Meta's Andromeda Update Broke Your Playbook — Here's the New One
If the last few months running Meta ads for your beauty brand have felt… off, you're not losing your touch. Something really did change.
Campaigns that used to scale like clockwork are now wobbly. Winners flame out in days instead of weeks. And if your team has been grinding through the usual fixes — swapping audiences, refreshing creative, nudging budgets — you've probably noticed that nothing actually sticks.
The problem isn't the creative. It isn't the offer. It's the structure underneath all of it.
Meta didn't just ship another update. With Andromeda, they fundamentally rewrote how ads get analyzed, distributed, and sequenced. The algorithm now reads your creative in full — visuals, copy, audio, pacing, the whole environment — and matches it to users based on real behavior, not interest targeting.
That breaks almost every playbook we grew up on. Most brands are still optimizing for a model that no longer exists.
At Pennock, we run paid media for beauty, skincare, and lifestyle DTC brands every day, and we've spent the last quarter rebuilding our approach from the ground up around Andromeda. Today's briefing unpacks what we've learned — and what you can put to work in your account this week.
On the menu
- The Andromeda traps most beauty accounts are stuck in
- The new operating system for testing and scaling
- Why your creative is your targeting now
What happens when you actually fix the system
Before we get into the traps, let's talk about what "fixing the system" looks like in real numbers.
One of our skincare clients came to us with the exact symptoms above — flat ROAS, rising CAC, creative that used to print money suddenly underperforming. No product changes. No offer changes. Just a full structural overhaul aligned to how Andromeda actually distributes ads.
Fourteen days in, compared to their previous 30-day baseline:
- +48%ROAS
- −39%CPA
- +15%Conversion rate
- −29%CPC
- +27%CTR
Nothing about the brand changed. The system did.
If you want to see how your own account stacks up under Andromeda, we're running free audits for a handful of beauty and lifestyle brands this month.
The four Andromeda traps most accounts are stuck in
When performance feels inconsistent right now, it's almost never one bad ad. It's a pattern — and we see the same four showing up across brands.
1. The high-frequency trap
Your ads are running. Conversions are coming in. ROAS even looks decent. But under the hood, the algorithm is serving the same creative to the same audience on repeat, and view-through conversions are papering over the lack of real reach.
The question to ask stops being "is this converting?" and becomes "are we actually reaching new people?"
The metric we track at Pennock is net new reach. The goal: roughly 80% of the people you reach each week should be net new, not repeats from the week before. If that number is low, your campaigns are recycling the same audience and growth will eventually stall — no matter how efficient the dashboard looks.
2. The volatility trap
You push more spend, and suddenly CAC spikes, ROAS swings day to day, and nothing feels predictable. One day you're a genius, the next day you're panicking in Slack.
This isn't random. It's what happens when the algorithm scales without guardrails. As spend increases, Meta explores new pockets of users — some convert beautifully, others don't, and the swings show up in your performance metrics.
Our fix is to give the algorithm constraints it has to respect. That means cost caps and target ROAS campaigns. Instead of efficiency bouncing around, spend bounces around. The system quietly pulls back on days it can't hit your targets and leans in on days it can.
The real question to answer here: what matters more to you — stable spend or stable efficiency? You can't fully have both.
3. The anchor ad trap
You look at an account and one ad is eating most of the budget. Reflex says: "This is inefficient, kill it."
Don't.
That high-spend ad — the anchor ad — is usually the one dragging new people into the system. It may not be the best converter in the account, but it's feeding the funnel. The "efficient" ads you're comparing it to? They're often just finishing the job on users the anchor already warmed up.
Turn off the anchor and those efficient ads suddenly aren't so efficient anymore. The solution isn't deletion — it's introducing new concepts strong enough to challenge the anchor. Let the algorithm decide when something better deserves the spend. If the anchor truly gets unprofitable, cap or isolate it. Don't amputate it.
4. The testing cannibalization trap
You launch five new creatives. One or two hoover up all the spend in 24 hours. They look bad, so you shut them off. The rest never got enough impressions to prove anything. Congratulations — you just "tested" nothing.
