Meta Andromeda Killed Your ROAS. Here's the Only Way to Get It Back.

Most brands are blaming the algorithm. The data says the problem is your creative pipeline, and it was broken before Andromeda flipped the switch. Here's what's actually happening, and what the top performers across 170+ brands are doing differently.

What Andromeda Actually Did

In December 2024, Meta quietly deployed a new ads retrieval engine called Andromeda. By October 2025, the rollout was complete. By early 2026, every advertiser on Facebook and Instagram, regardless of budget, vertical, or sophistication, was running inside it.

The official Meta Engineering blog described it as a "next-gen personalized ads retrieval engine" built on a custom deep neural network running on the NVIDIA Grace Hopper Superchip. Model capacity increased 10,000x. Query speed improved 3x. Feature extraction got 100x faster.

The numbers are impressive. The consequences for most advertisers have been brutal.

Across a study of 3,014 advertisers, ROAS declined 7% on average during the rollout. Landing page conversion rates dropped 17%, from 3.5% to 2.9% industry-wide. High-ticket products saw ROAS collapse by 31%. Affordable products fared even worse, dropping 35%.

Inside CTC's own client portfolio, tracked in real time across 170+ brands on Statlas, the pattern was consistent: CPMs spiked +28% year-over-year in Q1. Conversion rates dropped. CTR actually climbed (+43%), which meant clicks were getting cheaper, but fewer of them were converting. The ads were reaching people. The wrong people.

The prevailing narrative: Meta broke something. Targeting stopped working. The algorithm changed and nobody told us.

That narrative is wrong. And believing it is why most brands will keep losing.

Why You're Diagnosing It Wrong

Here is the uncomfortable truth that the industry doesn't want to say out loud:

Andromeda didn't break your targeting. It revealed that you never really had a creative strategy. The algorithm is just a mirror.

Before Andromeda, Meta's ad system operated on a relatively simple principle: you define an audience, Meta finds people in that audience who might click. Targeting was the lever. Creative was the execution. You could win with a strong offer, a tight lookalike, and a handful of tested assets. The machine was doing the heavy lifting of audience selection, and you were feeding it a modest number of options to choose from.

Andromeda inverted this completely.

The new system doesn't start by asking "who should see this ad?" It starts by asking "which ad should this person see?" Meta's AI analyzes your creative, color, format, hook structure, social proof placement, emotional tone, narrative arc, product visibility, and maps it against behavioral and interest signals for hundreds of millions of users. It then routes impressions to the people whose behavioral fingerprint most closely matches what your creative signals.

Your creative is now your targeting.

This means that if you're running 4 ads and your best competitor is running 40, you're not just outgunned on volume. You're functionally invisible to entire segments of the market that your competitor's creative is reaching, segments that may be your highest-value customers.

Andromeda didn't change who your customers are. It changed the system that finds them. And that system now requires creative diversity at a scale that most brands have never invested in.

The Data: A 65% ROAS Gap Hiding in Plain Sight

CTC manages advertising for over 170 ecommerce brands, and Statlas, our internal analytics platform, gives us cross-portfolio visibility that most agencies don't have. When we looked at Meta performance across accounts with radically different creative testing cadences, one number stood out.

Sixty-five percent higher ROAS for brands testing 20+ new ads per month versus those testing fewer than 10.

Sixty-five percent. Not a rounding error. Not a single outlier account. A structural, repeatable gap that tracks directly with creative testing volume.

This finding is corroborated by third-party data. A study of over 1 million Meta ad creatives found that brands running 20+ new ads per month achieve 65% higher ROAS than those testing fewer than 10. Advantage+ campaigns with AI creative features deliver 22% ROAS improvements versus manually managed setups. A single ad set with 25 conceptually diverse creatives outperforms five separate, narrowly targeted ad sets, delivering 17% more conversions at 16% lower cost.

The top performers in this data set aren't running more ads randomly. They're running more conceptually distinct ads, different emotional triggers, different use cases, different personas, different formats, and letting Andromeda match each one to its most relevant audience segment.

The bottom performers are running 3-5 variations of the same ad.

How Andromeda Actually Evaluates Your Ads

To win, you need to understand what the machine is actually doing. Andromeda's retrieval process works in multiple stages:

1. Embedding extraction. The moment your ad goes live, Andromeda generates a multi-dimensional embedding, a mathematical fingerprint of your creative. It's reading visual elements, copy structure, product context, format signals, and engagement history.
2. Similarity matching. That embedding is compared in real time against behavioral embeddings for hundreds of millions of users, identifying people whose historical behavior patterns correlate with your creative's profile.
3. Hierarchical retrieval. Rather than scanning every live ad against every user, the system uses a layered hierarchy to efficiently narrow millions of candidates to thousands, then ranks them. This is where ad creative diversity pays off: the more distinct your creatives, the more nodes in the hierarchy your ads can occupy.
4. Entity clustering suppression. Here's the part most advertisers miss: Andromeda identifies visually similar ads and treats them as a single "entity." Run 10 variations of the same static product shot and they don't get 10 chances at distribution. They get one, then compete against each other for that slot.

Creative diversity isn't about volume. It's about occupying distinct conceptual territory that maps to different audience segments.

The practical implication: creative similarity scores matter. Top-performing accounts maintain creative similarity scores below 40% across their active ad sets. Your 10 "variations" may be functioning as 1. The brands seeing 65% higher ROAS are running creatives that are genuinely different from each other, not color-swaps.

