Legacy-style thinking
More decisions are locked in through structure and targeting.
Small segments each receive separate delivery space and signal.
The system has fewer messages and formats to match.
Account structure carries much of the hypothesis.
META ADS GUIDE
Meta Andromeda: What Changed and What It Means for Meta Ads
Meta Andromeda is not just another “algorithm update.” It is part of how Meta gets better at retrieving the most relevant ad candidates before final ranking happens. That does not make campaign strategy irrelevant — it changes how advertisers should think about structure, creative variation and automation.
Last updated: 19 September 2026
Retrievalfirst stage of ad candidate selection
Tens of millionsof ad candidates narrowed dramatically
+6%recall improvement in retrieval
+8%ads quality improvement on selected segments
These are Meta-reported system improvements — not guaranteed account-level outcomes.
01 — RETRIEVAL
What is Meta Andromeda?
Andromeda is Meta’s personalised ads retrieval engine. It operates in the first stage of Meta’s multi-stage ads recommendation system: before final ranking, it selects which ads are viable candidates for a particular person.
Meta describes the task as narrowing tens of millions of ad candidates to a few thousand relevant candidates. Larger, more sophisticated ranking models then predict value for the person and advertiser to determine the final set of ads.
→→→→
The full possible ad pool before personalised selection.
Source: Engineering at Meta — Meta Andromeda.
02 — SHIFT
What changed with Andromeda?
Andromeda increases the model complexity Meta can handle at retrieval. According to Meta, the system can learn higher-order interactions between people and ad data, making candidate matching more personalised at scale.
That becomes more relevant as creative volume grows. In the same broader period, Meta expanded Advantage+, automation and generative creative tooling. More possible ads make a stronger retrieval engine valuable—but do not make every input good.
Andromeda-era thinking
The system gets more freedom to match within healthy constraints.
More relevant candidates can be identified earlier.
Genuinely distinct messages broaden the possible matches.
Structure still matters, but input quality matters more.
Meta reported a +6% improvement in retrieval recall and +8% improvement in ads quality on selected segments after deployment across Facebook and Instagram. These are Meta-reported system metrics—not a promise of equivalent account results. Official source.
03 — DISTINCTION
Retrieval vs Ranking: why the distinction matters
When these concepts are blurred, advice quickly becomes misleading. Retrieval is about entry into the candidate set. Ranking is about evaluation among those candidates.
Retrieval
Which ads even get considered?
Andromeda narrows the enormous possible pool to a smaller personalised candidate set.
Ranking
Which candidates actually win the impression?
After retrieval, other models assess predicted value and help determine final delivery.
Andromeda does not automatically make your ad “win” — it helps Meta retrieve better candidate ads more effectively.
04 — CREATIVE
Why creative strategy matters even more now
As retrieval becomes better at matching ad and person, a varied creative library becomes more valuable. Variety does not mean five nearly identical videos with a new subtitle colour. It means genuinely different reasons to choose the product.
A strong portfolio can vary hooks, problems, value propositions, use cases, proof styles, UGC angles, visual formats and emotional framing. Quantity without meaningful variation is not creative diversity. Endless weak variants remain weak inputs.
Creative diversity matrix
Creative
Hook
Problem
Proof
Offer / value
Format
Creative A
Same
Same
Same
Same
Same
Creative B
Same
Same
Same
Same
Same
Creative C
Same
Same
Same
Same
Same
Creative D
Same
Same
Same
Same
Same
Creative E
Same
Same
Same
Same
Same
Five ads look different but repeat the same idea.
This is an illustrative teaching framework—not Meta account data.
05 — STRUCTURE
What does Andromeda mean for campaign structure?
Andromeda does not mean structure is irrelevant, targeting is dead, or one setup fits every account. It often means unnecessary fragmentation, many micro ad sets and duplicated angles deserve scrutiny.
The goal is not “simple because simple sounds nice.” The goal is structure that supports signal quality, delivery and creative learning without removing necessary separation between markets, economics or objectives.
Ad set 1Limited spend
Ad set 2Limited spend
Ad set 3Limited spend
Ad set 4Limited spend
Ad set 5Limited spend
Ad set 6Limited spend
Ad set 1Concentrated signal · varied angles
Ad set 2Clear commercial reason to separate
Many small ad sets can spread budget and signal, especially at limited volume.
Consolidation may support stronger signal concentration; it does not automatically win and should be assessed against actual volume and business objectives. Also read our guide to the Meta Ads learning phase.
06 — TARGETING
Does this mean targeting no longer matters?
No. Targeting, exclusions, geography, language, age, first-party data and business constraints can still be critical. The shift is away from hyper-manual micro-management by default and toward strong inputs plus room for the system to work.
Broad can create more delivery freedom, but constraints, data and economics determine whether it is right.
Creative matters, but offer, site, tracking, budget and margins still shape the result.
Structure still determines budget flow, signal distribution, objectives and required business separation.
Test meaningful hypotheses with enough room to read them; avoid testing for its own sake.
07 — ADVANTAGE+
How does Andromeda relate to Advantage+ and automation?
