Meta rebuilt its ad delivery infrastructure from the ground up in 2024 and 2025. The new system, called Andromeda, completed its global rollout in October 2025. A study covering 3,014 ecommerce advertisers across 73 countries, analysing 115.7 billion impressions and $834 million in ad spend throughout 2025, documents what changed and who got hit hardest.
Every advertiser running Facebook and Instagram campaigns is now operating under Andromeda, whether they realise it or not. The question is not whether the system affects your account. It does. The question is whether your creative strategy has adapted to how the new system actually works.
Most haven’t. Here’s what you need to understand, and what to do about it.
What Is Andromeda?
Andromeda is Meta’s AI-driven ad retrieval system. It operates at the stage before bidding when Meta evaluates millions of potential ads to show a user, and Andromeda is what narrows the field to viable candidates. It then works alongside GEM (Global Earnings Model), which handles ranking and pricing after retrieval. Andromeda decides which ads make it onto the shelf. GEM decides what gets featured.
The old retrieval system was rule-based and relatively simple. It couldn’t handle scale. With Advantage+ Creative generating hundreds of variations from a single ad, the previous infrastructure would have collapsed. Andromeda was built to replace it.
This is not a gradual evolution of the previous system. It is a different system with a different optimisation logic. Strategies built for the old model — narrow audience segments, lookalike seeds as primary targeting, multiple ad sets testing different audiences — no longer perform the way they did because they are solving the wrong problem.
The Core Shift
Before Andromeda, targeting defined who saw your ads. Under Andromeda, your creative defines who sees your ads. The algorithm reads your assets — visually, structurally, and in terms of messaging — and predicts which users are most likely to respond. Your interest targeting and lookalike audiences still exist, but they now function as soft suggestions rather than hard boundaries.
The creative is doing the targeting work.
That means the algorithm is no longer asking: “Who should see this ad?”
It is increasingly asking: “Which creative best matches this person right now?”
The Mechanics eCommerce Brands Need To Understand
Entity ID: Why Ten Variations Can Equal One Ad
Andromeda assigns an Entity ID to each ad based on its visual and structural characteristics, not the file itself, but the pattern it detects. Similar ads are collapsed into a single Entity ID and made to compete with each other rather than expanding reach. Ten product shots with different headline colours are not ten ads. They are one ad, given one retrieval ticket. The practical consequence: if your creative library is variations of the same concept, you are not building creative volume. You are building creative noise.
The Creative Similarity Score
Performance data from late 2025 suggests that a Creative Similarity Score above 60% triggers retrieval suppression.
When your ads are more than 60% similar, Andromeda treats them as redundant. The recommended target is a Diversity Index below 40% across active assets, meaning each asset should share fewer than 40% of visual and audio features with the others in the same campaign. This metric is now visible in Meta Ads Manager under delivery reports. Check it on your active campaigns before reading further.
Accelerated Creative Fatigue
Effective ad lifespan has compressed from 6–8 weeks pre-Andromeda down to just 2–4 weeks. The precision matching works both ways. Andromeda finds your best audiences faster, but exhausts them faster, too. Brands running a quarterly creative refresh are now running a cadence that is three to six times too slow. The operational implication is significant: you need a systematic pipeline for generating genuinely new concepts, not a periodic refresh of existing ones.
Conversion Signal Quality
Andromeda uses conversion signal quality as a factor in retrieval. Pixel-only tracking now penalises your ad quality score. Implementing the Conversions API with proper deduplication directly improves how often your ads make it through the retrieval gate. This is the least visible but arguably most structurally important point — poor tracking doesn't just make your reporting inaccurate, it makes your ads less likely to be shown.
