If your Meta ROAS fell off a cliff in the first half of 2026 and nothing you changed in Ads Manager brought it back, you're not doing anything wrong. Meta changed the game underneath you, and the fix isn't in the bidding tab.
To recover ROAS after the Andromeda update, stop editing bids and fix the two signals the new engine actually reads — your creative and your conversion data. Consolidate into broad Advantage+ campaigns, ship 15–20+ distinct creatives a month, and feed qualified-lead events back through the Conversions API so Meta optimizes toward real buyers. Agencies that do all three recover the reported drop in two to three weeks, because most of that drop was measurement and targeting signal, not lost sales.
Here's exactly what changed, why your old playbook stopped working, and the recovery sequence we run for clients.
What Andromeda Actually Changed
Andromeda is Meta's new ads retrieval engine. It finished rolling out to all advertisers by early 2026, and it fundamentally changed how ads get matched to people.
The old model routed your ad using the audience you built — interests, demographics, lookalikes. Andromeda mostly ignores that. It reads the creative itself — the visuals, the copy, the on-screen text, even the audio — and predicts who the ad is for from behavioral signals tied to that content. Your narrow interest stack isn't wrong anymore; it's just irrelevant. The engine decided it can target better from your creative than from your audience settings.
That single shift breaks a decade of agency muscle memory. Interest research, lookalike laddering, hyper-segmented ad sets — the levers most media buyers reach for first are the ones Andromeda pays the least attention to.
Why Your ROAS Dropped 15–30% Overnight
Two things stacked on top of each other, and it's worth separating them because they need different fixes.
First, Andromeda punished narrow setups. If your account was built on tight interest stacks and small lookalikes, performance genuinely slipped as the engine started favoring broad targeting plus creative signal.
Second — and this is the one that panicked everyone — a March 3, 2026 click-attribution change moved engagement actions like likes, shares, and saves out of click-through reporting. Reported ROAS dropped 15–30% overnight across accounts that had no change in actual sales. The money was still coming in. The number in Ads Manager just stopped counting the same events it used to.
So before you rebuild anything, do one boring thing: compare your reported ROAS to actual revenue in your CRM or Hyros for the same window. We do this first for every client that comes to us in a panic. Nine times out of ten, half the "drop" is a reporting artifact, and knowing that stops you from slashing budgets on campaigns that are still profitable. If you've never trusted your in-platform numbers, this is the moment to get server-side attribution right so you're managing to reality, not to a deflated dashboard.
The Recovery Sequence That Actually Works
Once you know how much of the drop is real, the recovery is the same three moves in the same order. Do them together — none of them works alone.
1. Consolidate into broad Advantage+ campaigns
Collapse your narrow ad sets. Andromeda rewards broad targeting because that gives it the widest pool to match your creative against. In practice, one broad ad set holding 20–25 diverse creatives beats five narrowly targeted ad sets — Meta's own data shows that structure delivering around 17% more conversions at 16% lower cost.
Advantage+ Shopping campaigns also got priority in the auction in Meta's January 2026 updates, with 15–22% lower CPA than manual setups because they tap real-time auction multipliers manual bids can't touch. If you're still running the old multi-ad-set, interest-stacked structure, that alone is costing you.
One caveat we tell every client: consolidation feels wrong at first. Watching six ad sets collapse into one broad campaign sets off every "but I'm losing control" instinct a media buyer has. You are losing control — of levers that stopped working. What you gain is a campaign the engine can actually learn from, because a single ad set with real budget exits the learning phase fast instead of a dozen tiny sets that never gather enough conversions to optimize. Give it seven days before you judge it, and don't edit it mid-learning — every meaningful edit resets the phase and buys you nothing but noise.
2. Ship creative velocity — for real
This is where most agencies stall, because it's actual work. Andromeda matches each creative to the audience its content implies, so the more conceptually distinct creatives you feed it, the more demand it can find. The number that matters: brands testing 20+ new ads a month are seeing roughly 65% higher ROAS than those shipping fewer than 10.
"Conceptually distinct" is the key phrase — not 20 color variations of the same ad. Different hooks, different personas, different awareness states (problem-aware, solution-aware, product-aware), different formats. You're not A/B testing a button color; you're giving the engine range. This is exactly why ad creative is now 70% of performance and why the words in your ads carry more weight than ever — Andromeda literally reads them.
Here's the volume-to-performance pattern we see hold up across accounts:
| New creatives / month | What Andromeda does with it | Typical outcome | |---|---|---| | Under 10 | Runs out of range fast; recycles the same audiences | Flat or declining ROAS | | 10–20 | Enough range to find pockets of demand | Stabilizes | | 20+ | Wide matching surface, distinct emotional triggers | ~65% higher ROAS vs. under-10 |
If your team can't hand-make 20+ good concepts a month, this is where AI-assisted production earns its keep — not to spray junk, but to turn one strong idea into ten honest variations a human still approves before it ships.
3. Fix the conversion signal so it stops learning from junk
This is the move almost nobody talks about, and it's the one that separates a recovery from a rebuild.
A default pixel setup fires a conversion for every form fill — bots, tire-kickers, misclicks, all of it. Andromeda and Advantage+ then dutifully optimize toward whoever fills the form cheapest, and you end up with a pile of garbage leads at a great cost-per-lead and a terrible cost-per-closed-deal. The platform did exactly what you told it to. You just told it the wrong thing.
The fix is server-side. Send Conversions API events only when a lead is actually qualified or accepted, not on the raw form fill. Now Meta learns from real buyers. Advertisers running a proper CAPI setup see about 17–18% lower cost per result, and lead quality starts visibly improving in weeks two to three as the algorithm retrains. To do this well you need your CRM talking to Meta — the same lead-routing and follow-up automation that qualifies the lead should be what fires the conversion event back. That's the loop that quietly fixes lead quality while everyone else is still blaming the creative.
What to Stop Doing
A short list, because the "don'ts" matter as much as the "dos":
- Stop editing bids to chase the number. Manual bid tweaks fight an engine that's optimizing on signals bids can't influence.
- Stop rebuilding narrow audiences. Interest stacks and micro-lookalikes are the setup Andromeda deprioritizes.
- Stop judging on in-platform ROAS alone. After the March 2026 attribution change, that number under-reports. Manage to cost-per-qualified-lead or real revenue.
- Stop making 20 versions of one ad and calling it velocity. The engine needs conceptual range, not cosmetic variety.
If you're weighing where the budget should even live after all this, we broke down Google Ads vs. Meta Ads for Dallas businesses — because sometimes the honest answer to "how do I fix Meta" is "shift 20% of it somewhere the signal is cleaner."
The Bottom Line
Andromeda didn't break Meta ads. It moved the controls. The agencies still panicking are the ones yanking on levers — bids, interests, audiences — that the new engine ignores. The ones recovering are doing three unglamorous things: consolidating into broad Advantage+ campaigns, shipping real creative volume, and feeding the algorithm clean, qualified conversion data so it optimizes toward buyers.
Do all three and the reported ROAS drop mostly reverses within a couple of learning weeks — because a good chunk of it was never a real drop, just a signal you weren't sending and a report that changed how it counts.
Want us to run the diagnosis for you? We'll pull apart how much of your ROAS drop is real versus reporting, audit your creative velocity and Advantage+ structure, and wire your Conversions API to fire on qualified leads instead of junk. Get a free automation audit and we'll show you exactly where the signal is leaking.