Goodbye to manual segmentation: here’s how Andromeda, Meta’s new AI, works

Marketing teams are increasingly investing in new ad formats and platforms. TV, radio, local digital media, or even paid newsletters are part of the mix.
The problem arises when those platforms don’t offer an API to integrate directly with analytics tools. The consequence: manual reports in Excel that remain isolated and make it difficult to get a complete view of the investment.

What is Andromeda?

Meta introduced Andromeda as its new internal machine learning engine for ad delivery: a system that processes millions (or tens of millions) of ad candidates, narrows them down to thousands, and then decides which one to show to each person in real time.

Among its technical innovations: deep neural network architectures, specialized hardware (such as NVIDIA’s Grace Hopper Superchip or Meta’s own MTIA), and a hierarchical indexing logic to handle the massive volume of possible creative assets.

Additionally, it operates in a layer prior to the final ad ranking, filtering millions of initial options down to only those most likely to perform, thereby improving the efficiency of the delivery system. This represents a direct update to the old retrieval engine, which could no longer handle the current scale of formats, ads, and user signals.

Meta also reported that Andromeda runs on high-performance internal infrastructure, integrating proprietary hardware with advanced GPUs to accelerate machine learning and reduce latency in ad delivery.

The effects already reported:

• Improvements in ad quality (~8%),
• Improvements in system recall (~6%),
• ROAS growth (~22% in some cases with Advantage+), according to industry sources.
  

Why is this a relevant change for marketing teams?

It’s no longer “just” about optimizing audiences or bids. With Andromeda, the focus shifts to price, creative volume, format diversity, and automated delivery.
Traditional micro-segmentation-based campaigns may fall behind. Meta seems to favor broader structures with many creative variations, so that the algorithm can “choose” the optimal combination.

This implies that interest- or behavior-based segmentation strategies lose weight compared to data quality and creative variety.
In this new environment, manual optimization gives way to collaboration between AI and the human team, where the marketer provides data, brand, and content, and AI delivers scale, speed, and continuous optimization.

In other words: “train the AI to work with you.”

And what does this mean for you as a marketer?

Some recommendations we can adopt right now (and that Bunker DB can help you with):

1. Diversify your creative assets: don’t just use a single image or video; create versions that address different angles, problems, or people.
2. Target broad audiences: instead of hyper-segmenting, give the algorithm room to learn who really responds; then adjust if necessary.
3. Data, connectivity, and integration: with so much “intelligence” in delivery, it’s crucial that your tools connect seamlessly—from measurement and attribution to your dashboards.
4. This is where our Connection Center module in Bunker DB comes in: you centralize data, audit connections, see which networks are active, and power that “human work + AI.”
5. Simplify campaign structures: fewer campaigns, fewer ad sets; let automation test multiple ad combinations within cleaner structures.
6. Ensure robust conversion data: verify that the pixel, Conversions API, and key events are properly configured; without quality data, Andromeda cannot learn effectively.

Evaluate new metrics: beyond CTR, look at CPA, ROAS, LTV, and how performance evolves across the broad groups the system identifies.

Strategic control vs. tactical control

Although the system operates with automation, you are still the guide.

Set the strategy, define the objectives, and monitor the results.

AI optimizes, but it doesn’t decide the “why.”

How do we see it in Bunker DB?

At Bunker DB, we believe that Andromeda represents a great opportunity for marketing teams who want to be more efficient, make better use of their data, and deliver better results.

Our platform—and specifically our Connection Center module—provides teams with tools to integrate multiple platforms (Google, Meta, TikTok, LinkedIn, and X, among others), centralize data, audit connections, and present insights quickly.

And by combining it with ADA AI, our analytical assistant, teams can converse directly with their data, create charts or insights on their Meta campaign performance, and train their own optimization decisions.

|

ADA AI - Bunker DB

Thus, when Meta delivers with Andromeda, you have the data “ready” to interpret, compare, and optimize.

You manage the strategy, AI handles the execution, and we give you global visibility to close the loop.

In summary

Meta is rewriting the way digital advertising is done with Andromeda.
It’s no longer enough to have good targeting or a good ad: the ecosystem demands lots of creative assets, connected data, broad audiences, and a holistic view.

And in that context, the ideal is to add Bunker DB to your team: work together to make Meta’s AI work for you, not against you.