- Andromeda is Meta's ad retrieval system: it picks a shortlist of a few thousand ads from tens of millions before ranking decides which one you see. Meta described it publicly in December 2024.
- Because retrieval now reads the ad itself, your creative does most of the targeting. One winning ad reaches one kind of buyer and saturates that pocket faster than it used to.
- Creative diversity means different ideas, not different edits. A new hook on the same angle reaches the same people; a new problem, persona or format reaches new ones.
- Scale by adding distinct concepts in batches and judging them on new-customer CAC from Shopify, not by pushing budget behind one ad until frequency climbs and CAC jumps.
- Andromeda cannot fix a weak offer, a slow page or a thin margin. It makes those problems show up faster.
In this article
For four months, one ad carried the account. A founder-shot video, a clear problem, a strong hook, and a cost per new customer comfortably under target. The brand did what every playbook says: find the winner, then pour budget into it. Daily spend went from $1,500 to $4,000.
Within five weeks, the winner stopped winning. Frequency in prospecting crept past three. CPMs rose. New-customer CAC went from $38 to $71. The team tried duplicating the ad, new thumbnails, new first lines of copy, a new audience. Nothing moved it back. The conclusion in the room was that Meta had "got worse".
Meta had not got worse. It had changed how it decides who sees an ad, and the account was still built for the old system. This article explains what changed, why a single great ad now saturates faster, and how to build a creative engine that scales in 2026.
What Andromeda actually is
Every time someone opens Instagram or Facebook, Meta has to choose one ad from tens of millions that are eligible to run. It does that in two broad steps. First, retrieval builds a shortlist of a few thousand ads that might be relevant to this person. Then ranking scores that shortlist and the auction picks a winner.
Andromeda is the new retrieval engine. Meta's engineering team described it in December 2024, reporting a 10,000-fold increase in model capacity at the retrieval stage, better recall and better ad quality on the segments it was tested on. The important part for advertisers is not the hardware. It is what the system now pays attention to.
Why the creative is now the targeting
A more capable retrieval system can understand far more about each ad: what it shows, who appears in it, what problem it talks about, how it is edited. It matches that against what it knows about each person. With broad targeting, which is now the default for most sales campaigns, the ad itself becomes the main signal Meta uses to decide who should see it.
That is powerful when it works. A video of a new mother using your product will be retrieved for people who look like new mothers. A video about joint pain after running will be retrieved for runners. You no longer need to build those audiences by hand.
It also has a cost. One ad now finds one pocket of people, and it finds them efficiently. Once that pocket has seen the ad a few times, there is nowhere new for it to go. Pushing more budget behind it does not reach new people. It shows the same ad more often to the same people, and to slightly worse matches at the edges. That is the curve the brand in the opening ran into.
One winning ad scaled from $1,500 to $4,000 a day: weekly new-customer CAC and prospecting frequency (illustrative)
Illustrative composite. Frequency shown x10 to share the axis (1.4 rising to 3.2).
| Label | New-customer CAC ($) | Frequency x 10 |
|---|---|---|
| Wk 1 | 38 | 14 |
| Wk 2 | 41 | 17 |
| Wk 3 | 47 | 21 |
| Wk 4 | 55 | 25 |
| Wk 5 | 64 | 29 |
| Wk 6 | 71 | 32 |
Creative diversity means different ideas, not different edits
When a winner fades, the instinct is to make more of it: the same video with a new hook, a new thumbnail, a new caption, a 9:16 cut. Those are edits. To the retrieval system they look like the same ad, and they get retrieved for the same people. They can help with fatigue a little, but they do not open new audiences.
A new concept changes at least one of four things that decide who the ad is for.
| Lever | What changes | Example for a sleep supplement |
|---|---|---|
| Problem | The pain or desire the ad talks about | Waking at 3am vs jet lag vs a racing mind before a big day |
| Persona | Who is in the ad and who it speaks to | New parent vs shift nurse vs frequent flyer vs retiree |
| Format | How the message is delivered | Founder talking to camera vs ingredient demo vs side-by-side comparison vs static review |
| Offer or product | What you are asking people to buy | Single bottle vs 3-month bundle vs trial pack |
Change one lever and you usually reach a different pocket. Change only the hook or thumbnail and you usually reach the same one.
How to structure the account for retrieval
The structure that suits Andromeda is simpler than most accounts we audit. The goal is to give the system enough different ideas to work with, and enough budget per campaign to learn which idea suits which person.
- Fewer campaigns, broader targeting. One or two prospecting campaigns with broad audiences usually beat eight narrow ones. Keep existing-customer exclusions in prospecting.