- 22%: ROAS lift from Advantage+ creative vs manual setup
- 17%: More conversions from 1 ad set with 25 diverse creatives vs 5 ad sets
- 2–4 weeks: Current effective ad lifespan, down from 6–8 weeks pre-Andromeda
- <40%: Target Creative Similarity Score across active assets
What The Shift Looks Like In Practice
Before Andromeda
- Multiple ad sets testing different audiences
- Lookalike seeds as a primary targeting strategy
- 6–8 week creative refresh cycle
- 5–6 ad variations per ad set
- Audience segmentation as the primary lever
- Interest targeting as a hard filter
- Pixel-based conversion tracking sufficient
Under Andromeda
- Fewer, broader ad sets with more creative diversity
- Lookalikes as signal input, not a hard boundary
- 2–4 week creative refresh cycle
- 10–25 conceptually distinct creatives per campaign
- Creative diversity as the primary lever
- Interest targeting as a soft bias only
- Conversions API with deduplication required
Andromeda doesn’t pick the winning ad. It matches different ads to different people based on predicted response. To do that effectively, it needs options. Ten variations of the same concept are not options.
What Genuine Creative Diversity Looks Like And What It Doesn’t
This is where most accounts get it wrong. When Meta says “diversify your creative,” the instinct is to produce more variations of existing concepts. More headlines. Different button colours. Cropped differently. These do not constitute creative diversity under Andromeda. They constitute creative noise. The same Entity ID, competing with itself.
The practical framework for building genuine creative diversity is to think in terms of concepts, not variations. Each concept should represent a distinct combination of: who the creative is for (persona), what outcome it focuses on (desire), and how familiar with the problem the viewer is expected to be (awareness level). Two ads built from different concept combinations will almost always register as genuinely distinct to Andromeda, regardless of whether they use the same product imagery.
Same product shot, five different headlines
- What Andromeda sees: One Entity ID — same visual pattern
- Retrieval outcome: Compete as one ad, not five
- Verdict: Noise
Same hero image, different background colour
- What Andromeda sees: One Entity ID — visual similarity >60%
- Retrieval outcome: Retrieval suppression triggered
- Verdict: Noise
Same concept, trimmed 2 seconds from the video
- What Andromeda sees: One Entity ID — structural similarity detected
- Retrieval outcome: Treated as duplicate
- Verdict: Noise
Product demo video vs lifestyle image
- What Andromeda sees: Different Entity IDs — format and visual pattern distinct
- Retrieval outcome: Separate retrieval tickets
- Verdict: Partial
Pain point hook ad vs social proof ad vs offer ad
- What Andromeda sees: Different Entity IDs — messaging and visual approach distinct
- Retrieval outcome: Each matches different user segments
- Verdict: Genuine
Static catalogue ad vs video catalogue ad (same product)
- What Andromeda sees: Different Entity IDs — format registers as distinct type
- Retrieval outcome: Doubles the retrieval tickets per product
- Verdict: Genuine
Different creator, same script
- What Andromeda sees: May cluster depending on visual similarity
- Retrieval outcome: Partial differentiation at best
- Verdict: Partial
Different concept: new hook, new format, new emotional trigger
- What Andromeda sees: Different Entity ID — new signal for retrieval
- Retrieval outcome: New audience segment reached
- Verdict: Genuine
The eCommerce-Specific Opportunity
The Confect study makes one finding that every ecommerce brand running Meta should pay attention to: e-commerce advertisers who adapt, particularly through Catalogue Ads, have a structural advantage under Andromeda that other verticals cannot easily replicate.
The reason is specific. E-commerce product catalogues contain inherent creative diversity — different products, different visual treatments, different price points, different use cases. Properly structured, each product in your catalogue represents a potential distinct Entity ID. A brand with 200 SKUs has 200 potential retrieval candidates before they've made a single creative decision.
The lever most accounts are not pulling: combining static catalogue ads with video catalogue ads for the same products. Top performers are 62% more likely to use Video Catalogue Ads, which register as entirely different entity types in Andromeda's clustering system compared to static catalogue ads of the same product. This doubles the retrieval opportunities for any product that has both formats without creating new creative concepts from scratch.
The eCommerce Quick Win
If you are running static catalogue ads and not video catalogue ads for the same products, you are leaving retrieval tickets on the table. Creating video versions of your top-performing catalogue products — even simple motion graphics or slideshow-style videos — registers as a genuinely distinct creative type and opens a separate retrieval path for every product it covers.
