Surviving Meta’s Andromeda Update: Your Action Plan

Meta’s new ad engine, Andromeda, is already shaking things up.

And if your CPMs or CPAs have been acting weird lately, this update could be the reason.

Andromeda changes how Meta’s system selects and delivers ads — shifting more control to machine learning and less to manual targeting.

In short, the old playbook doesn’t hit the same anymore.

Ahead: The deep insights + action plan you need to survive and thrive in this new era.

Andromeda Explained

First, What Is Andromeda?

Andromeda is Meta's new ad delivery engine.

Think of it as the brain that decides which ad to show to which person.

The old brain followed strict rules. The new brain plays the odds—it's constantly learning, adapting, and testing at massive scale.

What Actually Changed

The algorithm went from deterministic (rule-based) to probabilistic (best-guess).

Old way (Deterministic):

New way (Probabilistic):

Why this matters:

Your job isn't to find "the one winning ad" anymore:

Why Your Old Playbook Died

What's Dead:

What Works Now:

The shift is simple: The old system rewarded you for finding one perfect ad. The new system rewards you for giving it multiple strong options.

One winner worked when the machine was simple. Now the machine is smart, and it's hungry for choices.

What Most Advertisers Got Wrong

When Andromeda dropped, we watched the same mistake play out everywhere.

The narrative went: "The algorithm is smarter now, so just launch 100 ads and let it figure it out."

Sounds logical, right? It's not.

Think of it like this: If we gave you 100 white t-shirts with slightly different neck cuts, you still just have... a bunch of white t-shirts. But if we gave you 5 completely different outfits—a suit, gym gear, casual streetwear, beachwear, and formal attire—now you've got real options for different occasions.

That's the difference between creative volume and creative diversity.

Here's what actually happens when you dump 100 similar ads into an account:

You didn't give the algorithm diverse options. You gave it 100 versions of the same thing and asked it to pick.

Meta doesn't need more ads.

It needs more different ads—ones that target different personas, address different pain points, and are visually distinct.

Your Action Plan

STEP 1: Audit Your Current Structure

Before you change anything, you need to know what you're working with.

What to Review

Pull up your Meta Ads Manager and answer these questions:

Identify Your Winners

Now dig into performance. Which 3-6 ads have been consistently delivering over the last 30+ days?

Ask yourself:

Calculate Your Baseline

You need numbers to measure against:

STEP 2: Restructure for Consolidation

Alright, you've audited. You know what's broken. Now let's fix it.

The goal here is simple: Give the algorithm clean, concentrated data instead of fragmented garbage.

Account Structure

There are a few ways to restructure this depending on your budget and experience, but here's the setup we recommend for most:

├── Main Campaign
├── Ad Set 1: Proven Winners (80-90% budget)
│ ├── Winner Ad 1
│ ├── Winner Ad 2
│ ├── Winner Ad 3
│ ├── Winner Ad 4
│ └── Winner Ad 5
│
└── Ad Set 2: Testing (10-20% budget)
├── New Test Concept 1
├── New Test Concept 2
└── New Test Concept 3

Ad Set 1: Proven Winners

This ad set contains 3-6 ads maximum.

These are your proven, diverse concepts—the ones that actually work:

Budget allocation: 80-90% budget

This is your revenue engine. It's predictable & scalable.

Ad Set 2: Testing

This ad set contains 3-15 new concepts you're actively testing.

Kill protocol:

Budget allocation: 10-20% of your total spend.

This ad set exists to find improvements for your weakest link in the Winners ad set.

Budget Tier Adaptations

How you organize your Testing ad set also depends on how much you're spending:

More budget = more room to isolate variables. Less budget = keep everything tighter.

Targeting

Keep it stupid simple. Here's what your targeting should look like:

In short: Let your creative do the targeting.

STEP 3: Test Strategically

You've got the structure. Now let's talk about what actually goes inside it.

Like we said, Andromeda doesn't just want more ads. It wants different ads.

And there's a specific way to create "different" that the algorithm actually recognizes.

The 3-Second Rule (This Is Critical For Video Ads)

Meta considers video ads "the same" if the first 3 seconds look similar—even if your copy, CTA, and messaging are completely different.

Let me say that again: You can change every word in your video, but if the opening visual looks the same, Meta groups them as identical and throttles delivery.

What Meta flags as "similar":

What Meta sees as "different":

If someone scrolled past your ad on mute, would they think they'd seen it before? If yes—it's too similar.

The Persona-First Framework

Instead of making 20 versions of the same message, identify 4-5 distinct customer types who buy for different reasons.

Example: For a skincare product, you might have: One cares about speed (fast results). One cares about price (affordable solution). One wants proof it works (testimonials/science). One needs natural ingredients (clean beauty).

Create ads that speak to each motivation—with visually different settings and messaging that feels tailored to that specific audience.

Same product. Different reasons to buy.

What to Test (In Order of Impact)

1. Visual diversity (Biggest lever)

2. Angles: Different pain points or benefits (speed vs. cost vs. results)

3. Formats: Video vs. static vs. carousel—different people consume differently

4. Offers: "BOGO" vs. "50% off" vs. "Free shipping"

5. Messaging variations: Headlines, body copy, CTAs

Start at the top & test ONE thing at a time.

STEP 4: Scale & Optimize

You've restructured. You're testing diverse concepts. Now comes the hard part: Not touching anything.

Let me walk you through the actual timeline.

Days 1-3: Launch & Monitor

You're only checking for technical issues:

That's it. Don't look at CPA. Don't look at ROAS. The algorithm is still figuring things out.

Days 4-7: Initial Health Check

Now you can start looking at early signals.

Key metrics:

Outbound CTR (oCTR): Should be ≥ 0.8%

CPM vs. account average: Should be ≤ 1.5x your baseline

Frequency: Should be < 2.0

Kill protocol (only if BOTH are true):

Days 8-14: Performance Evaluation

Now you can judge based on actual conversions.

What you're looking for:

Don't kill ads with low spend if they have good efficiency. The algorithm might be holding them for specific micro-audiences.

Kill ads with high impressions but terrible CTR. Those are confirmed duds.

Day 15+: Time to Scale

Your ads are validated. Performance is stable. Now you can pour gas on the fire.

Standard Scaling Protocol:

If ROAS is at or above target:

Aggressive Scaling (Under $1k/day spend):

When you're spending less than $1k daily, the dollar risk is minimal.

When to STOP Scaling:

Hit the brakes immediately if you see:

These signals mean you've hit your campaign's ceiling. Don't force it.

Maintain current spend and focus on creative refresh instead.

Final Thoughts

Meta will change again. It always does.

And here's what we've noticed after a long time in this game: The tactics change. The principle doesn't.

Brands that thrive long-term aren't the ones who panic and chase every new playbook.

They're the ones who understand their numbers, build adaptable systems, and stay calm when everyone else panics.

Remember, you're not in this alone. We'll keep breaking down what actually matters, so you can focus on what works.

Milo Tran

Content Marketer at TrueProfit & eCommerce Operator

I’m part of the TrueProfit content team, but before that, I’m an eCom nerd with 3 years of hands-on experience. I’ve done a bit of everything, but running high-performing ads and making sense of the numbers? That’s my thing. If it’s written here, it’s because I’ve lived it.