What Winners Are Doing Differently

Across CTC's portfolio and third-party data from over a million Meta creatives, the behavioral gap between top and bottom performers is stark and consistent. Here's what separates them:

They treat creative as a product roadmap, not a campaign deliverable.

Bottom performers ship creative when they launch a campaign. Top performers maintain a continuous creative pipeline, briefing, producing, testing, and retiring ads on a rolling weekly cadence. When Meta's Andromeda compressed the effective ad lifespan from 6-8 weeks down to 2-4 weeks, teams with a pipeline barely flinched. Teams that treated creative as a project had to scramble every month.

They map creative to audience states, not audience segments.

Rather than building creative around demographic targeting that's now largely irrelevant, top performers build creative frameworks around customer awareness levels, emotional states, and use cases:

• Problem unaware - native-style content that surfaces the pain point
• Problem aware, solution unaware - educational content that introduces the category
• Solution aware - product-first content with social proof and specifics
• Gift/occasion context - occasion-specific creative that triggers different purchase intent
• Skeptic/comparison buyer - objection-handling creative, competitive framing

Each of these maps to a distinct behavioral cluster that Andromeda can find, if you give it the creative to work with.

They've gone all-in on Catalog Ads.

This is the single biggest structural gap between top and bottom performers. Top-performing brands generate 60% of their total Meta revenue from Catalog Ads. They deploy 72% more of them, allocate 117% more budget to them, and maintain libraries of 67+ live Catalog Ads versus the bottom-performer benchmark of fewer than 39.

The performance advantage is massive: Catalog Ads deliver 23% higher ROAS, 37% better CPA, and a 20-30% ROAS advantage maintained across the entire creative lifecycle. Accounts allocating 60-100% of Meta budget to Catalog Ads see 44% higher ROAS and 68% lower CPA than those that don't.

The CTC Playbook for the Andromeda Era

CTC's creative methodology, built on the Prophit Engine, was designed around systematic creative intelligence before Andromeda made it mandatory. Here's the framework we now run across every Meta account in the portfolio:

The Creative Volume Benchmark

Minimum viable creative output is now a function of monthly ad spend:

• Under $10K/month: 8-12 new, conceptually distinct creatives per month
• $10K-$50K/month: 15-25 new creatives per month
• $50K+/month: 25-40+ new creatives per month, with weekly refresh cycles

These aren't variations. These are conceptually distinct creative approaches, different angles, formats, emotional triggers, and customer states. If they look similar, Andromeda will treat them as similar.

The Creative Brief Framework (5 Dimensions)

Every creative brief at CTC maps to at least one of five strategic dimensions:

1. Awareness Stage - Where in the purchase journey is the target customer? Unaware, aware, considering, or ready to buy?
2. Emotional State - What is the customer feeling when they need this product? Frustrated, aspirational, gift-giving, skeptical?
3. Format Diversity - Are we covering video (UGC, polished brand, talking head), static (lifestyle, product-forward, infographic), and catalog formats?
4. Product/Use Case - Are different use cases, product variations, and collections each getting dedicated creative attention?
5. Audience Persona - Are distinctly different customer personas (demographics, lifestyle, occasion) represented in the creative pool?

What to Do Right Now

If you're reading this and your ROAS has declined since Q4 2025, here is the diagnostic framework, and the immediate actions that move the needle:

Step 1: Audit your creative similarity score.

Look at your active ads. If you can easily describe all of them in the same sentence, "product shot with lifestyle background and value prop headline," you have an entity clustering problem. Andromeda is routing all of those through the same retrieval node. You need genuinely different creative, not variations.

Step 2: Count your active Catalog Ads.

If the answer is fewer than 40, start here before anywhere else. Catalog Ads are the single highest-leverage asset in a post-Andromeda Meta account. If you don't have Design Rules set up in your catalog, set them up today, the adoption gap between top and bottom performers on this single feature is 103%.

Step 3: Consolidate your campaign structure.

If you're running more than 3-4 Meta campaigns and none of them is hitting 20+ conversions per week, you're fragmenting your signal. Consolidate your best-performing ads into a single, broad-targeted campaign. Give Andromeda a data-dense pool to optimize from.

Step 4: Build a creative cadence, not a creative sprint.

The ad fatigue window is now 2-4 weeks. That means you need a minimum of 2-3 new, conceptually distinct creatives entering the testing pipeline every week at moderate spend levels, more at higher spend. This isn't a campaign launch. It's an operating system.

Step 5: Change your success metric.

Stop optimizing ad-level ROAS. Start optimizing account-level efficiency, your aMER, and measure creative performance against incremental ROAS (iROAS), not platform attribution. Some of your best creatives are building the funnel, not closing it. Don't kill them before you understand what they're doing.

The Bottom Line

Meta's Andromeda update is not a temporary disruption. It's a permanent architectural shift in how the platform routes advertising to audiences. The old playbook, tight targeting, narrow audiences, a handful of proven creative assets, is functionally broken. Not because Meta changed the rules, but because the rules now expose what was always true: creative strategy was always the lever. We just had enough targeting crutches to avoid building it.

The data is clear. Across CTC's portfolio of 170+ ecommerce brands, brands testing 20+ new conceptually distinct ads per month are achieving 65% higher ROAS than those running under 10. That gap will widen as Andromeda matures. The brands that build systematic creative pipelines now will be the ones that compound into 2027. The brands that don't will keep blaming the algorithm.

The algorithm isn't the problem. It's just the mirror.

Want CTC to Audit Your Meta Creative Health?

We run our CRTV Score diagnostic across your Meta account and tell you exactly where the 65% gap lives in your account, and how to close it.

Talk to Us