- More concepts per campaign. Aim for somewhere between 8 and 20 distinct concepts live across your main prospecting campaigns, depending on spend.
- Add in batches, do not replace everything. Introduce three to five new concepts every two to four weeks. Turning off all the old ads at once resets what the system has learned.
- Let losers lose quietly. Meta will push spend to the concepts that work and starve the rest. You do not need to kill an ad the day it underperforms.
New-customer CAC at similar spend by number of distinct concepts live in prospecting (illustrative, same brand over three quarters)
Illustrative composite. Concepts counted as distinct problem, persona, format or offer, not edits.
| Label | New-customer CAC |
|---|---|
| 2 concepts | $68 |
| 6 concepts | $52 |
| 12 concepts | $44 |
Judge creatives on the right number
More concepts means more decisions about which ones to keep. Two traps are common. The first is judging on CTR, which rewards curiosity rather than buying intent; we covered that in high CTR, low sales. The second is judging on Meta-reported ROAS, which can be inflated by existing customers and changed definition in March 2026, when click-through attribution started counting only link clicks.
| Metric | Where it comes from | Why it matters |
|---|---|---|
| New-customer CAC by concept | Shopify first orders matched to the ad via UTMs | The cost of growth, not of reselling to existing customers |
| First-order value and product mix | Shopify | Some concepts attract buyers of your cheapest product |
| Contribution per order | Shopify plus your cost sheet | A concept that sells the low-margin product can look good on ROAS and lose money |
| 30 to 60-day repeat by concept | Shopify cohorts | Concepts attract different kinds of customer, not just different numbers |
| Frequency and CPM trend | Ads Manager | Early warning that a concept is reaching the end of its pocket |
The winner that stopped winning
- What the dashboard said
- One founder video drove most new customers for four months. Scaling it from $1,500 to $4,000 a day pushed new-customer CAC from $38 to $71.
- What was actually happening
- Thirty edits of the same video had been launched in six weeks. All were retrieved for roughly the same audience: women 35 to 50 with poor sleep after having children. Other groups that bought the product organically, shift workers and frequent travellers, had never been shown an ad made for them.
- Decision
- Test Brief four new concepts around different problems and personas, add them in two batches, cap the original ad's share of spend, and judge each concept on Shopify new-customer CAC over three weeks.
When a winner fades, do not make more of it. Make something for someone else.
What Andromeda cannot fix
Better retrieval finds better matches for the ads you give it. It cannot make a weak offer strong, a slow page fast, or a thin-margin product profitable. In fact, it makes those problems visible faster, because the system finds your best audiences quickly and then has nowhere else to go. If new concepts do not move CAC, look at the offer, the landing page and each product's break-even ROAS before briefing more creative. And check which product your ads lead with: your best seller might be a dead end.
A checklist for this month
- List every live prospecting ad and write its one-sentence concept. Count how many are truly distinct.
- Check frequency, CPM and new-customer CAC trends on your top three ads over six weeks.
- Look at your post-purchase survey or reviews for customer types and problems your ads never mention.
- Brief three to five new concepts, each changing at least one lever: problem, persona, format or offer.
- Consolidate campaigns where budget is spread thin, and keep existing customers excluded from prospecting.
Andromeda rewards brands that understand who their different buyers are and why each of them buys. That understanding lives in your order data, reviews and survey answers, not in Ads Manager. Connecting the two is what we do in a Blindspot Snapshot.
Questions founders ask
What is Meta Andromeda?
Andromeda is the retrieval stage of Meta's ad delivery system. It narrows tens of millions of eligible ads down to a few thousand candidates for each impression, which the ranking models then score. Meta's engineering team described it in December 2024, reporting higher recall and ad quality after deployment.
Did Andromeda make Meta ads more expensive?
Not directly. What changed is that accounts relying on one or two winning ads hit diminishing returns sooner, because those ads reach a narrower set of people. That shows up as rising CPMs, higher frequency and higher CAC on the same creative.
How many creatives should a DTC brand run?
There is no magic number, but most growing accounts we see do best with somewhere between 8 and 20 genuinely different concepts live across their main campaigns, refreshed in small batches every two to four weeks. Ten edits of one idea count as one concept.
Should I still use detailed targeting and many ad sets?
For most DTC brands spending under a few hundred thousand dollars a month, fewer campaigns with broad targeting and more creative variety work better. Splitting budget into many small ad sets slows learning. Exclusions for existing customers are still worth keeping in prospecting.
How do I know if my ad is saturated?
Watch frequency, CPM and new-customer CAC together on your top ad over four to six weeks. If frequency rises past about 2.5 to 3 in prospecting while CAC climbs and CTR falls, the ad has reached most of the people it will convert. More budget will buy expensive customers.
