Meta Andromeda, the retrieval layer in Meta's ad system, filters millions of ads down to a short list before any auction starts. If your creative doesn't signal relevance to the right person, your bid never fires.
The campaign looks right. The offer is solid. The budget is approved. And spend just sits there, flat, like a bobber on a dead lake.
Most marketers assume the ad lost the auction. Usually it never got in.
Here's what's actually happening. Meta rebuilt the front end of its ad system. It's called Andromeda, and its job is retrieval: before any auction runs, it pulls a short list of the ads that look most relevant to the person scrolling. Meta's CFO, Susan Li, described it on the July 2025 earnings call as the stage "where we select the few thousand most relevant ads from tens of millions of potential candidates[1]." Meta says the 2025 Andromeda update drove nearly 4%[2] higher conversions on Facebook Feed and Reels. Your bid only matters if you make that list.
So the question isn't "how much should we bid?" It's "is our creative giving Meta a reason to pull it for this person?"
That changes how you build ads. Here are the five things we change first when Untethered Office takes over a Meta account.
Are your best ads fighting each other?
Your best ads fight each other when they share a fingerprint. Meta assigns every ad one from its visual, copy, on-screen text, and audio, and when two match, it picks one and benches the rest.
Meta reads everything in your ad: the visual, the copy, the on-screen text, the audio. It turns all of that into a kind of fingerprint, a location on a map of what the ad means and who it's for. And it doesn't read it once. It re-checks that fit against every person, every scroll.
Here's the trap. You find a winner, so you make ten versions of it. New headline. Different button color. Same video with a new first frame.
To you, that's testing. To Meta, that's the same ad ten times. Same fingerprint, same spot on the map. So it picks one and benches the rest.
That's not a scaling problem. That's you outbidding yourself for the same tiny corner of the market while the rest of it goes unreached.
Scale comes from spread. Ads that are actually different land in different places, in front of different people.
Before you duplicate a winner again, pull your last ten ads and count how many would actually look different to Meta.
Why can't you test your way through everything?
Good creative judgment is pattern recognition built from doing the work, not from testing everything. You pick an angle, test it on purpose, and read what comes back.
The usual answer to "we need more variety" is a testing grid. Every segment, every visual style, every awareness stage.
Run the math. Twenty segments, fifteen visual styles, five awareness stages. That's 1,500 combinations before anyone writes a line of copy. No budget we've ever seen survives that.
Think about bass fishing. The guy who grids the whole lake, one cast every ten feet, eventually finds fish, but it takes all day and most of his gas. The guy who catches them reads structure: weed edges, drop-offs, where the bait goes when the water warms up. He's using everything he learned on every trip before this one.
That's creative judgment: pattern recognition from doing the work. It sounds like this: "This buyer, at this stage, needs this angle." Then you test that, on purpose. Write the angle down before you launch, so you know what you're reading when the results come back.
What is awareness, and why does it change the ad?
Awareness isn't a personality type. It's where a specific person is standing on the road to buying, right now.
We talk about "unaware" and "product aware" buyers like they're personality types. They're not. It's where a person is on the road to buying, right now. The same contractor can be unaware in March and ready to switch yards in June.
Every ad has one job: meet the person where they're standing and move them one step forward.
Here's how that looks for a lumberyard selling to builders and remodelers:
- Unaware. Show them a problem they've stopped noticing. "Your crew lost 40 minutes this morning waiting on a delivery nobody confirmed."
- Problem aware. Name the pain they already feel. "Short loads are why the schedule slipped. Not your framers."
- Solution aware. Prove the approach is right. Why a dedicated yard rep beats calling a pro desk and hoping.
- Product aware. Win the comparison. In-house truss design and package pricing, against the yard they use now.
- Most aware. Remove the last bit of friction. "Contractor account approved in 24 hours. First delivery this week."
If every ad you run is a product-aware "here's our yard" ad, you're fishing one small, crowded pond. To grow, you need an ad for every stop on the road. Sort your current ads by the five stages, then build the next one for the stage that's empty.
How do you build real creative variety?
You build real creative variety by changing what the ad looks like, not just what it says. Turn one dial at a time.
Copy persuades. The visual tells Meta, and the person, who the ad is for. And creative carries more weight than most teams think: NCSolutions research puts it at 49%[3] of the incremental sales from advertising, while marketers guess it's closer to 20%[4]. So when you want a new fingerprint, change what the ad looks like, not just what it says.
A useful way to do that is four dials:
- Format: what the ad is pretending to be. A jobsite interview, a yard walk-through, a text-message screenshot, a customer testimonial.
- Face: who's on camera. The yard manager, an outside sales rep, a builder who buys from you every week.
- Style: how it's made. Phone video, polished production, a simple animated cutaway of a wall section.
- Setting: where it happens. The yard, the truss shop, a framed house, the cab of a truck.
Hold three dials. Turn one.
Say your yard walk-through with the manager is working. Keep the script. Reshoot it on a jobsite with a repeat builder telling the same story. Same message you already proved, brand new fingerprint, new set of people. That's not a tweak. That's a new ad.
Pick your best ad, pick one dial, and reshoot it this week.
