Meta Andromeda is the retrieval engine Meta built to decide which ads qualify to enter the auction before any bidding happens. It is not a single unified targeting system for your whole account — it is the first filter in a two-stage delivery process, and understanding what it actually does changes how you should think about creative strategy. This post covers what Andromeda is, what it is not, and the specific implications for small business advertisers managing their own Meta accounts.
Are your Meta ads keeping up with how delivery actually works now?
We help local businesses build ad creative and account structure that performs under Meta’s current delivery system.
The Quick Take: How Meta Ad Delivery Changed
| Before Andromeda | After Andromeda |
|---|---|
| All eligible ads entered a single massive auction simultaneously | A retrieval stage filters ads before they reach the auction |
| Manual interest targeting told Meta who to find | Creative content signals tell the retrieval system who an ad is for |
| More ad variations meant more testing data | Similar ads get grouped together and share one auction slot |
| Audience settings were the primary targeting lever | Creative diversity is now the primary targeting lever |
| Small tweaks to existing ads counted as new creative tests | Only conceptually distinct creatives register as separate auction candidates |
The Takeaway: Andromeda did not make Meta ads more complex to manage — it made creative quality and conceptual diversity the single most important variable in account performance.
💡 Pro Tip: The most common misreading of Andromeda is treating it as a reason to produce more ads. Volume alone does not help if the ads are conceptually similar. The system groups similar creatives together regardless of how many you upload. Focus on conceptual range, not raw output.
Table of Contents
→ What Meta Andromeda Actually Is
→ How the Two-Stage Delivery Process Works
→ What Entity IDs Are and Why They Change Your Creative Strategy
→ Why Creative Is Now the Primary Targeting Signal
→ How to Structure Your Account for Andromeda
→ What the Learning Phase Looks Like Under Andromeda
→ The Bottom Line on Meta Andromeda
→ FAQ: Common Questions
What Meta Andromeda Actually Is
Meta Andromeda is a retrieval engine — specifically the first stage of Meta’s ad delivery process, responsible for filtering tens of millions of eligible ads down to a few thousand candidates before any auction occurs. Meta introduced it in late 2024 and rolled it out broadly through early 2025. It runs on specialized hardware including the NVIDIA Grace Hopper Superchip and Meta’s own MTIA chips, which give it the processing power to analyze ad content at a scale the previous system could not reach.
What Andromeda changed is not who wins the auction — that is still determined by bid, predicted engagement, and relevance score in the ranking stage. What Andromeda changed is who gets to enter the auction at all. Before the system existed, Meta’s delivery pipeline ran a simpler pre-filter. Andromeda replaced that filter with deep neural networks capable of reading the visual content, text, and semantic meaning of an ad in real time, then matching it to users whose current behavior patterns align with what that ad is about.
One important clarification worth making directly: Andromeda is not the same thing as Meta’s Advantage+ suite, and it is not the same as GEM (Generative Ads Model), which is a separate Meta AI system focused on ad creative generation. Andromeda is specifically the retrieval layer. It operates under the hood on every Meta ad account regardless of your campaign settings. For a broader view of how Meta paid media strategy fits into local business marketing, the paid media services page covers the full picture.
💡 Pro Tip: You do not activate or configure Andromeda. It is infrastructure, not a setting. What you control is the quality and diversity of the creative inputs you give it. Think of it the way you think of a search engine — you cannot change how it indexes content, but you can make your content easier to index correctly.
How the Two-Stage Delivery Process Works
Meta’s current ad delivery works in two sequential stages: retrieval, then ranking. Understanding the difference between these two stages is the key to understanding why creative matters more than it used to.
In the retrieval stage, Andromeda scans the full pool of eligible ads and selects a shortlist of candidates that appear relevant to a specific user at a specific moment. This happens in under 300 milliseconds — faster than a blink. If your ad does not pass this retrieval filter, it never reaches the auction, regardless of how high your bid is. The retrieval stage is the bouncer. The ranking stage is what happens once you are already inside.
In the ranking stage, the shortlisted ads compete in the traditional auction. Meta evaluates bid, predicted engagement rate, and ad quality to determine which ad actually gets shown. This stage has not changed fundamentally. The critical shift Andromeda created is that the ranking stage now only matters for ads that survived retrieval. An ad with a high bid and strong historical performance can still be excluded entirely if Andromeda does not identify it as relevant to the user being evaluated.
What Entity IDs Are and Why They Change Your Creative Strategy
Entity IDs are the internal labels Andromeda assigns to each ad based on its semantic content — the visual elements, message, tone, and format taken together. Ads that are conceptually similar get assigned the same Entity ID. When multiple ads share an Entity ID, they do not each get their own slot in the retrieval shortlist. They compete with each other for a single slot.
