Andromeda made Meta's ad delivery smarter at finding purchases. It didn't make it any better at finding profit. Here's what changed, what Meta has actually published, and how ecommerce brands keep margin, stock and measurement in their own hands.
On this page
- What is Meta Andromeda?
- What changed for Meta ads after the Andromeda update
- The profit problem: Meta optimises for what it can see
- The inventory problem: Advantage+ doesn't know what's in the warehouse
- How to measure Meta ads in the Andromeda era
- Your 30-day Meta Andromeda checklist
- The bottom line on Meta Andromeda
Meta Andromeda is the ad retrieval engine behind Facebook and Instagram ads, the system that decides which of the tens of millions of eligible ads even get considered for each person, and since 2025 it has quietly rewritten how ecommerce brands should run Meta ads. Fewer ad sets, broader targeting, far more creative variety, and more of the account handed to Advantage+.
What most explainers miss is the part that matters to a founder: Meta's AI optimises for the purchases and revenue it can see, not the margin you keep. Give it a discounted bestseller and a full-price bundle and it will scale the one with the higher ROAS, even if it makes you less per order.
Prefer to watch? At our DTC Builders NYC panel, operators from Koio, Thesis and Prospa talked through what Andromeda changed for Q4 creative, landing pages and allowable CAC.
What is Meta Andromeda?
Meta Andromeda is Meta's personalised ads retrieval system: the first stage of ad delivery, where Meta narrows "tens of millions of ad candidates into a few thousand relevant ad candidates" before its ranking models pick the ad you see. Meta Engineering announced it on 2 December 2024.
In plain English, Andromeda doesn't pick the winning ad, it builds the shortlist. If your ad doesn't make the shortlist for a person, it never reaches the auction for them, however good your bid. Meta says the system lifted recall by 6% and ads quality by 8% on selected segments.
Andromeda, GEM and Lattice: Meta's ad AI stack
| System | What it does | What Meta has reported |
|---|---|---|
| Andromeda | Retrieval: shortlists a few thousand ads from tens of millions | 10,000x model complexity, +8% ads quality |
| Lattice | Ranking architecture: fewer, larger models that learn across objectives and placements | Almost 12% higher ad quality, up to 6% more conversions |
| GEM | Generative Ads Recommendation Model, a foundation model that teaches Meta's ranking models | +5% ad conversions on Instagram and +3% on Facebook Feed in Q2 2025 |
| Sequence learning | Reads the order of actions a person takes before and after seeing an ad | +3% conversions in tests on selected segments |
Meta's GEM post and an August 2026 engineering post credit its newer models with a 6% lift in Instagram conversions. Those are Meta's figures, not independent tests.
What changed for Meta ads after the Andromeda update
The shortlist is now built per person, by models that read the ad itself. That changes four things.
1. Creative diversity beats creative volume
In April 2025 Meta published The creative advantage: unlocking the power of diversification with Meta Andromeda. Its core advice: run a wide range of ads with different themes, messages and visuals, because creative is now how the system finds the right people. Meta's example brand, Dribbleup, went from 3–4 new creatives a week to almost 50 with steady performance.
The word that matters is different. Ads with the same creator, setting and a new headline get treated as one idea, so 10 near-identical ads behave like 1. Meta hasn't published a similarity threshold, but the rule holds: change the motivator, the person, the format or the setting, not just the copy.
What counts as a new concept for an ecommerce brand:
- A different motivator: sleep, confidence, saving time, gifting, status, value
- A different person: customer, founder, expert, creator in a different setting
- A different format: UGC video, static comparison, carousel, catalog, founder talking head
- A different stage: problem-first for cold traffic, proof and reviews for people who already know you
2. Fewer ad sets, broader targeting
When the model does the matching, slicing budget across 12 interest ad sets mostly slices its data. Advantage+ audience treats your audience inputs as suggestions, so the common structure is 1 sales campaign with 1–2 broad ad sets per country or offer.
3. Advantage+ is now the default, not the option
Advantage+ shopping campaigns are now Advantage+ sales campaigns, with catalog ads running inside every one. Meta says the streamlined setup cut cost per action by about 9% on average, and its Marketing API v25.0 (February 2026) made Advantage+ the default for sales campaigns. If your agency still builds manual campaigns, ask why.
Advantage+ creative can also generate backgrounds, text and formats from your assets, and Meta reports about 11% higher click-through rates. Use it for variety, but review the output: the AI doesn't know which claims you can make or which products you can afford to discount.
4. The landing page is part of the ad
This was the strongest point on our panel. Edward Weston put it bluntly: "Andromeda is reading your landing pages. It is matching it back to your creative." When the ad promises one benefit and the page sells another, conversion drops, so brief the page and the creative from the same concept.
