For years, a large part of Meta Ads management happened inside the audience builder.
Advertisers separated interests into ad sets, tested different lookalike percentages, built retargeting windows, and tried to find the audience combination that produced the lowest acquisition cost.
Those controls still have uses. But Meta's advertising system has become much better at deciding which eligible person should see which ad.
Andromeda is one part of that change.
Its practical importance is not that targeting has disappeared. It is that Meta can evaluate far more potential matches between people and ads than older retrieval systems could handle.
That changes where advertisers can create an advantage.
If Meta has more freedom to identify the right person, feeding the system ten versions of essentially the same message is not much of a creative strategy.
The question becomes less “Which tiny audience should see this ad?” and more “Do we have enough genuinely different ads for Meta to find the message that matters to different buyers?”
What Meta Andromeda actually does
Andromeda is Meta's personalised ads retrieval system.
Retrieval happens early in the ad recommendation process.
When someone opens Facebook or Instagram, Meta cannot run its most expensive ranking models against every eligible advertisement on the platform. There are too many possibilities and very little time.
Instead, the process works in stages.
According to Meta, retrieval starts with tens of millions of potential ads and reduces them to a few thousand relevant candidates. Later ranking systems evaluate that smaller group in more detail before deciding which ads are ultimately shown.
Andromeda was built to make that first stage much more capable.
Meta says its new architecture increased retrieval-model capacity by roughly 10,000 times compared with the previous approach. That allows the system to model more complex relationships between people and ads while still working within the strict latency requirements of ad delivery.
This distinction matters.
Andromeda is not the entire Meta Ads algorithm. It does not independently decide your budget, final ad ranking, offer, landing page, or business strategy.
It improves Meta's ability to identify promising ad candidates before later parts of the system make further decisions.
Why creative volume suddenly matters more
Meta did not build Andromeda only because audience prediction became more complicated.
It also had to deal with a much larger number of ads.
Advantage+ tools can automate parts of audience selection, budget allocation, placement, and creative delivery. Meta's generative AI tools can also produce additional creative variations.
That means the recommendation system has many more possible ads to consider.
Andromeda was designed partly around this increase in creative volume.
But there is an important distinction for advertisers:
more creative files do not automatically mean more creative choices.
If twenty ads all make the same argument, they may give Meta twenty executions of one idea rather than twenty genuinely different ways to match a customer with the product.
The “perfect audience” matters less when Meta can explore
This does not mean audience targeting is dead.
Geography still matters. Language can matter. Legal restrictions matter. Product availability matters. Existing-customer strategy matters. Some products genuinely serve narrow groups of people.
What has changed is the amount of work that needs to happen through manual interest segmentation.
Meta's Advantage+ systems increasingly give its models more room to find people likely to respond rather than requiring the advertiser to define every potential buyer in advance.
That makes an old problem more obvious.
You can spend hours arguing about whether an audience should contain “Running”, “Fitness”, or a 3% lookalike while every ad in the campaign says exactly the same thing.
The targeting looks sophisticated.
The communication is not.
For many eCommerce campaigns, the larger opportunity is now to improve what Meta has available to show once it finds a potential buyer.
Creative variation is not creative diversification
One of the easiest mistakes in paid social is counting files instead of ideas.
A campaign may contain ten ads and only one concept.
It helps to separate four different types of creative work.
Variation
A variation changes part of an existing ad without changing the central idea.
Examples include:
- A different opening line
- A different product photo
- A new background
- A shorter caption
- A different call to action
Variations can improve execution, but they usually give the customer the same basic reason to buy.
Iteration
An iteration improves an existing ad based on something you learned.
Suppose people watch the first half of a product demonstration but leave before the main benefit appears.
The next version might move that benefit into the opening seconds.
Or perhaps a customer testimonial performs well, but viewers drop before the strongest quote. The next iteration could lead with that statement.
Iteration should have a reason behind it.
“We needed three new ads this week” is a production schedule, not a learning.
Angle
An angle changes the reason the customer should care.
The product stays the same, but the buying motivation changes.
A travel bag might be advertised around:
- Fitting more into hand luggage
- Keeping belongings organised
- Avoiding checked-bag fees
- Durability
- Fast airport travel
- Looking better than a traditional travel backpack
Those are different arguments.
Concept
A concept is the larger creative idea that delivers the message.
