Meta Andromeda Broke Your Winning Campaign Structure — How to Rebuild ROAS Without Starting From Zero
The short answer: Meta's Andromeda update rewired how the auction selects and serves ads at the retrieval layer — making legacy campaign architectures like 3:2:2 and tightly segmented interest stacks actively fight against the algorithm instead of working with it. You do not need to start from zero, but you do need to restructure around signal quality, creative velocity, and off-platform attribution before you touch your budget.
What Meta Andromeda Actually Did (And Why Your Old Structure Stopped Working)
Andromeda is Meta's overhaul of its ad retrieval and ranking infrastructure — the system that decides which ads enter the auction in the first place. Before Andromeda, campaign structure gave the algorithm meaningful directional signals. Tight ad set segmentation, stacked interest layers, and controlled audience splits helped the system find buyers efficiently.
After Andromeda, the retrieval layer became dramatically more autonomous. Meta's system now ingests a far wider candidate pool before ranking, which means your carefully constructed audience fences are no longer guardrails — they're friction. The algorithm routes around them anyway, and in doing so, it penalizes accounts that force it to work against its own retrieval logic.
The result is what thousands of media buyers have reported over the past several months: campaigns that ran profitably for a year going cold in a week, CPAs jumping 30–50% with zero changes on the advertiser's side, and the deeply unsettling feeling that the platform is no longer responding to optimization inputs the way it used to.
This is not a testing problem. It is not a creative fatigue problem. It is a structural mismatch between how your account was built and how Meta's system now operates.
The Specific Mechanics That Broke Your Framework
Interest stacking fragmented signal. Andromeda's retrieval layer learns from aggregated behavioral signals across a broad candidate set. Splitting audiences into narrow interest segments starves each ad set of the conversion volume it needs to exit the learning phase and stabilize — so you end up with multiple ad sets in perpetual learning, burning budget without generating usable data.
3:2:2 multiplied creative surface area without multiplying signal. Running three campaigns, two ad sets, and two creatives per ad set made sense when structure guided delivery. Now it splits your conversion signal across too many containers, and the algorithm cannot identify a winner fast enough before budget is exhausted.
Audience exclusions created delivery gaps. Exclusion lists that previously prevented overlap now create holes in delivery that Andromeda's retrieval system doesn't know how to fill efficiently, leading to erratic spend pacing and inflated CPMs.
Campaign budget optimization lost its predictability. CBO's internal allocation logic changed alongside Andromeda. Budgets that previously distributed sensibly across ad sets now concentrate unpredictably, often starving your best-performing creative while overserving underperformers during the new learning cycle.
Before You Rebuild: Diagnose What Actually Broke
The worst mistake you can make right now is launching a restructure before you understand your specific failure mode. Andromeda did not break every account the same way. Spend 48 hours in diagnosis before you touch a single campaign setting.
Step 1 — Pull Your Pre/Post Andromeda Performance Baseline
Andromeda's rollout accelerated through late 2024 and into 2025. Pull a 90-day performance window and segment it into pre-impact and post-impact periods. Look specifically at:
CPM trend (rising CPM with flat CTR = auction-level problem, not creative problem)
Frequency vs. reach ratio (if frequency is climbing but reach is flat, your audience fences are trapping delivery)
Learning phase percentage across active ad sets (more than 30% of spend in learning = structural fragmentation)
Cost per purchase or cost per lead by ad set, not campaign average (averages hide which containers are actually broken)
Step 2 — Separate Platform Attribution From Real Attribution
This is non-negotiable. Meta's self-reported numbers became less reliable as Andromeda changed how conversion signals are weighted in the ranking model. If you are optimizing against numbers that the platform itself is misreporting, every decision you make compounds the error.
