Meta Andromeda is Meta's upgraded ad retrieval engine, the stage that scans millions of ads and shortlists the best ones for each person before ranking picks the winner. Meta Engineering detailed it in December 2024. The big shift for advertisers: your creative is your targeting now. Andromeda matches creatives to people, not audiences to ads, so the winning move in 2026 is one broad campaign fed with a big library of genuinely different creatives. This guide gives you the structure and the creative system to win, from V8 Media, the team behind R2+ billion in client sales.
The rules changed and nobody sent you an email
If your Meta Ads performance feels like it is auditioning for a soapie, you are not imagining things. The rules changed, quietly.
Meta Andromeda is the rebuild underneath Facebook and Instagram ads that makes your old targeting habits look like strategy from 2017. The punchline is simple. Your creative is your targeting now.
If you have been obsessing over exclusions and micro-segments, you have been polishing the door handle while the building moved. Andromeda rewards variety, not duplicates. It prefers broad pools, then lets the creative signal do the sorting. It does not need you to find the audience. It needs enough distinct creative angles to find the audience for you.
What Meta Andromeda actually is
Andromeda is Meta's upgraded ad retrieval engine, explained publicly by Meta Engineering in December 2024. Retrieval is the step where Meta shortlists which ads even get a chance to compete for a person, before ranking decides the final winner. Meta's own engineering post describes it scanning tens of millions of active ads and narrowing them to thousands of candidates for each person in milliseconds. Analysts like Jon Loomer have unpacked what that shift means for advertisers.
Here is the part that changes your job. Andromeda uses computer vision and AI to read your creative. The visuals. The hook. The on-screen text. Even the audio. It assigns each ad a kind of identity, then matches it to the people most likely to respond. It is matching creatives to users, not audiences to ads.
Meta built this because the number of ad creatives exploded. Automation like Advantage+ pushed volume up. The system needed to evaluate far more options, faster. The polite version is "smarter indexing". The honest version is "we can test more ads than you can micromanage".
Why creative is the new targeting
Nielsen has measured creative quality at over half of an ad's impact, and Meta's own research agrees. The data has said the quiet part out loud for years. Creative drives more than half your results. Not targeting. Not budget. Creative. That is not a motivational poster. It is a warning label.
Andromeda pushes that into your daily operations. Targeting is broad now. So the platform leans on your creative for signal. Your visuals, your spoken words, your on-screen text, your headline, your first line of copy all become classification signals. The transcript of your video matters. The product in use matters. The person holding it matters.
Your creative becomes a map. Meta reads it and decides "this is skincare for sensitive skin" or "this is a first-time-buyer offer", finds the responders, and shifts spend to what converts. That is why duplicates do not count as variety. Meta needs distinct angles, not the same ad wearing different socks. Keep changing the background while repeating the hook and you give Andromeda déjà vu. It responds by ignoring you, the way South Africans ignore loadshedding schedules that never match reality.
Old way vs the Andromeda way
| Old playbook | Andromeda | |
|---|---|---|
| You control | Audiences, interests, exclusions | Creative angles and offers |
| Structure | Many ad sets, split funnels | One broad campaign, one ad set |
| What wins | Finding the right audience | Feeding enough distinct creative |
| Biggest lever | Targeting | Creative variety |
Why your old winning ad suddenly died
You had a winner. It ran for months. Then it tanked, and you changed nothing.
That is Andromeda. When retrieval changed, the pool it pulls from changed too. Your one hero creative now competes inside a faster, broader search. One ad cannot cover every pocket of buyers. It never could. The old system just hid that from you.
The fix is not to hunt for another single winner. It is to stop relying on one. Feed the machine ten strong angles and it finds ten pockets you never could. Lose one to fatigue and the other nine carry you. One hero ad is a single point of failure. A creative library is resilience.
The new structure: one campaign, one ad set, many ads
The Andromeda-friendly structure is boring on purpose. One campaign with campaign budget optimisation. One ad set with broad settings. And a chunky creative library at the ad level. The simplicity is the strategy.
Why does it work? Because consolidation helps the algorithm learn faster. One place, not thin data spread across silos. Split your budget across top, middle, and bottom funnel campaigns and you starve each of learning signal. Then you call Meta "unstable" while you are the one chopping the learning up like biltong.
How many creatives? Industry analyses point to 15 to 30 genuinely different creatives live at once for accounts seeing strong results, with a practical floor of around 6. For a mid-sized SA brand spending R10,000 to R20,000 a month, a working starting point is 12 to 16 distinct creatives. The exact number is not the point. Enough variety for Andromeda to find pockets is the point.
This is also why constant tiny tweaks hurt you. Tinker daily and you reset the learning every time. Launch a batch. Let it run. Kill the losers. Plan the next batch. That is it.
What "variety" actually looks like
Variety means different angles, not cosmetic edits. If you can describe two ads with the same sentence, you made duplicates. Tools and analyses, like Segwise, now even talk about a creative similarity score. Assets that are too alike, above roughly 60% similar, get suppressed in retrieval. So aim for genuinely distinct concepts. Here is what Meta can read as distinct signals:
- A demo solving the problem in the first 3 seconds.
