Recently, you may have had your feathers ruffled by Meta’s new Andromeda update. Depending on where on the internet you’re looking, Andromeda may be a seismic shift that’s completely upending the way we build ads, or it may just be a molehill dressed up as a mountain.
If your charity is running Meta Ads as part of its fundraising and marketing efforts, this is relevant to you.
So what is Andromeda, exactly, and what do charities need to know about it?
Andromeda is the latest shift in a push towards automation
Andromeda is an engineering and infrastructure upgrade to how Meta’s algorithm decides what ads get shown to which users. It capitalises on:
- larger deep learning models
- better GPU / AI hardware
- faster inference
- larger candidate pools
It’s a pretty big shift, but also one that’s a natural next step in Meta’s push towards more AI-driven, signal-based automation.
In summary: if your ad strategy has already been evolving over the last few years to accommodate changes in Meta’s algorithm, then this update is not seismic for you.
But if you’re still running ads the same way you were three or four years ago, then this update will probably feel pretty huge.
Meta’s ad retrieval system needed an update
In order for your ads to appear on Meta, its machine learning has to go through a process to retrieve ads, then rank them in an auction.
Previously, advertisers could define who they wanted to target by interest, demographic or behaviour, or by customised lists. Meta matched these targeting requirements to individual users on the platform. Campaigns would enter an ad auction in which ad units competed against those from other advertisers who were targeting the same audience. The auction picked the winner, and ads were placed.
Andromeda upgrades Meta’s ad retrieval system, and shifts performance to being more signal-based. The ad auction still exists, so factors like predicted action rate, ad quality signals and bid and value optimisation are still key to campaign planning.
To strengthen signals, biggest testing pools and more creative diversity is needed. But we’ll come onto that in a moment.
Why did Meta make the change?
The broader industry context is that Meta’s come under fire in the last decade over its handling and sharing of user data.
The same targeting precision that’s allowed us to run successful fundraising, activism or awareness campaigns has also allowed other advertisers to influence political elections and decision-making. The same algorithms that help us to capture audiences with strong affinities are the same ones that have allowed tribalism and polarization to develop.
Platforms are under pressure to resolve this, and one way we’ve seen this play out on Meta is in the gradual disappearance of highly specific targeting options.
In the past, you could target someone who likes Oxfam, listens to grime music and is likely observing Ramadan. But now these signals are disappearing.
While this is on balance an ethical development, it does throw up some practical challenges when it comes to ad delivery.
AI has become more threaded through advertising
AI isn’t actually new to the advertising space. For years now, what ads you see, when and on what platforms has been decided by machine learning.
But in the last few years, automation tools like Advantage+, generative copy and dynamic creative have made it easier to create and test at scale – which is seriously putting Meta’s machine learning to the test.
Where you once might have had 3 tightly targeted ad sets, and 2-3 creative within each, advertisers are now throwing a huge number of ad sets and creatives at a single campaign. Meta now has to process millions of variants. The system has become a bottleneck.
Meta is in a content recommendations arms race
We only have to look at TikTok to understand how addictive, and therefore profitable, a good content recommendations algorithm is. To retain market share, Meta needs to catch up.
Instagram has already seen some pretty big shifts in content recommendations, and Meta is soon to follow.
The key elements that underpin good content are:
- They’re personalised
- They’re contextual
- They’re behaviour-driven
That requires analysing huge signal sets about both users and creative content. So Meta upgraded the retrieval engine, expanded deep learning models and improved infrastructure.
Andromeda is shifting the process to being signal-led
Andromeda is Meta’s new AI ad-retrieval engine. Its job is to ask, before the auction starts, “Which ads should we consider showing this person?” The shift moves from “Who should see this ad?” to “Which ad should this person see?”.
Before, Meta retrieved a set of eligible ads and entered those into the auction. Andromeda improves how Meta retrieves and scores candidate ads before they enter the auction.
- User opens Facebook/Instagram
- Andromeda scans eligible ad candidates
- It retrieves a shortlist of likely relevant ads
- Those ads enter the auction
- Auction chooses the final ad
How big the shift is depends on how up-to-date your current approach is
If you’ve been steadily improving and evolving your ad delivery strategy, testing Advantage+ settings, using broad audiences, and developing a diverse range of creative in format and design, you’re already leveraging machine learning. So this update doesn’t really change that much.