This is the mechanic that breaks the old playbook of uploading many slight variations of a winning ad. Webtopia’s breakdown of Entity IDs puts it clearly: you may think you launched five new ads, but if they look and feel similar, Meta treats them as one. Minor edits — headline swaps, color changes, small copy tweaks — do not create new Entity IDs. Andromeda analyzes the underlying concept, not the surface details.
The practical implication is significant for small business advertisers. Running ten variations of the same product photo with different captions does not give you ten auction entries — it gives you one, shared among all ten. True creative diversity means different formats (video vs. static vs. carousel), different hooks (problem-led vs. benefit-led vs. testimonial), and different visual environments — not the same idea reskinned.
💡 Pro Tip: A practical starting point for local businesses is to build creative around three genuinely different angles: one that leads with the problem your customer is experiencing, one that leads with the outcome or transformation, and one that uses a real customer perspective or story. These three conceptual territories are distinct enough to generate separate Entity IDs and reach different user segments.
Why Creative Is Now the Primary Targeting Signal
Andromeda reads the content of your ad — visuals, copy, format, tone — and uses that analysis to determine which users the ad is relevant for. This is what people mean when they say “creative is the new targeting.” The system infers audience from content rather than waiting for the advertiser to specify one.
An ad showing a specific problem — say, a disorganized home office — signals to Andromeda which users are currently displaying behavioral patterns around home organization, remote work, or productivity. The advertiser does not need to manually target “people interested in home office.” The creative does the targeting work by being specific and clear about who it is speaking to. Vague, generic creative that could apply to anyone tells Andromeda very little, which is why it tends to underperform against focused, specific creative even when the product is the same.
This shift also explains why broad audience targeting has become the recommended approach. When you restrict targeting heavily with interest stacking and demographic filters, you are limiting Andromeda’s ability to find the users it has identified as the most relevant match for your creative. Narrow targeting tells the system where to look. Broad targeting with strong creative tells the system what to look for. The second approach consistently outperforms the first under the current delivery architecture, according to Jon Loomer’s analysis of the Andromeda update.
How to Structure Your Account for Andromeda
The account structure changes Andromeda rewards are consolidation at the campaign and ad set level, and diversification at the creative level. These two principles work in the same direction: pooling your budget and signal into fewer, broader containers gives the retrieval and ranking systems more data to work with, while ensuring the creatives inside those containers are conceptually distinct gives Andromeda more distinct auction candidates.
| Account Element | Andromeda-Aligned Approach |
|---|---|
| Campaign structure | Fewer campaigns with consolidated budget; Advantage+ Shopping or Advantage+ Audience where applicable |
| Audience targeting | Broad (age and location only); let creative signal do the audience work |
| Placements | Advantage+ Placements to allow the system to shift budget to lowest-CPA placement in real time |
| Creative volume | Fewer ads with genuinely different concepts; aim for distinct Entity IDs, not high raw count |
| Attribution window | 7-day click / 1-day view to maximize the conversion signal feeding the learning system |
💡 Pro Tip: If you are running five separate campaigns at $10 per day each, consolidating into one campaign at $50 per day is almost always a better approach under Andromeda. The system needs conversion signal concentrated in one place to learn efficiently. Fragmented budgets across multiple campaigns starve each campaign of the data it needs to exit the learning phase.
What the Learning Phase Looks Like Under Andromeda
The learning phase is the period after launching a new ad set when Meta’s system is gathering enough conversion data to predict delivery reliably. The threshold Meta has historically cited is approximately 50 optimization events per week at the ad set level — not the campaign level. This is a guideline, not a hard gate. Some ad sets stabilize with fewer events when the signal is clean and consistent; others remain in learning longer when signal is noisy or fragmented across too many ad sets.
One update worth noting: Meta has lowered the learning phase threshold from 50 to 10 optimization events for Purchase-optimized and Mobile App Install campaigns specifically. This change has not yet rolled out uniformly to all account types, so verify this in your own Ads Manager under the delivery status column rather than assuming it applies to your account.
The most common reason small business accounts stay stuck in “Learning Limited” is budget fragmentation. When spend is spread across too many campaigns or ad sets, none of them accumulate the conversion volume needed to exit learning. The fix is structural: consolidate campaigns, reduce active ad set count, and ensure each ad set has a realistic path to generating sufficient weekly events at its current budget. If your conversion event is Purchase and your budget cannot realistically drive 50 purchases per week, consider optimizing for a higher-funnel event like Add to Cart or Initiate Checkout to accumulate signal faster. You can shift to Purchase optimization once the account has enough learning data to work from.