Before rebuilding your campaign structure, install a parallel attribution layer:
Post-purchase survey (Fairing, Northbeam, or even a simple Typeform) asking customers how they found you — this is your ground truth for channel attribution
MER (Marketing Efficiency Ratio) calculated as total revenue divided by total ad spend across all channels — a platform-agnostic north star metric that Andromeda cannot manipulate
Incrementality test using Meta's own Conversion Lift tool or a geo-holdout — run this before restructuring so you have a clean baseline to measure against post-rebuild
The goal is to stop trusting the dashboard as your primary optimization signal and start trusting your own data stack. This is the single most important shift you can make in a post-Andromeda environment.
Step 3 — Audit Your Conversion Signal Quality
Andromeda's retrieval model is heavily dependent on the quality of conversion signals flowing back to Meta via the Conversions API. If your CAPI setup is incomplete, duplicated, or firing on low-intent events, you are training the algorithm on bad data — and it will find the wrong buyers at the wrong price.
Check your Events Manager for:
Event match quality score below 7.0 (anything lower means Meta cannot reliably match your conversions to real users)
Duplicate events (browser pixel + CAPI firing the same event without deduplication)
Optimizing for a proxy event (Add to Cart, Initiate Checkout) when you have sufficient Purchase volume — Andromeda rewards accounts that optimize for the event closest to revenue
The Rebuild Framework: Structure for How Meta Works Now
Once you have your diagnosis, you are ready to rebuild. The core principle is consolidation — fewer containers, stronger signal, more creative surface area. Here is the framework we use at Ise AI for accounts coming out of an Andromeda disruption.
Campaign Architecture: Consolidate to One or Two Campaigns
Stop running three to five campaigns simultaneously unless you have the budget to feed all of them adequate conversion volume. The threshold Meta needs to exit learning and stabilize delivery is approximately 50 conversion events per ad set per week. If your budget cannot support that across multiple campaigns, you are guaranteeing perpetual learning.
For most accounts under $30K/month:
One Prospecting campaign (broad or Advantage+ audience)
One Retargeting campaign (30-day website visitors, existing customers excluded from prospecting)
For accounts over $30K/month with strong conversion volume:
One Advantage+ Shopping Campaign (ASC) for proven evergreen creative
One manual campaign for new creative testing with controlled budget
One retargeting campaign with tight audience and aggressive offer
Audience Strategy: Broad Is the New Precise
This is the hardest pill for experienced media buyers to swallow, but it is what the data supports post-Andromeda: broad targeting with strong creative outperforms interest-stacked audiences in most verticals right now.
Andromeda's retrieval model is better at finding your buyer from a broad signal than you are at defining them through interest layers. The algorithm has access to behavioral data you do not — purchase history, app usage, content engagement — that makes your interest targeting a subset of what it already knows.
What this means practically:
Remove interest stacks from prospecting ad sets and test broad (age/gender/geo only)
Let Advantage+ Audience run with your pixel data as the primary signal source
Use creative to do the audience qualification work that interest targeting used to do — your ad should speak so specifically to your buyer that it self-selects
Creative Strategy: Volume, Velocity, and Variance
In the post-Andromeda environment, creative is the primary lever you still control. The algorithm decides who sees your ad — you decide what they see. This is where your competitive advantage now lives.
Creative volume: Run a minimum of four to six creatives per ad set. Andromeda's system needs creative variance to identify winners. Two creatives is not enough surface area for the algorithm to learn from.
Creative velocity: Introduce new creative every 10–14 days, not when you see fatigue. By the time frequency metrics show fatigue, you have already lost momentum. Build a creative pipeline that is always 2–3 concepts ahead of your current rotation.
Creative variance: Test across format (static, video, UGC, carousel), hook (problem-led, social proof, offer-led, curiosity), and angle (different pain points, different buyer personas). The goal is to give the algorithm enough signal to find your highest-converting combination — then scale that combination before it fatigues.