But if you’re still setting ads up the way we were doing it 4 years ago, with narrow groups of audiences, pen profiles, and 2-3 variants of still images with slight differences between them, then Andromeda is going to be a bit of a shock to your system.
What signals does Andromeda use?
While Meta doesn’t explicitly say, it’s possible that these play a role (based on the engineering notes on Andromeda):
User signals
- browsing behaviour
- recent actions
- content interactions
- engagement patterns
Context signals
- time of day
- content being viewed
- session behaviour
Creative signals
- visuals
- pacing
- emotion
- message
- Format
What has changed?
1. The system learns better from broader targeting
Stacking interests and creating highly specific audience segments is less necessary an approach than it used to be. (More on this in a later blogpost!)
You could assume people might respond to an ad, but their actual signals might indicate they’re not the right audience. Broad targeting helps Andromeda figure it out more effectively.
2. Creative diversity matters
A small number of creative variations, or a large number that all look really similar, means a higher chance they’re all filtered out.
Fewer, more diverse creative gives you a better chance of having a few succeed in the auction.
3. Campaign structures have been simplified
Large, broad campaigns often perform better because they give the algorithm more learning signals, and fewer specificities to take into account.
4. Creative lifecycle is faster
Meta’s learning and delivery cycles appear to have sped up. So advertisers have to be iterating and refreshing creative much more frequently than they were before.
What should charities do now?
1. Stop over-engineering targeting.
Instead of creating loads of ad sets that are all specifically targeted, create broad audiences and combine them when there is likely to be lots of overlap in interest or behaviour.
2. Invest more in diverse creative in greater volumes
Meta’s model learns from signals, and signals are stronger when variants are more diverse. This means that your copy and creative should have multiple hooks, different visual styles, and use different formats – from videos, to carousels, to static images.
Aim to have enough prepared that you can be responsive to performance.
3. Maintain breadth
Once, common convention was to start broad and then narrow down – retire creative that isn’t performing and develop more versions of the ones that are.
In the new system, this risks killing creative variation too early. We want to give the system time and flexibility to learn from variations. So replace your creative regularly, yes, but continually with new, fresh ideas – don’t narrow down into the individual variants that are working well.
4. Feed the model good conversion signals
Tracking accuracy is vital, because Meta needs healthy signals to be able to optimise. It’s more important than ever that your infrastructure is place, and that you have both cookie-based and server-side tools to record and analyse behaviour.
This includes Meta Pixel, Conversions API and consistent conversion volume.
Meta hasn’t said this explicitly, but from our experience, this also means being active on the platform year round, so that you’re feeding it with signals outside of just conversions. This puts you in a better position to compete at peak times of the year.
Note: if your charity works in health, financial hardship, or any other sensitive category, Meta’s January 2025 policy changes may have restricted your pixel from sharing conversion data entirely. That means the tracking infrastructure described above may not be available to you.
See our guide to navigating those changes before implementing anything in this section.
5. Simplify campaign set up
A more effective campaign structure that gives the algorithm maximum flexibility could be:
- 1 campaign
- 3-4 ad sets
- many diverse creatives
- broad targeting
A separate but related issue: Meta’s 2025 policy changes directly affecting charity campaigns
Andromeda is one part of a broader shift at Meta. Running alongside it are a series of specific policy changes – not algorithm updates, but rule changes. These have a direct and significant impact on how charities can use Meta’s advertising tools.
These are worth understanding separately, because they affect charities in ways that commercial advertisers mostly don’t feel.
Read our full guide on navigating Meta’s 2025 advertising changes.
How we can help your Meta campaigns
Meta advertising is shifting fast, and keeping up with both the algorithm changes and the policy changes takes time that most charity digital teams don’t have spare.
If you’d like a second pair of eyes on how your campaigns are set up, from creative structure, tracking, targeting, or navigating the restrictions, we can help.
Find out more about our Meta Ads service for charities, or drop Matt an email.
