What is Meta’s Andromeda algorithm? The complete guide for advertisers

If you’ve noticed your Meta ads behaving differently over the past year – delivery patterns shifting, audience targeting becoming less predictable, creative performance more volatile – you’re not imagining things.

Meta has fundamentally rebuilt how ads get matched to people. The system driving this is called Andromeda, and understanding it is now essential for any brand spending seriously on Facebook and Instagram.

This guide explains what Andromeda is, how it works, and what it means for your advertising strategy.

What is Meta Andromeda?

Andromeda is Meta’s AI-powered ad retrieval and ranking system. It replaced the previous ads infrastructure and represents the most significant change to how Meta serves ads in over a decade.

The name comes from the Andromeda galaxy – fitting for a system designed to process an almost incomprehensibly large number of potential ad-to-user matches in real time.

“It’s a mixture of hardware and software and AI algorithm that builds on the existing technology,” explains Josh Goldsmith, Head of Social at Nest. “Instead of picking from tens of thousands of ads, it’s actually able to pick from tens of millions of ads – that’s three orders of magnitude greater.”

Here’s the technical summary: Andromeda uses machine learning models to predict which ad will be most relevant to each individual user at any given moment. The scale of change is significant: before Andromeda, the system evaluated thousands of ads in each auction. After Andromeda, it can evaluate tens of millions of ads in fractions of a second.

This happens across Facebook, Instagram, Messenger, and the Audience Network – billions of predictions per day.

The technical details matter less than what this means in practice: Meta is no longer matching ads to audiences. It’s matching ads to individuals.

How Andromeda differs from previous systems

Meta’s old advertising system worked roughly like this:

1. You defined an audience (demographics, interests, behaviours, custom audiences)
2. Meta found users who matched that audience definition
3. Your ads competed in auctions for those users
4. The best-performing ads (by your objective) got more delivery

Targeting was the primary lever. If you could define the right audience, you could reach the right people. Creative mattered, but audience selection was where sophisticated advertisers focused their energy.

Andromeda flips this model.

The algorithm now does the targeting for you. When you use Advantage+ campaigns, broad targeting, or even detailed targeting, Andromeda looks at all eligible users and predicts – at the individual level – who is most likely to take your desired action.

Your audience inputs become suggestions rather than constraints. The system will find people outside your defined parameters if it predicts they’ll convert.

This is why many advertisers have noticed that broad targeting now often outperforms detailed targeting. It’s why Advantage+ Shopping campaigns can deliver strong results with minimal audience inputs. The algorithm has become better at finding your customers than you are.

The three things Andromeda optimises for

Andromeda’s predictions are built around three factors:

1. User value

How likely is this specific person to take the desired action (click, purchase, sign up, etc.) based on their historical behaviour, current context, and predicted intent?

Andromeda has access to signals you’ll never see: browsing patterns, app usage, purchase history across Meta’s ecosystem, time of day behaviour patterns, device switching habits, and countless other data points.

It builds an individual-level prediction model that updates continuously.

2. Advertiser value

What’s the expected value to the advertiser if this user sees this ad? This incorporates your bid, your optimisation objective, and the predicted conversion value.

Higher bids don’t guarantee delivery. A lower bid with a higher predicted conversion rate can win over a higher bid with lower predicted engagement.

3. User experience

Will showing this ad improve or degrade the user’s experience on the platform? Meta needs users to keep scrolling, keep engaging, keep coming back.

Ads that get hidden, generate negative feedback, or cause users to leave the platform get penalised. Ads that generate genuine engagement get rewarded.

Andromeda balances all three factors in real time, for every potential ad impression, billions of times per day.

Why creative is now your primary lever

Here’s the strategic implication that most advertisers haven’t fully absorbed:

If the algorithm handles targeting, creative becomes your main competitive advantage.

“Because the system can handle more ads, we should be feeding it more ads,” Josh explains. “Within a target audience of people who look quite similar, those people are going to resonate with different things. One user might resonate with UGC content. Others might be more interested in product shots with text overlays explaining the benefits – lo-fi content, hi-fi content, lifestyle shots.”

Think about it: every advertiser targeting “women interested in fitness” now has access to the same algorithmic targeting capability. You’re no longer competing on audience definition – Andromeda will find the right people regardless of how you set up your targeting.

What differs between advertisers is the creative you give the system to work with.

And this is where most brands are failing.

Andromeda is built for hyper-personalisation. It can serve different ads to different individuals based on predicted resonance. But it can only do this if you give it options.

If you’re running 10 ads, Andromeda has 10 options to choose from. If your competitor is running 200 ads with genuine variety across hooks, formats, angles, and styles, they’re giving the algorithm 20x more opportunities to find perfect matches.

The algorithm rewards advertisers who feed it volume and variety.

This isn’t speculation. Meta has been explicit about this in their guidance to large advertisers. The system performs better when it has more high-quality creative options to learn from.

What Andromeda needs from advertisers

Based on how the system works, here’s what Andromeda rewards:

High creative volume

The volume benchmarks have shifted dramatically:

Four or five years ago: Brands could succeed with 5 hero creatives. Find what works, scale it, refresh quarterly.

Today: Competitive brands are running 30-50 assets per month minimum. Leaders are running hundreds.

Tomorrow: The trajectory points toward 500+ ads per month for serious advertisers.