Where do the best segments actually hide?
The best segments aren't demographic labels. They're the reason behind the purchase, stacked two layers deep.
Most teams slice their market by what people buy or by demographics. Builders. Remodelers. Homeowners.
The better cuts are in the reason behind the purchase: what someone believes, who they see themselves as, what situation they're in, what they value, and what just happened to them.
The real move is stacking two of those together.
Take a custom builder doing four to six homes a year (that's who they are) whose old yard just got bought out and whose rep left (that's what just happened). That person doesn't need "Quality lumber, fast delivery." They need "Your yard got bought. Your rep is gone. Here's who picks up the phone now."
Nobody else is running that ad. And that's a repeat buyer, the customer a yard can count on and build pricing around, not a one-off homeowner pricing out a deck.
One test keeps this honest: a stack only counts if it needs a different ad. If a custom builder and a production builder would respond to the exact same message, you haven't found a new segment. You've found a new label.
To find your stacks, list your last ten good customers and write down what had just happened to each of them before they called. Any situation that shows up twice is worth its own ad.
How will you know any of this worked?
You'll know it worked when cost per lead drops and those leads become real accounts, not when click-through rate goes up.
None of this matters if you judge it by click-through rate. Meta grades its own homework, then hands you an A. A comScore and Pretarget study of display ads found the correlation between clicks and conversions was 0.01[5], about as close to zero as it gets.
Before you launch:
- Label every ad with the awareness stage it targets and the one dial you changed. That turns "we ran some new creative" into a test you can actually read.
- Judge on cost per lead, not engagement. Then check which leads became real accounts in your CRM. That's the only honest scoreboard.
- Expect trends, not perfect counts. Cookie consent, ad blockers, and modeled conversions mean the numbers are directionally right, not exact. Direction is enough to decide what to make next.
Then pick your next test, label it before it goes live, and check the CRM a month later to see which leads turned into accounts.
Bottom line
Making ads is cheap now. Mark Zuckerberg told investors in October 2025 that advertisers will increasingly be able to hand Meta a business objective and a credit card and "have the AI system basically figure out everything else that's necessary, including generating video or different types of creative that might resonate with different people[6]." Maybe. But choosing what to make is still the job, and that part stays with the person who knows the customer.
Here's where to start this week:
- Pull your last ten ads and count how many actually look different.
- Sort them by awareness stage and find the stage with nothing running.
- Build one new ad for that stage by turning a single dial.
- Label it, launch it, and judge it on cost per lead and CRM accounts.
Every round makes the next one better. The ads that win tell you more about your buyers, and your judgment about where the fish are gets sharper each time.
If your Meta spend is flat and you can't tell whether it's the creative, the targeting, or the tracking, that's exactly what the Untethered Office Account Teardown is for.
Frequently Asked Questions
What is Meta Andromeda, and why does it block my ads before the auction?
Meta Andromeda is Meta's ad retrieval system. Before any auction runs, it filters millions of ads down to a short list of a few thousand that look relevant to the person scrolling. Your bid only matters if your ad makes that list. Flat spend is usually a retrieval problem, not a bidding problem. If the creative doesn't signal relevance, the auction never sees it. To make the list more often, give Meta ads that look genuinely different from each other, not ten versions of one winner.
How many creative variations do I actually need?
A full testing grid breaks down fast. Twenty segments, fifteen visual styles, and five awareness stages produce 1,500 combinations before anyone writes a copy line. The practical answer is to build ads that are genuinely different by turning one creative dial at a time, so each one carries a different fingerprint and reaches a different set of people. Start with your best ad and change one dial: format, face, style, or setting.
Why is click-through rate the wrong way to judge Meta creative?
Meta grades its own homework, then hands you an A. Click-through rate tells you Meta approved the ad, not that it produced a lead worth money. Judge on cost per lead, then check whether those leads converted to real accounts in your CRM. In 2026, cookie consent and modeled conversions mean the numbers are directionally right, not exact. Report on two numbers: cost per lead and leads that became accounts.
What makes a psychographic stack more useful than a demographic segment?
A demographic segment labels who someone is: builder, remodeler, homeowner. A psychographic stack adds the reason behind the purchase: their situation, beliefs, or what just happened to them. Stack two layers and you get an ad nobody else is running. The test: if two "different" segments would respond to the same message, you don't have two segments. You have one with two labels. To find a real one, ask your best customers what had just happened when they first called you.
Related Reading
Sources
- Susan Li, CFO, Meta. Q2 2025 earnings call transcript, July 2025.
- Meta. Q2 2025 earnings call transcript: conversion lift on Facebook mobile Feed and Reels from Andromeda improvements, July 2025.
- Westwood One. Marketers Vastly Understate the Sales Effect of Creative (NCSolutions research), April 2024.
- Westwood One. Marketers Vastly Understate the Sales Effect of Creative (Advertiser Perceptions survey of marketers), April 2024.
- MarTech. Study: For Display Ads, Clicks Have Nearly Zero Correlation With Conversion (comScore and Pretarget), 2012.
- Mark Zuckerberg, CEO, Meta. Q3 2025 earnings call transcript, October 2025.