Account Structure Under Andromeda
The structural implication of Andromeda is that fragmented account architectures — many ad sets, narrow targeting, limited creative per set — are now actively counterproductive. A controlled test showed that a single ad set with 25 diverse creatives produced 17% more conversions at 16% lower cost versus a traditional 5-ad-set structure.
This is not primarily an efficiency argument. It is a signal density argument. Fragmented structures dilute the conversion data available to each campaign, which reduces Andromeda's ability to learn which creative-user combinations are working. Consolidation concentrates the signal. More signal means faster, more accurate retrieval decisions.
The Lookalike Question
Lookalike audiences are not dead, but they no longer function as a primary targeting strategy for most accounts. For accounts spending under £5,000 per month with limited conversion history, a 2–3% lookalike can still provide useful early-stage scaffolding. For established accounts with sufficient conversion volume, the algorithm's own behavioural signals already exceed what a lookalike seed can define. Adding a LAL constraint limits delivery without meaningfully improving targeting quality.
The Creative Pipeline Problem
The hardest part of adapting to Andromeda is not the strategy. It is the operational reality of building and sustaining a creative pipeline that produces genuinely distinct concepts every two to four weeks, at the volume the system now requires.
Most e-commerce brands are not structured for this. Creative production was historically a quarterly exercise or a campaign-by-campaign decision. Under Andromeda, it needs to be a continuous process with a systematic framework for concept generation, a clear brief structure that ensures new concepts are genuinely distinct rather than variations of existing ones, and a regular audit of Creative Similarity Scores to catch homogenisation before it triggers retrieval suppression.
The brands reporting 20–35% higher ROAS than legacy-structure accounts in 2026 are not necessarily spending more on creative. They are executing the brief more systematically — building 8–12 conceptually distinct concepts per campaign, refreshing on a 2–3 week cycle, and treating creative diversity as an operational discipline rather than a creative aspiration.
The Honest Constraint
Building 10–25 genuinely distinct creatives per campaign every 2–4 weeks is a significant resource commitment. For brands without an established creative production process, the right starting point is not trying to hit the volume ceiling immediately. It is auditing what you currently have, identifying the Creative Similarity Score on your active campaigns, eliminating genuine duplicates, and building toward the target volume over a defined period. The direction matters more than the immediate number.
What To Do This Week
Five actions, in priority order.
1. Check your Creative Similarity Score in Meta Ads Manager delivery reports. If any active campaign is above 60%, you have a retrieval suppression problem that is affecting performance today, not in the future.
2. Audit your Conversions API setup. If you are running pixel-only tracking, implement the CAPI with proper deduplication. This is a prerequisite for Andromeda to receive accurate feedback signals, and accurate signals directly affect retrieval frequency.
3. Identify your static catalogue ads and brief video versions. This is the highest-leverage low-effort action for ecommerce brands. It doubles retrieval opportunities per product without requiring new concept development.
4. Consolidate ad sets where you have multiple running with similar audiences. Fewer ad sets with more creative volume outperform fragmented structures. If you have five ad sets with four ads each, consider whether three ad sets with seven each would give the algorithm more to work with.
5. Map your active creative library against concept dimensions. List every active ad and assign it a persona, a desire focus, and an awareness level. Any two ads that share all three are genuine duplicates, regardless of how different they look. Any gap in the combinations is an opportunity for a new concept brief.
Andromeda represents the largest change to Meta advertising since iOS 14.5. The accounts that have adapted are reporting better performance than legacy-structure accounts. The accounts that haven’t are paying for reach that the retrieval system is suppressing before the auction even begins.
The strategy is not complicated. The pipeline discipline required to execute it consistently is. That is the actual work.
Need a fresh perspective? Let’s talk.
At 360 OM, we specialise in helping businesses take their marketing efforts to the next level. Our team stays on top of industry trends, uses data-informed decisions to maximise your ROI, and provides full transparency through comprehensive reports.
.png)