Examples include:
- A product demonstration
- A customer testimonial
- A before-and-after scenario
- A founder explaining why the product exists
- A comparison with an alternative
- An educational breakdown
- A problem-and-solution story
- A lifestyle scenario
A new hook on the same testimonial is not a new concept.
It is a variation or iteration of the testimonial concept.
Five ads can still be one ad
Imagine an eCommerce campaign contains:
- Two static images
- Two short videos
- One carousel
On paper, that looks diversified.
Now remove the formats and summarise what each ad says.
If every one can be reduced to:
“Our product is high quality and you should buy it.”
you do not have five ideas.
You have one idea displayed five ways.
We find it more useful to audit the message behind each asset.
| Question | What it reveals |
|---|---|
| What problem does this ad talk about? | Whether several customer problems are represented |
| What benefit leads the message? | Whether every ad depends on the same selling point |
| What proof does it use? | Whether the campaign relies only on claims or uses demonstrations, reviews, comparisons, or other evidence |
| What objection does it answer? | Whether the creative addresses reasons customers hesitate |
| Who is the situation relevant to? | Whether the ads naturally speak to different customer contexts |
| How is the idea presented? | Whether formats and concepts are actually different |
If every row produces almost the same answer, adding another visual variation probably will not create much new strategic range.
Creative can help Meta identify the right customer context
Creative does more than attract attention after targeting has already happened.
The ad itself contains information about the product, customer problem, use case, message, and likely relevance.
Meta's recommendation models can use relationships between people and ads when deciding which candidates deserve further consideration.
This makes meaningful creative diversity useful for another reason: different concepts can become relevant to different people for different reasons.
Consider one skincare product.
One customer may care about sensitive skin.
Another wants a shorter morning routine.
Another cares primarily about ingredients.
A fourth is trying to solve dryness while travelling.
One generic “better skin” ad forces all of those motivations into the same message.
Several genuinely different concepts give Meta more possible matches to explore.
Breadth first, then depth
There is no correct number of creatives that every campaign should run.
The required volume depends on budget, product range, markets, audience size, sales volume, and how many useful messages the brand actually has.
Instead of chasing a fixed number, we think about the creative library in two dimensions.
Breadth
Breadth means testing meaningfully different ideas.
That can include different:
- Customer problems
- Benefits
- Angles
- Concepts
- Proof points
- Objections
- Use cases
Depth
Depth means developing promising ideas further.
If a product-demonstration concept works, do not immediately abandon it because the campaign “needs something new.”
Test better hooks. Change the pacing. Improve the demonstration. Show a different use case. Add stronger proof. Try another creator.
Breadth helps discover ideas. Depth helps improve the ideas that deserve more investment.
A creative programme needs both.
Do not diversify randomly
Creative diversification is not an instruction to produce unrelated ads.
The concepts should come from actual customer and product knowledge.
Useful sources include:
- Customer reviews
- Support questions
- Sales conversations
- Product reviews and returns
- Search queries
- Comments on existing ads
- Customer interviews
- Reasons people choose one product over another
- Objections that repeatedly appear before purchase
Suppose customers repeatedly ask whether a sofa can fit through a narrow doorway.
That is not merely a support question.
It may be a creative opportunity.
A video showing the delivery process and modular construction may remove a concern that a generic lifestyle ad never addresses.
This is where human strategy still matters.
Andromeda can improve how Meta matches an ad with a person. It does not conduct your customer interviews.
Good creative cannot compensate for bad conversion data
Creative gets much of the attention in discussions about Andromeda, but Meta still needs reliable information about what happens after people see and click ads.
For eCommerce, that usually means accurate events around the buying journey.
Depending on the setup, those can include:
- Product views
- Add to cart
- Checkout
- Purchases
- Purchase value
- Product identifiers
The Meta Pixel can send browser-side events, while Conversions API provides another way for businesses to send marketing event data to Meta.
The objective is not to send every possible event.
It is to make sure the events used for measurement and optimisation represent what actually happened.
A duplicated purchase event does not become more accurate because Andromeda is sophisticated.
Neither does an incorrect purchase value.
The recommendation system can only work with the information available to it.
Broad targeting does not mean no targeting strategy
The discussion around Meta automation sometimes swings too far.
One side tries to control every interest manually.
The other says targeting no longer matters at all.
Neither is particularly useful.
Automated audience systems can give Meta more freedom to identify potential buyers, but businesses still have constraints that the algorithm should not decide on its own.
