The Meta Andromeda update is the change to how Meta retrieves and matches ads to people, and as of 2026 it means your targeting happens inside the creative instead of inside the audience settings. Meta reads every creative in the account, sorts your audience into buckets and sub buckets, then serves each person the ad it judges most relevant to them. Lookalikes, interest stacks and manual audiences decide less every month.
We have run the new structure across every ecommerce account we manage. On one account the overall return on ad spend (ROAS) went from 3.83x to 4.71x and one campaign inside it went from 4.30x to 6.08x, with no change to the audience setup. On another we replaced four testing and scaling campaigns with one holding a single ad set and 32 active creatives, and ROAS went from 0.70 to 1.37 in five days on almost the same spend, the single CBO test that moved ROAS 94.84%. Here is what changed, how to use Andromeda step by step, how to read the account, and what we changed again in 2026.
TL;DR
- Targeting moved into the creative. The creative mix is the audience lever now, not the ad set.
- Fewer ad sets, more ads. Stack 15 to 50 creatives in one ad set so the learning aggregates in one place.
- Every awareness stage inside. Without unaware and problem aware ads you keep hitting the same warm pool.
- Build audiences, do not exclude them. Run broad, but set up the engaged and existing customer audiences.
- Read frequency, not the ROAS column. Ad level frequency since launch separates a winner from a credit taker.
- ABO came back as a standing tool. When one winner starves the account, force the test in ABO.
What the Meta Andromeda Update Actually Changed
Andromeda is the retrieval stage of Meta’s ads system, the step before any ranking or bidding. Meta’s own engineering write up describes it as narrowing tens of millions of ad candidates down to a few thousand, on a design that raised model capacity 10,000x and delivered a 6% recall improvement plus an 8% ads quality improvement on selected segments. Meta now considers far more creatives per impression and picks between them far more precisely.
Before, you told Meta who to reach. After Andromeda the order is reversed, and the ad set became a container where learning is pooled rather than a targeting instruction. Two consequences follow. Splitting creatives across ad sets works against you, because you are splitting the data Meta learns from. And creative diversification became the targeting lever: the more genuinely different your creatives are in angle, format and awareness stage, the more different people Meta can reach.
It also changed what a winning ad is. Meta does not scale a creative, it scales a sequence. Several ads work on the same person in an order and the purchase is credited to whichever one they touched last, so the ROAS column reports who closed, not who did the work. Meta has kept shipping in the same direction since, with the Generative Ads Model and the Adaptive Ranking Model, both reporting conversion lifts on Instagram. Less manual control over who sees the ad, more weight on what the ad is.
Meta Andromeda vs Standard Meta Ads Targeting
Andromeda is not a campaign type you pick in Ads Manager, it is the same auction behaving differently, so the real question is which old habits still earn their place.
| What decides it | Standard Meta ads setup | After Andromeda |
|---|---|---|
| Who sees the ad | Interests, lookalikes, exclusions | The creative, matched by retrieval |
| Creatives per ad set | 4 to 6, one angle each | 15 to 50, every awareness stage inside |
| Where learning sits | Split across every ad set you opened | Pooled in one ad set, from ad level up |
| How you judge an ad | The ROAS column | Ad level frequency and incremental ROAS |
| Testing | Separate campaign, winners duplicated out | Tested and scaled in one campaign |
The difference is worth real money. On one client we added creatives to the ad set that already held the data instead of opening a new one. It went from 7 ads to 11 and account ROAS went from 3.83x to 4.71x. On a second account we added 10 creatives to an ad set holding 3, and it went from 4.24x to 5.43x.
How to Use Meta Andromeda in Six Steps
This is the order we run on a new account, and the order matters. Skipping the first step makes every number in the last one unreliable.
1. Fix the purchase event first
Open Events Manager, select the data set for the brand and read the Event Match Quality score on the Purchase event. It should be above 9, and the Conversions API (CAPI) should be sending server side events alongside the pixel with the customer parameters filled in. A low score means Meta is learning from a biased sample of your buyers.
2. Collapse the account into one CBO
The structure we start from is one campaign on campaign budget optimization (CBO), one ad set per persona, and 15 to 50 ads inside each one. Broad placements, no custom exclusions, every awareness stage covered. Retargeting and retention campaigns become optional, because the single CBO does both inside itself.
The rule for opening a second ad set is not a creative count, it is the learning. Learning happens at the creative level and aggregates at the ad set level, so the 50 conversion threshold applies to the ad set. While an ad set is still learning limited, adding creatives helps. Once it is out of learning and performing, we leave it alone and the next batch goes into a new ad set.
3. Build the audiences, do not exclude them
We do not exclude existing customers or engaged audiences, but we build both so Meta can see them. Website visitors, add to cart and checkout initiated go into the engaged audience. Pixel purchases from the last 180 days plus a dynamic Klaviyo customer list go into the existing customer audience. Skip this and a broad ad set spends most of your money on people who would have bought anyway.
4. Stack creatives across all five awareness stages
Map every creative to unaware, problem aware, solution aware, product aware or most aware before it goes live, then check the mix. Twelve ads where ten are product aware scale to a point, then ROAS collapses the moment Meta reaches colder audiences. The fix is filling the gaps, not adding ads.
5. Match the landing page to the awareness stage
A testimonial or product benefit ad can go to the product page. A native or unaware ad needs a listicle or an advertorial, because the person clicking it is reading, not shopping yet.
6. Graduate winners by post ID
When an ad proves itself, duplicate the post ID into the bid cap or cost cap campaign so the engagement and social proof travel with it. A fresh upload of the same file starts at zero.