The profit problem: Meta optimises for what it can see
Here's the part the agency guides skip. Andromeda, GEM and Advantage+ optimise towards the conversion and the value you send Meta, which for most Shopify stores is revenue. Meta never sees product cost, shipping, fees, returns or discounts, so contribution margin, the money left after those costs (see our contribution margin formula guide), is invisible to it.
So the better Meta's AI gets at finding purchases, the faster it can scale the wrong ones:
- Low-margin bestsellers, because cheap, popular products convert easily
- Discounted orders, because 25% off converts better than full price and revenue-based ROAS still looks fine
- Customers who never come back, because a first purchase is a first purchase to Meta whether that customer is worth £30 or £300
- Existing customers, because people who already love you are the easiest purchases to find
The fix is to give Meta better instructions and judge it on your numbers. Start with the break-even ROAS for each product line, because a 3.0 ROAS is great on a 63% margin product and a loss on a 30% one. Then work through five levers.
1. Optimise for value, not just purchases
Meta's value optimisation and the Highest value bid strategy chase purchase value rather than count, and a ROAS goal sets a floor. Set that floor from your break-even ROAS, and remember Meta's ROAS is revenue-based: 1.0 is break-even on revenue, not profit.
2. Send a margin-aware value (carefully)
Some brands pass a margin-adjusted purchase value through the Conversions API, so Meta optimises towards gross profit. It works, but it changes what every ROAS number in Ads Manager means, so label it and keep a revenue event for reporting. If tracking isn't solid yet, start with our server-side tracking guide.
3. Use value rules for customers worth more
Meta's value rules let you bid more for segments such as age, location or placement. Meta's own example raises bids 60% for a segment with 60% higher lifetime value (Social Media Today's rundown). Only use them where you have real customer lifetime value data by segment.
4. Split product sets and catalogs by margin
Catalog ads push whatever converts. Use product sets to keep low-margin, discounted or clearance SKUs out of prospecting, and give high-margin lines their own product set so they get budget on purpose. Meta's catalog fields support custom labels, so tag each SKU with a margin band and filter on it.
5. Watch new versus returning customers
Advantage+ sales campaigns find existing customers very efficiently, which flatters ROAS and does little for growth. Define existing customers in Ads Manager, use the new versus existing breakdowns, and judge prospecting on new-customer CAC. If first-order economics are weak, fix the offer, not the targeting (see profitable new customer offers and how and when to discount).
The inventory problem: Advantage+ doesn't know what's in the warehouse
Meta knows whether a catalog item is in stock, and out-of-stock items don't appear in ads. It doesn't know you have 40 units left and 6 weeks until the next delivery. That creates two expensive patterns:
- Spending into a stockout. Your best-converting SKU soaks up prospecting budget until it sells out, the winning creative goes dark, and you paid full acquisition cost for the last units you could have sold to returning customers.
- Ignoring overstock. High-margin stock sitting in the warehouse gets little spend because it converts slightly worse than the bestseller, until it becomes dead stock.
The fix is a weekly link between stock and spend: pull prospecting (or exclude the SKU from product sets) when days of cover drop below your reorder lead time, and push overstocked, high-margin lines with their own creative. Our inventory alerts guide covers the thresholds, and the free daily inventory actions skill in our AI Skills Library turns that into a short daily list.
How to measure Meta ads in the Andromeda era
When Meta's AI decides who sees what, its own dashboard becomes a less reliable judge, and Meta-reported ROAS can over-credit Advantage+ and view-through conversions. Measure it against your P&L instead.
| Metric | What it tells you | Why it matters now |
|---|---|---|
| MER (net sales ÷ total ad spend) | Whether total spend is growing the business | Doesn't depend on Meta's attribution |
| New-customer CAC | What a genuinely new customer costs | Stops Advantage+ hiding behind returning buyers |
| Contribution margin after marketing | Profit left after product, shipping, fees and ads | The number Meta can't see |
| Creative ranked by profit | Which ads make money, not just sales | Andromeda rewards variety, so you'll have more ads to judge |
| Change history | Who changed what, and when | Meta also makes automated changes you didn't ask for |
New to the first one? Read MER vs ROAS, and use the break-even MER calculator to see the MER your margins need. Track CAC payback period next to CAC.
Rank creative by profit, not ROAS
With 15 concepts live instead of 4, ROAS will mislead you on discounted or low-margin creative. StoreHero's Creative Overview puts every Meta ad next to the profit it made after product cost, shipping and fees, and marks it scale, keep or cut: a sale reel at 3.9 ROAS can lose €640 while a UGC video at 4.6 ROAS makes €2,860. Fíor Jewellery used it to scale ad spend 600% while growing contribution margin over 1,000%.
If you use Claude or ChatGPT, the free Meta Ads waste finder prompt measures Meta spend against blended CPA and new-customer contribution margin instead of Meta-attributed ROAS.