Examples include:
- Countries the business can actually serve
- Language requirements
- Age or legal restrictions
- Markets with different prices or margins
- Products unavailable in certain regions
- Customer exclusions
- Specific acquisition or retention objectives
Those are business rules, not predictions about which interest someone should have.
The role of targeting strategy is increasingly to define meaningful boundaries and useful signals rather than manually describe every person Meta is allowed to reach.
Campaign consolidation has a purpose, but it is not a religion
Another consequence of greater automation is the move toward simpler account structures.
Fragmentation can become a problem when several ad sets:
- Optimise for the same result
- Target substantially similar people
- Use similar creative
- Split a limited budget
- Have no meaningful business difference
In that situation, consolidation can give the delivery system a larger pool of data and budget to work with.
But “simpler” does not mean “put the whole business into one campaign.”
Separation still makes sense when there is a real reason.
Different markets may have different:
- Languages
- Margins
- Prices
- Inventory
- Legal requirements
- Offers
- Customer acquisition targets
A good campaign structure is not necessarily the one with the fewest campaigns.
It is the one where every separation has an explainable purpose.
Creative fatigue and creative limitation are different problems
Advertisers often use “creative fatigue” as a catch-all explanation when performance drops.
Sometimes the winning ad has genuinely been shown too often to the same potential customers.
But sometimes the problem is that the account never had much creative range in the first place.
If every ad communicates the same promise, launching another version of that promise may briefly redistribute delivery without introducing a new reason to buy.
That is a creative limitation problem rather than simply a fatigue problem.
Signs that the creative system deserves investigation include:
- Most active ads communicate the same benefit
- New launches mainly change hooks, colours, or creators without changing the argument
- One concept carries most of the account for months
- There are no ads addressing important customer objections
- Every ad assumes the customer already understands the product
- The brand has no creative for different use cases
- Creative production is based mainly on social trends rather than customer information
The answer is not necessarily “make more ads.”
It may be “find more useful things to say.”
A high click-through rate does not prove the creative is good
Creative should not be evaluated only at the ad level.
An ad can produce a lot of clicks because it is entertaining, surprising, or provocative while attracting people who are unlikely to purchase.
Another ad may have a lower click-through rate but bring customers who convert at a better rate and generate more profitable orders.
Follow the result further through the funnel.
Depending on the objective, that can mean looking at:
- Click-through rate
- Landing-page views
- Add-to-cart behaviour
- Checkout rate
- Purchase rate
- Cost per acquisition
- Purchase value
- New-customer rate
- Margin or contribution where available
The best creative is not always the ad that generates the cheapest click.
It is the creative that helps produce the business result the campaign exists to generate.
The landing page has to continue the argument
A strong ad can still fail after the click.
If a creative leads with “easy assembly in five minutes” but the product page never explains assembly, the customer has to start their evaluation again.
If an ad promotes free delivery but shipping charges appear unexpectedly later, the message and purchase experience are working against each other.
Creative performance therefore depends partly on what happens after Meta sends the visitor away.
We look for continuity between:
Ad concept → customer expectation → landing page → offer → checkout
This is also why a drop in advertising performance should not automatically trigger another targeting test.
The problem may be further down the journey.
Andromeda still does not know your margins
Meta can estimate which ads and people are more likely to produce the optimisation event it receives.
It does not automatically know every commercial detail behind that event.
For example, the platform may not know that:
- One product has much lower margin after fulfilment
- A high-selling item is about to run out of stock
- A certain market has expensive returns
- Existing customers are generating most of the revenue
- A promotion is deliberately sacrificing first-order margin
- A product is strategically important despite lower immediate sales
- The current spend level creates a cash-flow problem
- A high-performing creative is inconsistent with how the brand wants to communicate
This is where the phrase “let the algorithm decide” reaches its limit.
The algorithm can make delivery decisions inside the objective it has been given.
People still need to define whether those results make business sense.
A practical creative diversification framework
When we review a Meta account, we would rather map the existing creative library before immediately producing another batch of ads.
A simple framework looks like this:
| Layer | Question | Example |
|---|---|---|
| Customer problem | What is the customer trying to solve? | “Packing for a weekend without checking a bag” |
| Angle | Why should they care about this product? | “Save airline baggage fees” |
| Concept | How will we communicate that argument? | Side-by-side packing demonstration |
| Execution | How will the concept look and sound? | Creator video, product demo, static comparison |
| Variation | What small element are we changing? | Opening hook or first frame |
| Iteration | What are we improving based on previous performance? | Show the packed bag earlier because viewers dropped before the reveal |
This helps prevent a common production problem: jumping straight from “we need more creatives” to “make five more videos.”