- A founder explaining "why we made this" with a strong before and after.
- A customer testimonial with captions and one clear metric of change.
- A price-anchor offer, like a bundle, free-shipping threshold, or limited drop.
- An objection crusher, like "works on sensitive skin" or "fits plus-size comfortably".
How to set up an Andromeda-friendly campaign
Let us make this painfully practical. Your setup must stop fighting the system. Yes, it feels like less control. That is the point.
- Build for automation at the campaign level. Pick an objective that matches revenue, usually Sales with the pixel and conversion events tracked properly. Use campaign budget optimisation so Meta can move spend between ads without you playing traffic cop.
- Keep the ad set broad. Minimal restrictions. Only restrict where genuinely required, like gender-specific products or legal compliance. Stop stacking interests like a Jenga tower in a moving taxi.
- Load the ad level with real variety. Start with 10 to 16 distinct creatives on a modest budget, scaling to 20+ as production allows. Mix UGC, product demos, statics, carousels, and short vertical video. Each one a unique angle, not a cloned script.
- Kill non-spenders fast. If an ad consistently does not spend, Andromeda cannot find a responsive pocket for it. That is feedback, not rejection. Pause it. Replace it with a fresh angle.
- Batch your next creative drop. Replace losers continuously, but add major new batches on a steady cadence, every two to three weeks. That stabilises learning while expanding the search space.
Best practices for winning the creative-volume game
The brands that win under Andromeda do not just make more ads. They build a system that makes more learning. Otherwise you burn cash faster than Eskom burns diesel and call it "testing".
Build an angle library, not a mood board
A mood board is vibes. An angle library is money. Pull real customer questions, objections, and buying triggers from your inbox, WhatsApp, reviews, and support tickets. Turn them into themes. Then produce variations per theme. Problem-first hooks. Proof-first hooks. Objection-first hooks. Identity and offer hooks. That is variety without chaos.
Make the first 3 seconds earn attention
Meta has pushed "hook early" for years, and Andromeda raises the stakes. A weak opening is a weak signal, and the system learns "people do not care". Bold on-screen text, fast context, visible product use early. If you need 12 seconds to explain the product, the creative is the problem, not the market.
Measure like a business, not a hobby
Watch spend distribution, CTR, cost per purchase, and ROAS. But read them in context. If one ad gets spend and another does not, Meta is voting with money. Do not argue with the vote. Replace the weak signal with a stronger one.
Stop over-excluding, then acting surprised
Exclude too aggressively and you shrink the pool and the system's ability to find new pockets. Decide what matters most. Need new customers? Track new-customer acquisition cost and optimise toward it. Control the metric, not the illusion.
Benchmarks and reality checks
Numbers shift by vertical. But in most eCommerce accounts, a CTR of 1% to 2% is your sanity baseline. Stronger creative pushes it higher. Accounts that adopt the consolidated, creative-diverse structure? Industry reports often cite 20% to 35% better ROAS than the old way. Take that as direction, not gospel. The point is not the exact stat. Creative has always been the biggest lever. Andromeda just stopped hiding that.
This is exactly the system we build when we run client Meta Ads. For more on the fundamentals, see our guides on lead generation on Meta, scaling Meta Ads profitably, and why your Facebook ads get bad leads.
Frequently asked questions
What is Meta Andromeda?
Meta Andromeda is Meta's upgraded ad retrieval engine, detailed by Meta Engineering in December 2024. It shortlists which ads get a chance to be shown to each person, matching creatives to users rather than audiences to ads.
How does Andromeda change Meta Ads strategy?
It makes creative your main lever instead of targeting. The winning setup is one broad campaign with one ad set and a large library of genuinely different creatives, so the system can find pockets of buyers for you.
How many creatives do I need for Andromeda?
Aim for 15 to 30 genuinely different creatives for strong accounts, with a floor of about 6. A mid-sized SA brand can start with 12 to 16 distinct angles and scale as production allows.
Does Andromeda mean targeting is dead?
Largely, yes, for the old interest-stacking approach. Keep targeting broad and let creative do the sorting. Only restrict where genuinely required, like legal compliance or gender-specific products.
How often should I refresh creative under Andromeda?
Replace non-spenders continuously, and add a meaningful new batch every two to three weeks. Constant daily tweaks reset the learning and hurt performance.
Key takeaways
- Andromeda is Meta's retrieval engine (Meta Engineering, Dec 2024). Creative is the new targeting.
- It matches creatives to people, so it reads your visuals, audio, and copy as signals.
- Winning structure: one broad campaign, one ad set, a big library of distinct creatives.
- Aim for 15 to 30 genuinely different creatives. Duplicates do not count as variety.
- Build an angle library, hook in the first 3 seconds, kill non-spenders, batch new drops every 2 to 3 weeks.
