.
- The reporting is still yours. Meta decides who sees what; it does not tell you whether that turned into revenue.
Your next step
Open your main campaign at the ad level, last 14 days, and answer one question: how many ads got real spend? If the answer is one, your next creative is not a variation of it. It is a new angle. To run this read across several accounts without opening each one, see the plans: from US$ 31.20 a month on the annual plan, with the first 7 days for US$ 2.99.
Frequently asked questions
What is Meta Andromeda?
It is Meta's ad retrieval system: the stage that narrows tens of millions of possible ads down to a few thousand candidates for each person, before the final ranking decides what shows.
How many ads per ad set should I run after Andromeda?
Meta does not publish a fixed number. Start with a few genuinely different creatives and add more only if the budget can feed them; the Help Center lists running too many ads at once as one cause of Learning limited.
Do I need to change my targeting because of Andromeda?
Not necessarily. What changes is how much the creative matters: with broad audiences, the ad itself sorts who sees what. Very narrow targeting limits the options the system has to choose from.
How do I know if creative diversification is working?
Look at the ad level, not the ad set: share of spend per creative, frequency, CPM and cost per result. If one ad takes almost all the budget and the rest never get going, you have one creative, not variety.