Understanding Meta Andromeda: Why Your Creative Is Now Your Targeting

For years a simple logic applied in Meta Ads: whoever clicks together the right audience – interests, lookalikes, custom audiences – wins. This logic is outdated. With the AI system Andromeda, Meta has fundamentally rebuilt the way ads find their path to the right user. In short: it’s no longer your audience settings that decide who sees your ad, but your creative.

If you take paid social seriously, this is the most important structural change you should align your account to. In this guide you’ll learn what Andromeda is, which three shifts result from it, what a clean setup looks like and which mistakes you should avoid.

What is Meta Andromeda?

Andromeda is a machine-learning system that works at the so-called retrieval level. This level lies before the actual auction. Its task: from the huge pool of possible ads, make a manageable preselection for each individual ad placement – that is, decide which ads even enter the auction.

The decisive thing is how Andromeda makes this preselection. The system reads creative signals: the visuals, the hook, the copy and the context of an ad. From this it deduces for whom this ad could be relevant – and effectively predicts the right audience itself. Your manual audience settings thereby go from being a control to a rough guardrail.

According to Meta, this approach brings measurable improvements: the company cites around 6% better retrieval recall and about 8% higher ad quality. For you this means: the system gets better at bringing the right ad together with the right user – provided you feed it the right signals.

An image helps with understanding: picture the retrieval level like a bouncer who, out of millions of possible ads, creates a short guest list for each free ad placement. Only those on this list get into the auction at all and bid for the spot. Andromeda is this bouncer – and it doesn’t read your audience settings but your creative to decide who makes the list. This means a large part of the control you used to exercise in the interface migrates into the ad itself.

What concretely changes: three shifts

1. From interest to creative signal

The first shift is the most fundamental. In the past you told the algorithm via interests and lookalikes whom it should address. Today Andromeda reads your creative and decides itself who the right person is. The creative is thereby no longer just a message but a targeting instrument. What that means in practice – producing creatives in volume and variety – we go into in the guide to AI creatives for Meta & TikTok Ads; as a service we bundle it in Creative Engineering.

In practice this means: a creative that emotionally addresses a certain audience is assigned by the system to exactly that audience – entirely without you defining it manually. Interest stacks and lookalikes thereby lose effect. They rarely do harm, but they deliver less and less, because the system finds the relevant users via the creative signals anyway.

2. From many ad sets to few, broad ones

The second shift concerns your account structure. In the old world, fragmentation was a virtue: many ad sets, each with a narrow audience, to retain control. Under Andromeda this is counterproductive.

As a best practice you work with broad audiences and set Advantage+ Shopping as the default. Few ad sets with more budget beat fragmented structures, because each unit then gets enough conversions and signals to stabilise the system’s learning. If you spread the same budget across ten narrow ad sets, each one starves for data – and the AI can’t play out its potential.

3. From click tracking to signal quality

The third shift is the most invisible and at the same time the most consequential. When Andromeda decides based on signals, then the quality of these signals is the real lever. A brilliant creative is of little use if your tracking doesn’t cleanly report back to the system what happens after the click.

The fundamentals of good signal quality are concrete: the Pixel and Conversions API run in parallel, so events still arrive even when browser tracking is blocked. Event Match Quality should be above 7. And you define a prioritised conversion event that the system can clearly optimise towards, rather than confusing it with contradictory goals. This is exactly where it’s often decided in practice whether an account scales or gets stuck – and exactly where we as a Meta Ads agency start first.

The new setup playbook

If you take these three shifts seriously, a clear setup emerges:

• Broad instead of narrow: Start with broad audiences and Advantage+ Shopping. Let the system find the right users via your creatives, instead of restricting it upfront.
• Consolidated instead of fragmented: Use few ad sets with sufficient budget, so each unit gets enough signals.
• Signal quality first: Set up Pixel and Conversions API in parallel, aim for an Event Match Quality above 7 and define a prioritised conversion event.
• Creative as control: Treat your creative as what it has become – your most important targeting instrument. Variety of genuine angles beats variety of audiences.

This point about signal quality is also the reason why creative production has become the actual bottleneck. When the creative does the targeting, you need systematically tested, substantively different ads – no gut feeling. How to set up a scalable process for this, you can read in our guide to Creative Testing for Meta Ads.

Your product data also plays into it: Advantage+ Shopping is only as good as the catalogue that feeds it. A cleanly enriched feed delivers additional signals to the system – how this works, we show in the article on our Feed Hacking.

Common mistakes

Three patterns we see again and again when accounts are still run by the old logic:

Audiences too narrow. Whoever keeps building interest stacks and narrow lookalikes out of habit artificially constrains the system and gives away reach that Andromeda would tap better itself anyway.

Too many ad sets. Fragmented structures spread the budget so thinly that no unit collects enough signals. The result: unstable delivery and a system that never properly gets out of the learning phase.

Neglected signal quality. An account with patchy tracking gives the system false or incomplete feedback. Then Andromeda optimises against a distorted picture – and the best creatives can’t show their potential. Because under automation the question of what you’re even optimising for gets louder anyway, a look at modern measurement beyond platform ROAS in our article MER instead of ROAS is worthwhile.

Too much fiddling. Consolidated, broadly steered campaigns need some calm at the start. Whoever adjusts budgets and settings daily keeps throwing the system back into the learning phase and prevents exactly the stability it needs under Andromeda. Give new structures time to collect enough conversions before you judge. Premature interventions are one of the most underestimated efficiency killers – they feel like control but cost learning progress. As a rule of thumb: if you change something fundamental daily, you never measure a clean state but always only the transition.

What this means for your account

The short version: shift your energy from audience research to creative production and signal quality. These are the two levers that actually remain in your hands under Andromeda.

Concretely this means fiddling with audiences in the Ads Manager less often and more often producing new, substantively different creatives and testing them cleanly. It means consolidating your account structure instead of fraying it. And it means treating your tracking as a foundation, not as a tiresome obligation.

Whoever flips these priorities works with the system instead of against it – and that is exactly the difference under Andromeda between an account that scales efficiently and one that, despite budget, doesn’t get anywhere.