Is Meta's Andromeda Update Killing Your Ads? Here's How to Fix It

The Core Concept
Meta's Andromeda update moves the ad auction from 'audience-first' to 'creative-first' retrieval. Instead of you selecting who sees your ad, the AI analyzes your video's visual data to find users who have historically engaged with similar patterns.

The Strategy
To win, you must shift from 'perfecting one ad' to 'feeding the algorithm' with high-velocity creative testing. The goal is to produce enough distinct visual variations (10-20 per week) to allow the AI to find its own audiences through 'Creative Darwinism.'

Key Metrics

• Creative Refresh Rate: Target 10-20 new distinct hooks/week
• First-Stop Ratio: Aim for >30% (users stopping scroll in first 3s)
• Algorithm Learning Phase: Should exit in <48 hours with sufficient spend

Tools like Koro enable this by automating the production of UGC-style variants at scale.

Meta's Andromeda update is a fundamental re-architecture of the ad delivery system, powered by MTIA chips and NVIDIA Grace Hopper Superchips. It shifts the core mechanism of ad targeting from manual inputs (interests, lookalikes) to real-time generative inference based on creative assets.

Andromeda is a retrieval-based AI system that scans the content of your video ads—frames, audio, text overlays—to determine who should see them. Unlike previous updates that relied on user pixel data, Andromeda specifically focuses on 'creative signal density' to match ads to users based on their consumption habits [1].

The Technical Shift: Retrieval vs. Ranking

In the old world, you told Meta who to target (Retrieval), and Meta decided which ad to show them (Ranking). Under Andromeda, Meta handles both. The system retrieves users based entirely on the signals embedded in your creative. If your creative lacks clear visual signals—like a specific product use case or demographic cue—the system cannot find a relevant audience.

Why This Matters for D2C Founders:
If you are still running broad targeting with only 2-3 creative variations, you are effectively starving the algorithm. It has the processing power to find your customers, but you haven't given it enough 'food' (creative data) to learn who they are.

Manual targeting constraints are now actively penalizing performance. When you restrict audiences using 'Interest Stacks' or 'Lookalikes,' you prevent Andromeda's sequence learning models from exploring high-intent pockets of users that your manual settings exclude.

The 'Minor Tweak' Trap

I've analyzed dozens of accounts where founders believe they are testing creatives, but they are actually just testing 'minor tweaks.' Changing the button color or the first 2 seconds of text is no longer enough. Andromeda recognizes these as effectively the same asset.

To trigger a new audience search, you need Semantic Variety. This means changing the core visual composition, the avatar, the setting, or the spoken script.

If your ROAS has dropped, it is likely because your creative diversity score is too low. The algorithm is showing your same 3 ads to the same shrinking pool of people because it lacks the data to expand [3].

Creative Darwinism is the practice of launching a high volume of distinct creative assets and letting the algorithm 'kill' the losers and 'scale' the winners naturally. Instead of you guessing which hook works, you provide the system with 20 options and let user behavior decide.

Programmatic Creative is the use of automation and AI to generate, optimize, and serve ad creatives at scale. Unlike traditional manual editing, programmatic tools assemble thousands of variations—swapping hooks, avatars, and scripts—to match specific platforms instantly.

Why Volume is the New Targeting

Under Andromeda, every new creative asset is a new targeting set.

• Video A (Gym setting) finds fitness enthusiasts.
• Video B (Office setting) finds corporate workers.
• Video C (Home setting) finds stay-at-home parents.

If you only launch Video A, you are voluntarily ignoring the audiences for B and C. This is why volume matters. It's not about spamming; it's about casting a wider net of context.

The Koro Advantage:
This is where tools like Koro become essential infrastructure. Koro allows you to generate these distinct semantic variations without hiring three different film crews. By using different Indian AI avatars and backgrounds, you can create the 'Gym,' 'Office,' and 'Home' variations in minutes, satisfying Andromeda's hunger for signal density.

One pattern I've noticed is that the brands winning in 2026 aren't the ones with the biggest budgets, but the ones with the fastest feedback loops. Bloom Beauty (Cosmetics) is a perfect example of this shift.

The Problem:
Bloom was stuck in the 'Minor Tweak' trap. They had one winning ad format (a texture shot) and kept re-running slightly different versions of it. When Andromeda hit, their CPA doubled because the algorithm had exhausted the audience for that specific visual format.

The Solution:
They utilized Koro's Competitor Ad Cloner + Brand DNA feature. Instead of guessing new formats, they scanned the market for trending structures—specifically a 'Scientific-Glam' breakdown and a UGC testimonial format. They didn't just copy; they used Koro to inject their specific brand voice into these proven structures.

The Results:

• Metric: They generated 15 distinct new hooks in 48 hours.
• Metric: One of the new 'cloned' structures achieved a 3.1% CTR, becoming an outlier winner.
• Metric: The new creative beat their own control ad by 45% in ROAS efficiency.

By moving from manual iteration to AI-assisted generation, they stopped fighting the algorithm and started feeding it.

To replicate this success, you need a workflow that prioritizes volume and variety over perfection. Here is the 30-day playbook for implementing Creative Darwinism.

Phase 1: The Audit (Days 1-5)

Check your current 'Signal Density.' Look at your last 10 ads. If they all feature the same person, the same background, or the same first 3 seconds, you fail the audit. You need to identify 3 distinct 'Angles' to test (e.g., Social Proof, Problem/Solution, Unexpected Use Case).

Phase 2: The Generation (Days 6-10)

Use an AI tool to build out your assets. You cannot do this manually without burning out.

Manual vs. AI Workflow

Phase 3: The Launch & Kill (Days 11-30)

Launch all 10-20 variations in a Broad Targeting CBO (Campaign Budget Optimization) campaign.

1. Launch: Set a budget that allows for at least 500 impressions per ad.
2. Wait: Do not touch it for 48 hours. Let Andromeda test.
3. Kill: Turn off any ad with a 'Thumbstop Ratio' (3-second view / Impressions) under 25%.
4. Scale: Move the winners to a scaling campaign.

Koro's Role: Koro excels at the 'Generation' phase, allowing you to flood the 'Launch' phase with high-quality, distinct options. However, for highly complex storytelling ads that require physical product interaction (like applying a specific serum texture to real skin), a traditional shoot might still be necessary for your 'Hero' asset.

Success in the Andromeda era requires looking beyond just ROAS. ROAS is a lagging indicator. You need leading indicators to know if your creative strategy is working.

1. Creative Refresh Rate
How many new concepts are you launching per week? The benchmark for aggressive growth is now 5-10 completely new concepts weekly. If you are below this, you are vulnerable to fatigue.

2. Signal Diversity Score
This isn't a Meta metric, but an internal one. Categorize your live ads. Do you have a mix of UGC, Static, and Polished Video?

• Micro-Example: If you have 5 videos but they are all 'Talking Head,' your diversity score is low. Add a 'Product Demo' or 'Unboxing' style using Koro to boost this.

3. Thumbstop Ratio
The percentage of people who watch the first 3 seconds.

• Benchmark: <20% is poor. 20-30% is average. >30% is a winner.
• If your Thumbstop is low, your hook is the problem. Use Koro to regenerate just the first 3 seconds of your video with a more aggressive visual or statement.

In my experience working with D2C brands, those who obsess over Thumbstop Ratio consistently outperform those who obsess over audience settings. The hook is the new targeting.