This sounds extreme if you’re used to traditional creative production. But Andromeda treats each piece of creative as a hypothesis to test. More hypotheses = faster learning = better performance.

Genuine creative variety

Volume alone isn’t enough. Running 500 variations of the same concept – same hook, same format, different colours – doesn’t give Andromeda new information.

The system needs genuine creative variety:

• Different hooks (problem-focused vs. aspiration-focused vs. social proof vs. urgency)
• Different formats (static, video, carousel, UGC, polished brand content)
• Different angles (features vs. benefits vs. lifestyle vs. testimonials)
• Different visual styles (bright vs. muted, busy vs. minimal, people vs. product)

Each distinct creative approach gives Andromeda a new vector to explore.

Fresh creative

On Andromeda, overcoming creative fatigue means staying ahead of it.

What worked last month is delivering to a shrinking pool of users this month. You need a continuous pipeline of fresh creative to maintain delivery.

Strong conversion signals

Andromeda learns from conversions. The more conversion data you can provide – and the faster you can provide it – the better the system performs.

This means:

• Proper Conversions API implementation
• Optimising for the most valuable conversion events you have sufficient volume for
• Ensuring your tracking captures the full customer journey

Weak or delayed conversion signals handicap Andromeda’s ability to find your best customers.

How to win with Andromeda: the practical playbook

1. Embrace broad targeting

Stop fighting the algorithm with narrow audience definitions. In most cases, Advantage+ audiences or broad targeting will outperform detailed targeting.

Your historical audience definitions were useful when targeting was manual. With Andromeda, they often constrain the system from finding valuable users you’d never have thought to target.

Test broad. Let the algorithm learn. Focus your energy on creative instead.

2. Restructure your account for simplicity

Andromeda performs better with consolidated campaign structures. Splitting budget across dozens of ad sets fragments learning and limits the system’s ability to optimise.

The trend is toward fewer campaigns with more creative variety within each. Advantage+ Shopping campaigns are the extreme version of this – one campaign, broad targeting, let the algorithm figure it out.

3. Invest in creative volume

This is the big one. If you’re currently producing 10-20 ads per month, you need to find a way to produce 10x that.

This doesn’t necessarily mean 10x the budget. It means:

• Using AI-assisted production to create variations efficiently
• Building modular creative systems where elements can be recombined
• Repurposing UGC and influencer content

The brands winning on Meta have built creative engines that can produce at scale.

4. Build a testing infrastructure

With high creative volume, you need systems to:

• Track which concepts are working vs. which variations of concepts are working
• Identify fatigue before it tanks performance
• Feed winning insights back into production quickly

Manual spreadsheet analysis doesn’t scale to 200+ ads. You need automated creative analytics.

5. Close the loop between performance and production

Andromeda rewards advertisers who can respond quickly to what’s working.

If your creative production cycle is 3-4 weeks, by the time you’ve acted on an insight, the moment has passed. The algorithm has moved on. Your competitors who can iterate in days, not weeks, will compound their gains while you’re still waiting for assets.

The brands pulling ahead have closed the loop – performance signals flow directly into production, and new creative ships fast enough to capitalise on what’s working right now.

Common mistakes advertisers make with Andromeda

Over-constraining targeting

Adding too many targeting restrictions limits Andromeda’s ability to find valuable users. Every exclusion, every narrow interest, every demographic constraint shrinks the pool the algorithm can explore.

Under-investing in creative

Spending 90% of your budget on media and 10% on creative is backwards in the Andromeda era. Creative is your primary performance lever. Under-resourcing it is like having a powerful engine with no fuel.

Optimising for the wrong events

If you’re optimising for link clicks but care about purchases, you’re training Andromeda to find clickers, not buyers. Optimise for the conversion event that actually matters, even if it means a longer learning phase.

Changing too much too fast

Andromeda needs stability to learn. Making major changes to campaigns – new creative, new targeting, new budgets – all at once resets learning and tanks performance. Make changes incrementally so you can attribute results.

Ignoring creative fatigue

Running the same ads indefinitely because they “still work” ignores individual-level fatigue. Your aggregate metrics might look stable while delivery is concentrating on an ever-shrinking pool of users who haven’t seen the ad enough times to tune it out.

The future of advertising is individual

Andromeda represents the logical endpoint of digital advertising’s evolution across three eras:

Era 1 (Pre-Internet – One-to-Many): For decades, you made one TV ad, one billboard, and broadcast it to everyone. Production was slow and expensive, but you only needed a handful of assets per year.

Era 2 (Internet Advertising – One-to-Some): Digital let you target segments: women aged 25-34 who like yoga, parents searching for school supplies. Instead of one ad, you needed ten or twenty variations, but the workflow stayed manageable.

Era 3 (AI-Powered Advertising – One-to-One): This is where everything breaks. Meta’s Andromeda engine doesn’t think in segments. It thinks in individuals, serving the right creative to the right person at the right moment.

This shift has profound implications. Targeting as a discipline becomes less important. Creative strategy becomes central. Production capacity becomes a competitive advantage.

The advertisers who thrive will be those who can produce enough high-quality, varied creative to feed the algorithm what it needs – and who can iterate fast enough to compound their learnings over time.

Andromeda isn’t the future. It’s the present. The question is whether your creative workflow is built for it.

FAQ

Not sure how many ads your brand needs to perform on Andromeda? Check out our Meta ads calculator.