Keep a change history
Andromeda-era accounts change constantly, and not all of it is your team: Advantage+ applies automated changes and accepted recommendations move budgets. When contribution margin drops, ask "what changed?" Our Google and Meta change history guide and the ad account change history audit line up every edit against the profit move after it.
Your 30-day Meta Andromeda checklist
- Week 1, measure. Work out break-even ROAS for each product line, set a baseline for MER, new-customer CAC and contribution margin after marketing, and start reviewing change history weekly.
- Week 2, protect margin. Take low-margin, discounted and low-stock SKUs out of prospecting product sets, tag the catalog by margin band, and test value optimisation with a ROAS goal set from break-even.
- Week 3, rebuild creative. Count live concepts, not live ads. Brief 5–8 new motivators, people or formats, each with a matching landing page.
- Week 4, consolidate. Merge overlapping ad sets into 1–2 broad ones, rank every creative by profit, and scale, keep or cut on margin.
The bottom line on Meta Andromeda
Andromeda is good news for brands that give Meta genuinely different ads and judge the results on profit. Feed it variety and better values, keep it away from stock you can't afford to discount, and measure it on contribution margin, not on the ROAS it reports about itself.
StoreHero ranks every Meta creative by profit after product cost, shipping and fees, tracks MER and new-customer CAC daily, and flags the ads to pull before a product sells out.
Key takeaways
- Meta Andromeda is ad retrieval: it shortlists a few thousand ads from tens of millions for each person before ranking picks the winner.
- Creative diversity is now the main targeting lever. Different motivators, people, formats and settings count; copy tweaks on the same shoot don't.
- Run fewer, broader ad sets and expect Advantage+ to be the default setup for sales campaigns.
- Meta optimises for the revenue it can see, so it can scale low-margin products, discounted orders and returning customers. Set ROAS floors from break-even and split product sets by margin.
- Link stock to spend: pull ads before a sell-out and push overstocked, high-margin lines.
- Judge Meta on MER, new-customer CAC, contribution margin after marketing and creative ranked by profit, and keep a change history.
Frequently asked questions
What is Meta Andromeda?
Meta Andromeda is Meta's personalised ads retrieval engine, announced by Meta Engineering in December 2024. It is the first stage of ad delivery on Facebook and Instagram: it narrows tens of millions of eligible ads down to a few thousand candidates for each person, and Meta's ranking models then choose which ad is shown.
When did Meta Andromeda launch?
Meta announced Andromeda on 2 December 2024 and rolled it out through 2025. Many advertisers and agencies say they felt the biggest shift in delivery and creative testing in late 2025.
How does Meta Andromeda work?
Andromeda uses much larger machine learning models (Meta says 10,000x more model capacity) running on NVIDIA Grace Hopper and Meta's MTIA chips to match each person with a shortlist of ads. It reads signals from the ads themselves, which is why creative diversity now does much of the work that audience targeting used to do. Ranking systems such as Lattice and the GEM foundation model then pick the final ad.
What does post-Andromeda mean?
Post-Andromeda describes the way Meta ads are run since the update: fewer and broader ad sets, Advantage+ campaigns as the default, and far more genuinely different creative concepts per campaign, because the system uses creative variety to find the right people.
How do you activate Andromeda in Meta ads?
You don't. Andromeda runs automatically for every advertiser as part of Meta's ad delivery system. What you control is what it has to work with: broad targeting, consolidated campaigns, a wide range of distinct creative, accurate conversion values and clean product sets.
How many ads should you run per ad set after Andromeda?
Meta doesn't publish a number. A common working range is 8 to 15 genuinely distinct concepts per campaign, but the count matters less than the variety: ten ads with the same creator, setting and new headlines behave like one idea.
Is Andromeda the same as Advantage+?
No. Advantage+ is the suite of automated campaign tools advertisers choose in Ads Manager, such as Advantage+ sales campaigns, Advantage+ audience and Advantage+ creative. Andromeda is the retrieval system underneath Meta's ad delivery, built partly to handle the much larger number of ads that Advantage+ and generative AI create.
Sources
- Meta Engineering: Meta Andromeda, next-gen personalized ads retrieval engine (Dec 2024)
- Meta for Business: AI innovation in Meta's ads ranking (Mar 2025)
- Meta for Business: The creative advantage, diversification with Meta Andromeda (Apr 2025)
- Meta Engineering: Meta's Generative Ads Model (GEM) (Nov 2025)
- Meta Engineering: A multi-stage architecture for Meta's ads ranking (Aug 2026)
- Meta for Developers: Graph API and Marketing API v25.0 (Feb 2026)
- Meta: Advantage+ sales campaigns
- Meta: Catalog ads (formerly Advantage+ catalog ads)
- Meta for Developers: Catalog fields reference
- Social Media Today: Meta value rules rundown (Aug 2025)
- DTC Builders NYC panel: Meta's Andromeda is reading your landing pages (StoreHero, Oct 2026)
