Meta Andromeda Update: What Changed for Ecommerce Ads

Everybody's blaming Andromeda.

Your cost per purchase jumped. One ad is eating 80% of the budget and the other 11 are starving. And every agency newsletter in your inbox says the same word.

Here's the short answer. The Meta Andromeda update replaced the first filter in Meta's ad system, the step that decides which ads even get a shot at a person. That filter now reads your creative (the image, the video, the words) to work out who the ad is for. So creative diversity does the targeting work your audiences used to do. Ten ads that look the same get treated like one ad. Five ads built for five different buyers get five chances to find people.

But Andromeda only decides who clicks. Your product page decides whether that click pays. That's the half nobody in the ad-buyer newsletters talks about, and it's where the money is.

What is the Meta Andromeda update, in plain English?

Meta's own engineers wrote it up in December 2024. Their post calls Andromeda a "personalized ads retrieval engine," and retrieval is the first stage of a multi-stage system. It takes tens of millions of ads and cuts them down to a few thousand candidates for each person, every time they open Facebook or Instagram. The later ranking models only ever see what Andromeda lets through.

The numbers Meta published in that engineering post on the Andromeda retrieval engine: a 10,000x increase in retrieval model complexity, +6% recall, and +8% ads quality on the segments they tested.

Then in December 2025, Meta's business team connected the dots for advertisers. In its guide to creative diversification, Meta says its ads system "analyzes creatives and groups together ads that share key visual and thematic attributes," and that learnings are shared at the creative level, not just the ad level.

Read that twice. If your ads look alike to the machine, they get pooled. Your 12 ads become maybe three ideas competing for the same people.

Andromeda doesn't care how many ads you uploaded. It cares how many different ideas it can see.

What has Meta actually said, and what are advertisers reporting?

Most Andromeda content blends what Meta confirmed with what some agency saw in three accounts. Keep them apart.

Claim	Who says it	How solid it is
Retrieval narrows tens of millions of ads to a few thousand per person	Meta engineering, Dec 2024	Confirmed by Meta
10,000x retrieval model complexity, +6% recall, +8% ads quality	Meta engineering, Dec 2024	Confirmed by Meta, on tested segments
The system groups ads with similar visual and thematic traits	Meta for Business, Dec 2025	Confirmed by Meta
Distinct concepts beat small tweaks of one concept	Meta for Business, Dec 2025	Meta's recommendation
Ad lifespan shrank from 6 to 8 weeks down to 2 to 4 weeks	Confect study of 3,014 ecommerce advertisers	Vendor data with a stated method
Narrow interest targeting is now mostly redundant	Agencies and tool vendors	Widely reported, not a Meta statement
Budget increases now spike costs faster	Founders in forums	Anecdotal

Confect sells creative tools, so it has a horse in this race. But it published its method: 3,014 ecommerce advertisers, $834 million in spend and 115.7 billion impressions across 2025. You can read their Andromeda study here.

The founder complaints are real, too. One Shopify merchant on the forums scaled an Advantage+ shopping campaign from $2,000 a day to $5,000 a day and watched their cost per thousand impressions nearly double. Was that Andromeda, or the budget jump? Nobody could prove it either way. That's the honest state of most Andromeda stories.

Why does creative diversity now do the targeting?

Here's the thing. Before, you told Meta who to find: women 28 to 45, interested in yoga, lookalike of past buyers. The ad was the message. The audience was the aim.

Now the ad is both.

Say you sell a cooling bamboo sheet set. Under the old model you'd run one angle ("sleeps cooler than cotton") at five hand-built audiences. Under Andromeda, the move is the reverse. One broad audience, five angles that each describe a different buyer:

• The hot sleeper who kicks the covers off at night
• The menopause buyer who's tried three "cooling" sheets already
• The couple where one runs hot and one runs cold
• The gift buyer who wants something that feels expensive
• The eco buyer who cares what the fabric is made of

Each of those ads tells the system something different about who should see it. The system goes and finds those people. That's the whole mechanism, and it's why Meta keeps saying diversification instead of iteration. Same photo with a new headline is iteration. A new person, a new problem, a new scene is diversification.

I'll say the unpopular part. I'd rather run five ads that disagree with each other than 40 ads that agree. The 40 get pooled into one idea. The five get five different crowds.

So why is my account still bleeding after I fixed the creative?

This is the part that burns founders.

You do everything right. Five concepts, broad audience, Advantage+ on. Andromeda goes out and finds five different buyers, just like Meta promised. Then all five land on the same product page. And that page was written for one of them.

The hot sleeper lands and sees "400 thread count, deep pocket fit." The menopause buyer sees the same thing. The gift buyer sees no gift box, no "arrives in 3 days," nothing. Five promises. One page that keeps only one of them.

Your ads got smarter. Your page didn't. Now you're paying Meta to find people your page can't close.