Creative Diversification Is the New Targeting
Diversification does not mean a new background or a different button colour. It means a different angle, concept, awareness stage and format. The more specific a creative is to one kind of person, the better Meta can find that person for you.
These are the formats holding spend across our accounts as of 2026. Statistic led unaware statics, the figure written large with the source underneath. Comic book drawings of the scene the customer is stuck in, paired with long form primary text. Native long copy ads, a phone native image plus 1,500 characters of story. Raw point of view live reaction UGC, no face, almost no text on screen. Recorded founder to customer calls, where the customer does not know the call is coming. Four panel process statics, where panel two covers what people already tried.
Three things hold across all of them. The problem takes most of the image and the product stays small. The person is called out broadly first and the story gets specific afterwards, because over specific persona hooks shrink the pool and push frequency up fast. And on young skewing audiences the base stays real, with AI applied on top of real footage.
Native long copy ads running right now
How to Read the Account After Andromeda
Open an account and you find one ad at 5.5x ROAS with three purchases next to one at 1.5x with a single purchase, and the 1.5x gets switched off. It is usually the wrong call. The 5.5x ad sits at the end of a sequence the 1.5x ad started, and killing the starter kills the finisher.
Frequency is the tell, read at the ad level and since that ad launched, not over a rolling seven day window. An ad set at 2.5 can be healthy, because people are seeing several different creatives. One ad at 2.5 is one creative repeating itself, and Meta starts charging you more for it.
| Ad level frequency | What it means | What we do |
|---|---|---|
| Below 1.5 | Reaching new people, room to grow | Keep it and scale it, this is the top of funnel |
| 1.5 to 2 | Mixed audience | Keep it, check it weekly |
| Above 2 | Talking to the same pool | Keep it for profit, never scale it |
The second read is the gap between reported ROAS and incremental ROAS at the ad level. A small gap means the ad is producing sales that would not have happened without it. A big gap means it is credited for purchases that were coming anyway, and the gap tracks frequency almost exactly. We have seen an ad show 6.89x in the default view and 0.81x on incremental attribution to new customers. Our post Andromeda analysis system has the column setup for this read.
Andromeda Review: What We Changed Again in 2026
The first version of this playbook was close to absolute: one CBO, consolidate everything, ABO only as a rare exception. A year of running it says otherwise. One CBO with one ad set per persona and 30 to 50 creatives is our default starting point as of 2026, not the answer for every account.
When one big winner holds most of the budget, Meta has little reason to move spend onto a new creative, so fresh tests sit at zero. We now launch those same creatives in an ABO (ad set budget optimization) campaign to force spend into them, accepting that we pay for the test. On one client those ad sets returned 4.32x on average on $1,371 of spend, and the winners graduated back into the CBO by post ID.
We also run bid cap and cost cap at the same time, with different jobs. Bid cap holds the proven top of funnel winners, 50 or more purchases and frequency below 1.5, duplicated by post ID. It is the maximum bid Meta enters the auction with, not a ceiling on cost per purchase, which is why a bid above your target cost per purchase can still work when the creative has strong engagement. Cost cap holds those winners plus the profitable middle and bottom of funnel ads.
A healthy account has 2 to 4 top of funnel creatives sharing the spend rather than one ad carrying everything, plus middle and bottom of funnel ads that look excellent in the ROAS column and hold little spend. Reading that second group as the winners keeps accounts stuck.
Setting up Andromeda campaigns for luxury brands
Luxury is the one category where we slow the consolidation down. The buying pool is smaller, so a broad ad set with 40 creatives pushes frequency up faster than on a supplement brand. We split campaigns by average order value (AOV) range so Meta does not dump budget on the cheapest line, weight the mix towards problem aware and solution aware rather than offer led, and read ad level frequency weekly. The full version is in the playbook we run for luxury brands post Andromeda, and Skin Design London is the UK luxury skincare account we built it on.
Related Results From Skaleit
We ran this structure on a jewelry brand and it produced +43% ROAS in 30 days while scaling. For the strategy behind the sections above, see the consolidated ad set strategy, the creative sequencing method, and the ads testing structure we use to choose between ABO and CBO.
FAQ
What is the Meta Andromeda update?
It is the change to how Meta retrieves and matches ads to people. Andromeda narrows tens of millions of eligible ads down to a few thousand candidates per impression, and Meta reports it raised retrieval recall by 6% and ads quality by 8% on selected segments. Targeting now happens through the creative rather than the audience settings.
How do you use Meta Andromeda in an account that is already running?
Check Event Match Quality on the Purchase event first, then collapse the ad sets into one CBO with one ad set per persona and move the creatives into the ad set that already holds the most data. Build the engaged and existing customer audiences instead of excluding them, fill the gaps in the awareness stage mix, and read ad level frequency before you turn anything off.
How do you set up Meta Andromeda campaigns for luxury brands?
Split campaigns by average order value range so Meta does not spend the budget on the cheapest line, keep creatives per ad set lower than on a mass market brand because the pool is smaller, weight the mix towards problem aware and solution aware, and read ad level frequency weekly. The consolidation still applies, it just moves slower.
Is Meta Andromeda better than standard Meta ads targeting for luxury products?
It is not a setting you switch between, it is how the auction works now, so a manual setup competes against it rather than instead of it. On luxury products, broad plus creative diversification beats interest stacks and lookalikes on reach and cost, but it needs tighter frequency discipline because the pool is small.
Is ABO dead after the Andromeda update?
No. As of 2026 ABO is a standing tool. When one winner starves the CBO we launch the new creatives in an ABO campaign to force spend into them, then graduate the winners back by post ID. Some accounts run better in ABO permanently.
Want This in Your Account?
If you are spending $30K or more a month on Meta and your ad sets are still split by angle, get in touch with us and we will look at your account before the first call.