What Is Andromeda?
In early 2024, Meta began rolling out Andromeda — a complete overhaul of its ad retrieval and ranking infrastructure. Unlike incremental algorithm updates, Andromeda replaced the core system that decides which ads get shown to which users. Meta's own engineering blog described it as a move from a two-stage retrieval model to a deep learning system capable of evaluating vastly more ad candidates per auction in real time.
The practical result: Meta can now consider a far larger pool of ads for each impression slot, and rank them with significantly more contextual nuance. More signals. Faster decisions. Higher bar for relevance.
Why Your Old Playbook Is Breaking
If your campaigns were structured around the pre-2024 playbook — tight interest stacks, heavy audience exclusions, manual bidding everywhere — you're fighting the new system rather than working with it. Andromeda is optimised to reward signal density, not audience precision.
- Hyper-narrow audiences starve the algorithm of the volume it needs to learn
- Manual bidding overrides that worked pre-Andromeda now interfere with real-time auction dynamics
- Fragmented campaign structures split conversion data across too many ad sets, slowing learning phase exit
- Frequency-based creative refresh cycles no longer apply — Andromeda serves creative more dynamically based on predicted engagement
The Three Structural Shifts That Matter
1. Consolidate. Hard. Andromeda learns faster with fewer, larger ad sets. Merge your audiences into 2–3 consolidated ad sets and let Meta's system find the buyers within them. Campaign Budget Optimisation (CBO) becomes non-optional at scale — it gives the algorithm maximum flexibility to allocate where signals are strongest.
2. Feed it conversion events, not just clicks. Andromeda is a conversion signal machine. The more purchase (or lead) events flowing through your pixel and CAPI setup, the more precisely it learns who to serve. If you're optimising for traffic or reach because your conversion volume is low, you're training the wrong model. Even at low budgets, optimise for the deepest event you can get 20+ of per week.
3. Creative is now the primary targeting signal. With Andromeda's expanded candidate pool, the creative itself tells Meta who to find. A strong hook targeting a specific pain point trains the algorithm far more effectively than a demographic filter. This is why broad targeting — combined with precise creative — now outperforms interest stacking consistently.
What to Do This Week
- Audit your campaign structure — if you have more than 5 ad sets per campaign, consolidate
- Verify CAPI is installed and passing match quality score of 7.0+ in Events Manager
- Switch to Advantage+ audience on at least one campaign and compare ROAS over 14 days
- Test 3–4 creative angles simultaneously — Andromeda distributes spend based on real-time engagement signals, so more variety means faster learning
The accounts scaling fastest right now are those that have embraced what Andromeda is designed to do: find buyers at scale from broad inputs, using conversion data and creative quality as the signal. Stop constraining it. Start feeding it.