If Meta costs have been swinging — CPCs unstable, cost per consult unpredictable, campaigns that used to work now stalling — something underneath changed. In December 2024 Meta rolled out Andromeda, a rebuild of how ads are matched to people. By July 2025 it was on for everyone. There was no splash page in Ads Manager and little training for reps.
Delivery has three stages: retrieval (billions of ads down to a few thousand candidates), ranking, then auction. For years retrieval was the bottleneck. The old system took shortcuts and leaned on your targeting because it could not search a large enough pool in time. Advertisers compensated with tight interest stacks, micro-segments, and generic hooks — “busy professional trying to lose weight” — because Facebook needed you to narrow the field.
That era is over. Andromeda is a faster retrieval engine on new hardware. Throughput is much higher. The candidate pool is much larger. The ad's content — message, visual, format, angle — now does most of the targeting. Your interest stack does less, and often hurts.
Why the old health playbook stalled
Tight targeting shrinks the pool Andromeda wants to search. One or two creatives gives retrieval nothing to route. Near-duplicates — same video, new first three seconds, swapped headline — are not diversification. Meta wants meaningfully different concepts: different pains (coverage, side effects, “do I qualify,” clinician access), different hooks, different formats (9:16, 1:1, static, carousel), different offers.
You are no longer hunting one winning ad. You are building a set of ads that can match different people inside the same certified audience: someone ready for a consult, someone still comparing compounds, someone who needs the Page to look like a real clinic.
The structure that fits
The old shape was many ad sets, few ads each. The post-Andromeda shape is one campaign per goal, one broad ad set (Advantage+ audience, minimal hard exclusions), and a large set of genuinely different ads inside it — tens, not two. Age floors still make sense in health. Value rules can down-weight weak segments without deleting them from exploration. Interest targeting and lookalikes as the primary split usually do not.
Inside that set, mix funnel stages: education and “why now,” direct problem-solution and proof, then objection and offer. Andromeda coordinates sequence if you give it the material. Batch launches monthly or twice a month so you are not resetting learning every other day. Kill ads that get no spend after a week. Keep the ones that spend and hit cost per qualified consult or purchase.
Format still matters. The same script should exist as short vertical, square UGC, a static, and a carousel where it earns it. Different patients consume the feed differently. Same claim, different vehicle — and the claim still has to be one your certification allows.
Optimise for a qualified event
Broad retrieval plus a junk lead event is how you buy volume that never becomes a patient. Add friction on the form. Send qualified actions server-side. Optimise for booked calls, starts, or purchases — not raw submits. Clean the pixel. Andromeda is only as good as the outcome you tell it is good.
What died: the six-ad habit, tight interests as strategy, splitting ad sets by funnel stage, and the hunt for a single unicorn creative.
What did not die: account health. Andromeda also makes it easier for Meta to deprioritise advertisers it does not trust. Hidden restrictions on a cold health account will cap a beautiful creative portfolio the same way they capped the old structure. Diversification is the retrieval requirement. Platinum-grade infrastructure is still what keeps a certified brand in the pool when the classifier gets nervous.