Andromeda decides who clicks. The page decides whether the click pays. Only one of those is on your side of the screen.

If you want the tactical version of fixing that handoff, I wrote a whole breakdown on writing a Shopify product page for Facebook ads traffic. The short version: the first screen has to finish the sentence the ad started, for every angle you run.

How much can you afford to pay per click after Andromeda?

Here's the math. And this is the number that settles most Andromeda arguments.

Your page sets your ceiling. Run it on a store like this (a hypothetical). Conversion rate 1.2%, average order value $70. That means revenue per visitor is $0.84. On 10,000 clicks, that's $8,400.

At a 60% gross margin, each visitor earns you $0.50 in gross profit. So $0.50 is the most you can pay per click and break even. Pay $0.90 per click on those 10,000 clicks and you've spent $9,000 to make $8,400 in revenue. Your return on ad spend is 0.93. You're losing money before a single product ships.

Now watch what happens when the page gets fixed. Same traffic, same ads. Conversion rate 2.4%, average order value $85. Revenue per visitor is $2.04. On the same 10,000 clicks, that's $20,400.

At the same 60% margin, each visitor earns $1.22 in gross profit. The same $0.90 click that was killing you is now comfortably profitable. Return on ad spend: 2.27.

Page as-is (1.2% x $70) $8,400

Page rebuilt (2.4% x $85) $20,400

Conversion rate x average order value x 10,000 clicks. Same ads, same $0.90 cost per click.

Nothing about Andromeda changed between those two scenarios. The only thing that moved was what each person was worth after they landed. That's why I tell founders to stop asking "what's my cost per click?" and start asking "what's my revenue per visitor?" There's a longer explanation of that shift in revenue per visitor vs earnings per click.

Plug in your own numbers below. Your cost per click is only expensive relative to what the page earns.

What does this look like with real numbers?

The math above is hypothetical. This part isn't.

A bedding brand we worked with had a cooling bamboo sheet page converting at 1.0% with a $125 average order value. That's $1.25 per visitor, or $12,500 on 10,000 visitors. After the page rebuild, that product ran at a 4.3% conversion rate and a $254 average order value. Revenue per visitor: $10.92. On the same 10,000 visitors, that's $109,200. See the full case study numbers. Real client numbers, not typical results, and not a promise of what your store will do.

At $1.25 per visitor, almost any click price is scary. At $10.92, you can outbid most of your category and still keep margin. The page is what lets you afford the auction.

This is also why more ad spend doesn't fix a conversion problem. Scaling a $0.84 page from $2,000 a day to $5,000 a day just buys you more of a loss, faster. Paid social is only one channel, so it helps to see where it fits in a full ecommerce traffic strategy.

What should you change first?

Two moves, in this order.

First, audit your creative for sameness. Line up your active ads. If you can't name the different buyer each one is for, they're probably getting pooled. Cut to your strongest three to five concepts and make each one about a different person with a different problem. If you run Advantage+ sales campaigns, this matters even more, because the campaign hands Meta almost all of the targeting.

Second, write down the promise each concept makes, and check that your product page keeps every one of them in the first screen. A headline or a short block per buyer type is usually enough. If you're sending cold traffic, this guide to product pages for cold traffic covers the trust gaps they bring with them.

AI now makes both jobs faster. Meta says over a million advertisers used its generative AI tools to make more than 15 million ads in a single month. The page side matters more, though, because it's the part your competitors still ignore.

Frequently asked questions

What is the Meta Andromeda update? It's the retrieval engine that picks which ads even get considered for each person. Meta says it narrows tens of millions of ads to a few thousand candidates, with a 10,000x jump in model complexity.

Does Andromeda make audience targeting useless? Much weaker, not useless. Meta groups ads by their visual and thematic traits, so your creative now tells the system who the ad is for.

How many creatives do I need? Meta hasn't published a number. Five truly different concepts beat 40 small variations.

Why did my Facebook ads get worse after the Andromeda update? Usually pooled look-alike ads, faster creative wear, or a page that only speaks to one of the buyers your ads now attract.

Does the Andromeda algorithm affect my product page? It changes who lands there. Your page decides whether they buy, and your revenue per visitor sets the most you can pay per click.

What to do next

Take your best-selling product. Pull last month's conversion rate and average order value for it, multiply them, then multiply by your gross margin. That's your break-even cost per click. If it's below what Meta charges you right now, no creative strategy will save the account. The page is the fix.

---

Book Your Profit Audit

Andromeda is getting better at finding your buyers every month, and every one of them lands on a page that either keeps the ad's promise or wastes the click. Book a profit audit and we'll show you what each Meta click is worth on your store today, then rebuild a high-converting product sales page in less than 15 minutes.

Book Your Profit Audit →

Or go here to check it out → revenueflows.ai

P.S. Andromeda can find a thousand new buyers for you by Friday. It can't make a $0.84 page worth a $0.90 click.