Andromeda is Meta's new ad delivery model, and by the time you're reading this, it's been live across most accounts for a few months. It changes how Meta selects who sees your ads, how creative gets matched to audiences, and how aggressively the algorithm overrides the targeting you set.
We ran the same campaigns on identical budgets pre- and post-Andromeda across six brands in our portfolio. The verdict: it works, but only if you change how you brief it.
What actually changed.
Three structural shifts matter:
- Audience targeting got mostly ignored. Advantage+ became the default surface. Detailed targeting still exists, but Andromeda overrides it more aggressively than the previous model. Your audience definition is now closer to a suggestion than a constraint.
- Creative became the primary signal. Andromeda weighs creative hooks, formats, and aspect ratios more heavily when deciding who to show ads to. Diverse creative = wider delivery. Single-creative campaigns starved fast.
- Conversion event quality matters more. Server-side tracking went from helpful to non-negotiable. Brands with degraded pixel signal saw immediate CPM and CPA inflation.
What to test first.
Five things, in this order:
- Consolidate ad sets. Drop down to 1–3 ad sets per campaign. Andromeda needs scale to optimize and gets confused by over-segmentation.
- Audit your CAPI implementation. If you're not running server-side with deduplication and event match quality > 7, fix it before you change anything else.
- Ship 8–12 creative variants per week. Hooks, formats, founder, UGC, motion. The volume itself is the lever.
- Set learning budgets aggressively. Andromeda needs ~$200/day per ad set minimum to learn properly. Below that and you're in a permanent learning phase.
- Watch frequency, not just ROAS. The biggest tell of a campaign past its prime is rising frequency with declining CTR, refresh creative before ROAS drops.
What didn't change.
The fundamentals still matter. Offer strength beats creative. Margin discipline beats topline growth. Mobile-first design beats desktop-first. Andromeda just made these things more obvious because the algorithm punishes weakness faster.
If you're seeing ROAS drift on Meta, the Andromeda excuse is real, but it's also a useful diagnostic. Brands with a strong offer, healthy tracking, and an active creative pipeline are coming out of this transition ahead. The rest are losing share.