Andromeda is the codename for Meta's new ad recommendation model, deployed across Facebook and Instagram in late 2024. Technically, it's a 10,000x scale-up of the previous ranking system. Practically, it's a generational change in how Meta decides which ads to show to which people.
The previous system optimised based on a relatively narrow set of signals: who you targeted, what creative you ran, what conversion event you optimised for. Advertisers had real control over each of these. The model learned, but it learned slowly and within the constraints you set.
Andromeda flips this. The new model processes vastly more signals, learns much faster, and gets dramatically better results when it has room to explore. Tight constraints (narrow audiences, limited creative variations, rigid campaign structures) actively hurt performance now, because they stop the algorithm from doing the work it's designed to do.
Meta hasn't publicised Andromeda heavily because the implications are uncomfortable: the agencies and advertisers who built their craft around the old controls are the ones now hurting most. The brands and operators who lean into the new model are seeing CPA drops, scale unlocks, and reach expansions that weren't possible before.
If your agency is still talking about "finding your perfect customer through interest layering" in 2026, they're running 2024 strategy on 2026 infrastructure. That gap costs you money every week.