Meta positions Andromeda as one of the AI optimisation technologies applied within Advantage+. That does not make the two terms identical: Advantage+ is a broader automation family; Andromeda is retrieval technology within delivery.
Better automation still depends on better inputs. The advertiser’s job shifts toward strategy, creative, economics, tracking and decision quality—not away from responsibility.
Creative inputsMessage · proof · format
Audience and signal inputsEvents · data · constraints
Automation systemsAdvantage+
RetrievalAndromeda
Ranking and deliveryPredicted value · auction
Business outcomesMeasured beyond the system label
Meta reported that more than 1 million advertisers used its generative AI tools to create more than 15 million ads in one month. Meta also reported a 22% ROAS increase among advertisers who had not previously used Advantage+ creative and enabled its AI-driven targeting features, plus an estimated +7% conversions for businesses using image generation. These are Meta’s own figures, scoped to the described groups, and are not universal expectations. Official source.
08 — PLAYBOOK
How should you respond in practice?
The response should be disciplined, not dramatic. Use Andromeda as a reason to improve inputs and remove constraints that serve no real purpose.
Keep separations that have a clear delivery or business reason.
Plan distinct angles, hooks, use cases, proof and formats.
Check events, deduplication and alignment with the commercial goal.
Use margin, CPA, profit and incrementality—not myths about the system.
A new message teaches more than another nearly identical variant.
Ensure budget and volume can support the chosen structure.
Give the system room, while setting direction through creative, economics and measurement.
Andromeda response scorecard
Answer for your current setup. The result is guidance—not a benchmark or Meta diagnosis.
Are we spreading spend across too many ad sets?
Do we have genuinely different creative angles?
Is our tracking clean enough for decisions?
Are we relying on audience splitting more than message quality?
Are we evaluating creative diversity properly?
Answer the questions to get a focused review direction.
Guidance only. No answer creates an authoritative score or guarantees performance.
09 — EXAMPLE
A simple e-commerce example
Illustrative example—not account data.
- Ad sets
- 6
- Creative
- Similar, repeated
- Signal
- Split
- Ad sets
- 2
- Creative
- Genuinely varied
- Signal
- More concentrated
The point is not that fewer ad sets always win. It is that stronger creative diversity and less unnecessary fragmentation can create better conditions for retrieval, delivery and learning. Brand B still needs clean tracking, a competitive offer and sound economics.
Judge the result against the business—for example using your break-even ROAS.
10 — ANTI-MYTH
What Andromeda does NOT mean
- Manual strategy is not dead; its focus shifts toward better decisions and inputs.
- Creative quality does not stop mattering because AI helps with matching.
- “Run broad everything” is not a universal prescription.
- Campaign structure is not irrelevant; unnecessary complexity is the problem.
- Established principles do not become obsolete overnight.
- Advantage+ alone cannot repair a weak offer, poor tracking or bad economics.
Andromeda improves retrieval. It does not override the reality of your business.
11 — CONTEXT
Where does Andromeda fit in?
Andromeda makes most sense as part of a broader Meta Ads shift—not as an isolated switch an advertiser turns on.
More delivery decisions are handled dynamically by the system.
Automation expands through a broader product family.
Tools can increase the volume and variation of ad inputs.
Stronger personalised retrieval supports candidate selection at scale.
Contextual evolution—not a complete history of Meta’s models or a claim that every element launched simultaneously.
FAQ
Frequently asked questions about Meta Andromeda
Meta Andromeda is Meta’s personalised ads retrieval system. It operates in the first stage of the multi-stage system, selecting a smaller relevant candidate set before more sophisticated models rank those candidates.
No. “The algorithm” is often shorthand for many systems. Andromeda has a specific retrieval role. Ranking, auction, pacing, measurement and other systems also affect final delivery.
It strengthens the case for avoiding unnecessary fragmentation so signal and creative can work within a healthy structure. It does not mean every account should use the same campaign setup.
Not as a universal rule. Fewer ad sets can often concentrate signal when volume is limited, but separate markets, economics or business objectives can justify real separation.
No. Geography, language, age, exclusions, first-party data and business constraints remain relevant. The balance simply shifts away from unnecessary micro-segmentation.
Tests should create meaningful differences in message, hook, proof, use case and format. Minor cosmetic variants do not give the candidate pool the same breadth as genuinely distinct concepts.
Retrieval determines which ads are even considered relevant candidates. Ranking then assesses expected value among those candidates and helps determine final ad delivery.
Meta describes Andromeda as one of the AI optimisation technologies used within Advantage+. Andromeda is retrieval technology; Advantage+ is a broader automation product family.
No. The retrieval system is part of Meta’s delivery infrastructure. The practical response should still fit the account’s volume, budget, creative capacity and commercial objectives.
Not automatically. Diagnose unnecessary fragmentation, creative repetition, measurement quality and economics first. Change only what has a clear account-specific rationale.
META ADS
Andromeda doesn’t remove the need for strategy — it raises the bar for it
As Meta improves retrieval and automation, structure, creative quality, tracking and commercial clarity matter even more. JLDigital helps e-commerce brands turn that shift into practical account strategy.
See how we work with Meta Ads