Meta optimizes on a 7-day click / 1-day view window. If you're making kill decisions at 48 hours, you're reacting to noise. Testing has to respect the timeframe the system is actually learning over, and it has to give every concept a fair shot at distribution.
That usually means controlling spend distribution in testing, using campaign structures that reduce algorithmic bias, and letting tests cook long enough to be statistically real. Otherwise, the system picks a favorite too early and everything else gets thrown out before it ever gets a chance.
The new operating system
Under Andromeda, performance is really a function of how well you structure your inputs so the algorithm can learn, sequence, and scale on its own. There are two connected halves to that system: a testing engine and a scaling engine.
The testing engine. We test in batches of 5–7 creatives at a time, each built around a distinct concept and persona. If the ads look too similar, Meta reads them as the same ad and just picks one — so variety isn't optional, it's the whole point. We give each ad roughly 2x target CPA in spend before making a call, and we lean on ABO campaigns during testing because they let us control spend distribution and give every concept a real shot.
The scaling engine. Winners graduate into CBO campaigns where the algorithm handles sequencing — which ad to show, to whom, in what order. That's where Meta's new sequential intelligence actually earns its keep. From there, we set outcome-based guardrails — target ROAS and cost caps — that tell the system what's acceptable, not how to find customers. If it can hit the target, it spends. If it can't, it pulls back. Volatility ends up in your spend, not your CAC.
The overall shape: fewer campaigns, fewer ad sets, more data flowing through each one. Two layers running in parallel — broad prospecting to feed the system, retargeting to capture the value. Both depend entirely on the quality of the creative you're putting in.
Creative is now the targeting
This is the shift most brands haven't internalized yet. Under Andromeda, you aren't really choosing the audience anymore. Your creative is. Meta reads the whole ad — visuals, copy, tone, setting — and decides who it's for, where it fits in the sequence, and when to serve it.
Creative is the targeting. Not metaphorically. Literally.
Build personas, not variations
The old way: take a winner, swap the headline, tweak the hook, try a new background color. Call it a test.
That doesn't work anymore. The algorithm looks at the whole ad, decides two variants are essentially the same thing, picks one, and ignores the rest. You think you ran a test; you actually ran a coin flip.
What works now is building distinct personas with genuinely different angles. For a skincare brand, that might look like:
- The ingredient nerd who wants the clinical breakdown
- The sensitive-skin sufferer who's been burned by every "gentle" product on the shelf
- The lifestyle buyer who wants the routine, the aesthetic, and the glow
- The derm-adjacent skeptic who needs authority and data
Each one is a completely different doorway into your brand. One leads with science. One leads with empathy. One leads with aspiration. Each tells the algorithm something specific about who this ad is for — and that's what lets Meta go find more of those people.
The visual has to match the angle
Messaging alone isn't enough, because the algorithm is reading the whole creative environment. If your visual contradicts your message, you weaken the signal and Meta gets confused about who to serve it to.
- A clinical angle needs clean visuals, white space, clinical cues, authority figures.
- A lifestyle angle needs real environments, movement, and natural light.
- A premium angle needs elevated settings and styling that reinforce status.
Make it visually obvious what the ad is and who it's for. The algorithm isn't guessing anymore — it's reading.
The mindset shift: testing under Andromeda looks more like exploration than refinement. You're mapping directions, not polishing one idea. The more distinct concepts and personas the system has to work with, the more it can find for you.
Sum it up
Andromeda fundamentally changed how Meta analyzes and distributes ads, which means your old optimization playbook is now actively working against you.
- Audit your structure first. Most of the pain isn't in the creative — it's in the four traps above.
- Let spend fluctuate, not efficiency. Use cost caps and target ROAS to keep CAC steady while the algorithm finds its way.
- Build for distinct personas, not variations. Creative is your targeting now. Treat it that way.
If you want a second set of eyes on your account, we're offering free Andromeda audits to a small group of beauty and lifestyle brands this month.
See you next time. — The Pennock team