Budget Controls: Build Your Own Guardrails
One of the most damaging consequences of Andromeda is that budget pacing became less predictable. Campaigns that previously spent smoothly now spike and stall, and Meta's native budget controls are not granular enough to prevent costly overruns — especially for agencies managing multiple client accounts.
Until Meta provides better native controls, build your own:
Set campaign-level spending limits in addition to daily budgets
Use automated rules to pause campaigns if CPA exceeds your threshold by 20% for more than 24 hours
Implement a third-party budget monitoring tool (Revealbot, Madgicx, or custom API alerts) that catches anomalies before they become client crises
Never rely on a Meta account rep's recommendation to increase budget during a performance dip — this is the most common and most costly mistake we see in disrupted accounts
The Advantage+ Question: Use It, But Use It Carefully
Advantage+ campaigns are not the enemy. They are a tool that works well in specific conditions and poorly in others. The frustration most advertisers feel toward Advantage+ comes from being pushed into it without understanding the tradeoffs.
Advantage+ works well when:
You have a proven creative asset with demonstrated conversion history
Your pixel has strong purchase signal (1,000+ purchase events in the past 30 days)
You are scaling a known winner, not testing new angles
Advantage+ works poorly when:
You are in a niche vertical where Meta's broad targeting finds the wrong buyer
Your creative is not yet proven and you need controlled testing conditions
You need to exclude specific audiences (existing customers, competitors) with precision
The practical answer: run Advantage+ alongside a manual campaign, not instead of one. Use ASC to scale proven creative and use manual campaigns to test new hypotheses. This gives you the algorithm's efficiency where it works and your own judgment where it doesn't.
What to Expect During the Rebuild Period
Restructuring a disrupted account is not a 48-hour fix. Here is an honest timeline:
Days 1–3: Diagnosis, attribution setup, conversion signal audit. Do not touch live campaigns until this is complete.
Days 4–7: Consolidate campaign structure. Pause fragmented ad sets. Launch new consolidated structure with broad audience and minimum four creatives.
Days 7–21: Learning phase. Expect CPAs to be elevated and delivery to be uneven. Do not optimize aggressively during this window — every edit resets the learning phase and costs you data.
Days 21–35: First real optimization window. Use your off-platform attribution data (MER, post-purchase survey) to evaluate performance, not Meta's dashboard alone. Kill underperforming creatives, introduce new variants, and begin scaling the consolidated structure that is working.
The accounts that recover fastest from Andromeda disruption are the ones that resist the urge to make constant changes during the learning window. Patience during days 7–21 is genuinely the hardest and most important part of the rebuild.
The Control You Still Have (And How to Use It)
The narrative that Meta has taken all advertiser control away is not quite accurate — but the control that remains is different from what it used to be. Understanding where you still have meaningful leverage is what separates accounts that recover from ones that keep chasing the algorithm.
You control creative. This is your highest-leverage input. The algorithm decides delivery; you decide message. Invest here first.
You control your offer. A structurally sound campaign with a weak offer will not recover. If your CPA is broken, audit your landing page conversion rate and your offer mechanics before assuming the problem is in the ad account.
You control your signal quality. CAPI setup, event match quality, and conversion event selection are all within your control and directly influence how well Andromeda's retrieval model finds your buyer.
You control your attribution framework. Choosing to optimize against MER and post-purchase survey data instead of Meta's dashboard numbers is a decision that puts you back in the driver's seat — even when the platform's own reporting is unreliable.
You control your guardrails. Automated rules, spending limits, and third-party monitoring tools give you meaningful oversight within an automated system — without requiring you to operate in a fully black-box environment.
When to Call the Rebuild Complete
Your rebuild is working when:
MER has returned to within 15% of your pre-Andromeda baseline
Less than 20% of active ad set spend is in the learning phase
Post-purchase survey data shows Meta as a consistent top-two acquisition channel
You have a creative pipeline with at least three new concepts ready to deploy
Your off-platform attribution and Meta's reported numbers are directionally aligned (they will never match exactly)
At that point, you have not just recovered — you have built a more resilient account structure than you had before Andromeda. One that is less dependent on any single structural framework and more dependent on signal quality, creative strength, and platform-agnostic measurement. That is an account that can survive the next algorithm change, whatever it turns out to be.