The Bottom Line on Meta Andromeda
Andromeda did not make Meta advertising harder — it made the rules more transparent. Creative quality and conceptual diversity have always mattered. What changed is that the system now enforces this directly. Ads that are vague, generic, or conceptually identical to each other no longer get the benefit of the doubt in the retrieval stage. They simply do not enter the auction as often.
For small business advertisers, the practical shift is manageable. Stop optimizing for the number of ads running and start optimizing for the number of genuinely distinct concepts in your account. Build creative around specific problems, specific outcomes, and specific customer perspectives. Give the retrieval system clear signals about who each piece of creative is for, and let broad targeting give it the room to find those people.
The advertiser’s job under Andromeda is not to outsmart the system — it is to feed it better inputs than your competitors do.
🎯 Not Sure if Your Meta Ads Are Set Up for How Delivery Works Now?
We audit Meta ad accounts for local businesses and identify exactly where account structure, creative strategy, or budget consolidation is limiting performance. Book a free 15-minute call to find out where your account stands.
15 minutes. No pitch. Just a clear picture of what your account needs.
Frequently Asked Questions About Meta Andromeda
What is Meta Andromeda?
Meta Andromeda is the retrieval engine Meta introduced in late 2024 that filters tens of millions of eligible ads down to a few thousand candidates before the auction begins. It is the first stage of a two-stage ad delivery process and operates on all Meta ad accounts automatically — it is not a setting you turn on or configure.
Is Meta Andromeda a setting I need to turn on in Ads Manager?
No. Andromeda is infrastructure — it runs automatically on every Meta ad account. You cannot activate or configure it directly. What you control is the quality and conceptual diversity of the creative inputs you give it, which determines how well it can match your ads to relevant users.
What is an Entity ID in Meta Andromeda?
An Entity ID is the internal label Andromeda assigns to each ad based on its semantic content — visuals, copy, format, and tone taken together. Ads that are conceptually similar receive the same Entity ID and compete with each other for a single retrieval slot rather than each getting their own auction entry. Only genuinely different creative concepts generate separate Entity IDs.
Why does everyone say creative is the new targeting with Meta Andromeda?
Because Andromeda reads the content of your ad — visuals, copy, and message — to determine which users it is relevant for, rather than relying solely on the interest or demographic settings the advertiser specifies. Specific, focused creative tells the retrieval system exactly who the ad is meant for. Vague, generic creative gives it very little to work with.
Should I use broad targeting or interest-based targeting with Andromeda?
Broad targeting — age and location only — is the recommended approach under Andromeda. Heavy interest stacking limits the system’s ability to find the users it has identified as the best match for your creative. When your creative is specific and clear about who it speaks to, broad targeting gives Andromeda the freedom to find those people more efficiently than manual interest selection does.
Will my ad costs go up because of Meta Andromeda?
Andromeda is a neutral system — it does not inherently increase or decrease costs. What changes is the efficiency of your spend. Advertisers who adapt their creative strategy to the retrieval architecture often see lower cost per acquisition because the system matches their ads to higher-intent users. Advertisers who do not adapt may see performance decline as similar creatives compete against each other for the same auction slot.
What is the learning phase and how does it work with Andromeda?
The learning phase is the period after launching a new ad set when Meta is collecting enough conversion data to optimize delivery reliably. The threshold is approximately 50 optimization events per week at the ad set level — this is a guideline, not a hard requirement. Meta has also lowered the threshold to 10 events for Purchase-optimized and Mobile App Install campaigns specifically, though this rollout is not yet universal. Consolidating budget into fewer ad sets is the most effective way to exit learning faster.
Does Andromeda affect small business advertisers differently than large ones?
The mechanics apply to every advertiser, but small businesses face a specific challenge: limited budget makes it harder to consolidate signal and exit the learning phase quickly. The most effective counter-strategy is structural consolidation — fewer campaigns, fewer ad sets, and a smaller number of genuinely distinct creative concepts rather than a high volume of similar variations.
How is Meta Andromeda different from Advantage+ campaigns?
They are different systems. Andromeda is the retrieval engine that operates on all ad accounts regardless of campaign type. Advantage+ is a suite of automated campaign formats — Advantage+ Shopping, Advantage+ Audience — that are designed to work with Meta’s AI infrastructure including Andromeda. Using Advantage+ campaign types does not activate Andromeda; Andromeda is already running. Advantage+ formats simply give the system more automation latitude over budget, targeting, and placement decisions.
Should I delete my old interest-based audiences now that Andromeda is live?
Not immediately. The right approach is to test a broad audience campaign alongside your current interest-based setup and compare cost per result over a meaningful time window — at least two weeks. Many accounts find broad outperforms interest-based targeting under Andromeda, but the outcome varies by account, creative quality, and product type. Let the data from your specific account make the decision.