First decide what those videos need to say.
What strong Meta Ads management looks like now
The work has not disappeared because Meta automates more of the delivery.
It has moved.
Less value may come from manually creating dozens of similar interest-based ad sets.
More value comes from decisions such as:
- What conversion event should Meta optimise for?
- Is the tracking reliable?
- What business constraints need to remain fixed?
- Which creative ideas have not been tested?
- Which ideas deserve more iterations?
- Does the offer still make sense?
- Does the landing page support the ad?
- Are new customers commercially valuable?
- Are margins and stock compatible with current spend?
Those are not decisions Andromeda can make for the business.
They are the inputs that determine whether Meta's increasingly capable recommendation system has something useful to optimise.
Do not replace audience obsession with creative obsession
There is a risk in turning the current shift into another oversimplified rule.
“Audience targeting is everything” was never universally true.
“Creative is everything” is not universally true either.
A campaign can fail with excellent creative because:
- The product is poorly priced
- The offer is weak
- The checkout is broken
- The landing page is slow
- Purchase tracking is wrong
- Margins cannot support the acquisition cost
- Stock is unavailable
Creative diversification matters because it gives Meta more meaningful options when deciding which ads may be relevant to different people.
It does not remove the rest of the eCommerce system.
Andromeda changed the retrieval system, not the need for strategy
Meta Andromeda gives the platform a much more capable way to retrieve potentially relevant ads from an enormous pool of candidates.
That makes narrow manual audience engineering less central to many campaigns and makes the quality of the available ads more important.
But the practical lesson is not simply “go broad and make more creatives.”
It is:
Give Meta room to find customers where that makes sense, then give it genuinely different, useful messages to choose from.
Build breadth by testing different customer problems, angles, concepts, and proof.
Build depth by improving the ideas that show promise.
Keep tracking accurate.
Maintain the business constraints the platform cannot know for you.
And judge the result through the economics of the actual business rather than platform metrics alone.
At Creative Brackets, we treat Meta Ads as part of the wider eCommerce system: campaign structure, creative direction, tracking, landing pages, offers, customer acquisition, and the commercial result after the purchase.
If your campaigns keep spending but another audience test is not fixing the problem, the constraint may be somewhere else.
FAQ
1. What is Meta Andromeda?
Andromeda is Meta's personalised ads retrieval system. Retrieval is an early stage in the ad recommendation process that reduces a very large pool of eligible ads to a smaller set of relevant candidates. More sophisticated ranking systems then evaluate those candidates before final delivery.
2. Does Andromeda mean interest targeting no longer works?
No. The useful conclusion is not that manual targeting has become useless. Meta's automated audience and recommendation systems can now take on more of the work of identifying likely responders, which can reduce the need for excessive interest segmentation. Business constraints, exclusions, markets, customer strategy, and some targeting decisions still matter.
3. What does creative diversification mean?
Creative diversification means giving the campaign meaningfully different ideas rather than only producing visual variations. Different angles, concepts, customer problems, proof points, objections, and use cases create more strategic diversity than changing a background colour or the first sentence of the same ad.
4. How many creatives should a Meta campaign have?
There is no universal number. The appropriate amount depends on budget, product range, market size, creative capacity, and how many useful concepts the brand has to test. Ten near-identical ads can provide less real diversity than three strong ads built around clearly different ideas.
5. Does broad targeting mean Meta decides everything?
No. Advertisers still define the campaign objective, conversion setup, budget, creative, offer, and important business constraints. Broader targeting gives Meta more freedom within that setup; it does not give the platform knowledge of every margin, stock issue, strategic priority, or customer-quality concern inside the business.
6. What is the difference between a creative angle and a concept?
An angle is the reason the customer should care, such as convenience, durability, price, or solving a specific problem. A concept is the larger idea used to communicate that angle, such as a product demonstration, customer testimonial, comparison, educational video, or founder story.
7. Should we stop testing audiences and only test creative?
No. Audience tests can still answer useful questions. The point is to avoid using another audience as the automatic response to every performance problem. Before rebuilding targeting, check whether the creative library contains genuinely different ideas, whether the conversion signal is reliable, and whether the post-click experience supports the ads.