Meta can now consider more advertising options and understand more of the behaviour behind a purchase. For D2C brands, the implications go well beyond producing more creative.
A customer sees one of your ads and keeps scrolling.
A few days later, they watch a product deconstruction, then a UGC ad, then a customer testimonial ad, then a founder message, then a ‘POV: you’ve just… <insert customer problem you are solving>’ ad… then they purchase.
Meta's recent AI advances are making its systems more capable of recognising signals across that kind of journey, in more of a linear pattern. And the latest Andromeda announcement represents an important step in how many advertisements Meta can evaluate in the first place.
The message for D2C businesses isn't necessarily 'make more ads'. It's time to think even harder about which creative messages customers need, how they work together and whether those interactions ultimately generate profitable growth.
What is Meta Andromeda 2.0?
Andromeda is Meta's AI-powered ad retrieval system. When an advertising opportunity appears on Facebook or Instagram, Meta needs to identify the most promising potential ads from a huge eligible pool.
Retrieval is the first stage of that process. Its job is to narrow the options quickly so more detailed ranking and auction processes can decide what is ultimately shown.
Introduced in 2024, Andromeda was already designed for that task. On 28 October 2026, Meta announced its next generation, supported by a redesigned machine-learning architecture and GPU pipeline.
'Andromeda 2.0' is useful industry shorthand, but it isn't a new feature advertisers need to switch on in Ads Manager.
1. Meta can now score every eligible ad at retrieval
This is arguably the most important and most overlooked part of the announcement.
Earlier retrieval systems faced a trade-off: there were too many possible ads to apply especially sophisticated scoring across the whole eligible pool within the time-frame available. They relied on efficient ways of narrowing the candidates.
Meta says the new system can score every eligible ad at the retrieval stage, with performance approaching that of its more detailed ranking models.
In production, Meta reports throughput exceeding one billion ads per second on a single GPU server, around 100 times its 2024 system. It also attributes a 2.4% increase in Facebook conversions since 2025 to Andromeda improvements.
These are Meta's platform-level figures, not a forecast of the uplift an individual account should expect.
The important distinction: being scored is not the same as being shown. Ads must still meet eligibility and targeting conditions, and the system still selects the strongest candidates for later ranking and auction stages.
But an advertiser's relevant creative has more opportunity to be meaningfully evaluated rather than eliminated through a more constrained early-stage process.
2. Meta can understand more of the engagement journey
The second development is about how Meta interprets customer behaviour.
In the latest Andromeda algorithm, Meta describes incorporating sequential ad-engagement events, changing interests and signals from engagement across different surfaces.
Meanwhile, its broader ranking and recommendation research, including GEM, has expanded the ability to learn from the order and timing of customer actions.
This is important, right… Andromeda is responsible for retrieval; GEM and related ranking models perform different jobs. It would be misleading to credit every improvement in sequence modelling to Andromeda alone.
Meta's August 2026 research found that modelling a mixture of behaviour - such as views, clicks and purchases - was more useful than modelling only one type of action.
In other words, engagement isn't necessarily just a collection of isolated events. The relationships between interactions can help predict what a customer may respond to next.
Linear content strategy is somewhat back, you as the marketer just aren’t in control like you once were (back in the day).
A hypothetical customer journey might look like this:
Discovery: a customer watches a short video introducing your product natively and on current .
Consideration: they engage with an ad explaining how it works.
Reassurance: they click a creative featuring customer reviews.
Purchase: they convert after seeing UGC social proof.
Meta's models can make better use of patterns like these when predicting relevance.
However, this does not mean Meta has announced a fixed ad-sequencing product that automatically serves every customer those four creatives in that order. Nor does it give marketers a transparent report of every engagement's incremental contribution.
So perhaps think of it as better prediction from behavioural sequences not necessarily guaranteed, marketer-controlled storytelling.
3. Andromeda isn't working alone
It's worth separating three related developments that are often bundled together in industry commentary.
Andromeda - retrieval: evaluates the eligible ad pool and identifies promising candidates efficiently.
GEM - broader recommendation intelligence: learns from rich histories of engagement and supports increasingly sophisticated predictions across placements and objectives.
GenRecs - semantic understanding: helps Meta reason about what an ad or product represents, rather than relying exclusively on its existing engagement history. That is particularly relevant when a new product or creative has little performance data.
Together, these capabilities are moving Meta further towards interpreting not just who engages, but what the advertisement communicates and how interests evolve.
For founders, the creative strategy is becoming more important, not less.
4. The creative opportunity is bigger than volume
Creative diversification remains relevant, extremely relevant in fact. But simply producing 20 versions of the same message doesn't necessarily give Meta 20 meaningfully different propositions to evaluate, in fact it can often cannibalise and bring your account wide CAC up.
Imagine a premium meal delivery brand.
The business could test four distinct customer motivations:
Convenience - reclaim time during a busy working week.
Performance - make nutrition easier around exercise and training.
Quality - understand ingredients, taste and the eating experience.
Value - compare the cost benefits with conventional shopping behaviour.
Then it could think about the role each creative plays.
A discovery ad might introduce the problem and the brand. A demonstration might help somebody understand the product. Customer testimonials might address hesitation. A clear offer or product message might help convert somebody already considering a purchase.
These aren't instructions to force a rigid funnel on every customer. Different people will engage in different orders, and some may convert immediately.
Instead, we're creating a connected library of useful messages that gives the advertising system more relevant options across different customer needs and stages of consideration.
The objective is more creative diversity.
Five actions we'd recommend D2C brands take now
1. Map your current creative coverage
Audit every live ad, not simply the number of assets. Identify the different customer motivations, product benefits, objections and stages of consideration represented.
2. Build customer insight into the brief
Use post-purchase surveys, customer reviews, interviews, customer service themes and retention analysis to learn what really influences purchase decisions. This gives your creative strategy something commercially meaningful to test.
3. Create connected, distinct creative concept
Develop a mix of discovery, explanation, proof and conversion-oriented messages. Test genuinely different angles as well as formats, rather than relying on cosmetic variations alone.
4. Keep signals and account fundamentals healthy
Check Pixel and Conversions API implementation, event quality, campaign objectives, catalogue accuracy and the conditions required for ads to be eligible. More sophisticated AI cannot compensate for unreliable conversion signals.
5. Measure business outcomes, not just platform outcomes
Monitor acquisition cost, contribution margin, repeat purchase behaviour and customer lifetime value where appropriate. Consider incrementality tests when decisions or budgets justify them.
The bigger picture for D2C growth
As Meta automates more of the decisions involved in advertising delivery, marketers don't become irrelevant.
Their responsibility just moves.
Less time spent trying to second-guess the system's next targeting decision. More time understanding the customer, developing distinctive propositions, connecting creative messages and deciding what performance actually means for the business and in turn feeding Meta’s algorithm.
Andromeda's expanded retrieval capacity matters. So does the wider progress in sequence and semantic modelling.
But the most valuable response isn't necessarily to increase production overnight or tear apart an account that's already performing.
It's to build a stronger creative strategy around the people you're trying to reach, using what you know about those you have successfully done so in the past.
Better algorithms can give you more opportunities. Better customer understanding helps you make the most of them.
How we can help
Built To Rise supports ambitious D2C brands with hands-on growth marketing leadership, customer-led Growth Diagnostics and practical projects that build the foundations for sustainable growth.
If you're unsure whether your next move should be more paid media spend, a better creative strategy or a clearer understanding of your customer, that's exactly the sort of commercial question we help founders answer.