Frequently Asked Questions
What is Meta Andromeda and when did it roll out?
Andromeda is Meta's overhaul of the ad retrieval and ranking infrastructure that determines which ads enter the auction. It is not a single update with a public launch date — Meta rolled it out progressively through 2024 and into 2025. Most advertisers began reporting significant performance disruption in Q3–Q4 2024, with the impact becoming widespread by early 2025.
Why did my Facebook ads stop working after Andromeda?
Andromeda changed how Meta's system retrieves candidate ads before ranking them. Campaign structures built around tight audience segmentation, interest stacking, and multiple parallel ad sets now fragment your conversion signal and create friction for the algorithm's retrieval logic — causing elevated CPAs, erratic delivery, and campaigns that appear stuck in learning. The structure that worked before now actively works against the system.
Should I delete my old campaigns and start fresh?
In most cases, no. Your existing campaigns carry conversion history and pixel signal that has value. The better approach is to consolidate — pause fragmented ad sets, reduce campaign count, and rebuild within existing campaign containers where possible. Starting from zero means starting a new learning phase with no historical signal, which typically makes performance worse before it gets better.
Is Advantage+ Shopping Campaign (ASC) worth using after Andromeda?
Yes, in specific conditions. ASC works well when you have proven creative assets and strong pixel purchase history (1,000+ purchase events in 30 days). It works poorly for creative testing, niche verticals, or accounts with thin conversion data. The best approach is running ASC alongside a manual testing campaign rather than replacing manual campaigns entirely.
How long does it take to recover ROAS after restructuring?
Expect a 21–35 day recovery window. The first 7–21 days after restructuring are a learning phase where CPAs will be elevated and delivery uneven. Resist making frequent changes during this window — every edit resets the learning phase. Meaningful optimization should begin around day 21, with full recovery typically visible by day 35.
Should I trust Meta's reported ROAS numbers after Andromeda?
Use them as a directional signal, not a source of truth. Meta's attribution changed alongside Andromeda, and self-reported platform numbers are not reliable enough to be your primary optimization metric. Build a parallel attribution framework using post-purchase surveys, Marketing Efficiency Ratio (total revenue ÷ total ad spend), and periodic incrementality tests to get an accurate read on true performance.
Does broad targeting actually work better than interest targeting now?
For most accounts, yes — in prospecting. Andromeda's retrieval model has access to behavioral signals that make your interest-layer definitions a subset of what it already knows. Broad targeting with strong, specific creative outperforms interest stacks in the majority of verticals post-Andromeda. The creative now does the qualification work that interest targeting used to do.
How do I prevent budget overruns during the rebuild period?
Set campaign-level spending limits in addition to daily budgets, create automated rules that pause campaigns when CPA exceeds your threshold for more than 24 hours, and use a third-party monitoring tool for real-time alerts. Do not rely on Meta's native controls alone — they are not granular enough to prevent costly overruns, especially across multi-account agency portfolios.
What is the minimum budget needed to exit the learning phase after restructuring?
Meta's system requires approximately 50 conversion events per ad set per week to exit learning and stabilize delivery. Work backward from your current CPA to determine the minimum daily budget per ad set. If your budget cannot support 50 events per week across multiple ad sets, consolidate further until each active ad set has enough budget to hit that threshold.
Can Ise AI help rebuild our account after Andromeda?
Yes. Andromeda recovery is one of the most common briefs we work on right now. We start with a full structural audit, attribution layer setup, and conversion signal review before touching a single campaign — because rebuilding on a broken measurement foundation just recreates the same problem. If your account has been disrupted and you want a second opinion on what actually broke, reach out to the Ise AI team.