Key Takeaways
- Creative is now the targeting — Andromeda reads your ad to decide who sees it. Audience selection matters far less than it used to.
- Lookalike audiences are effectively dead — We stopped using them across our portfolio. Broad targeting with strong creative outperforms every time.
- Advantage+ Shopping (ASC) should be your primary campaign type — It delivers 17% lower CPA than manual campaigns and handles most of the heavy lifting.
- Creative volume is the new media buy — Brands testing 20+ new ads per month are seeing 65% higher ROAS than those testing fewer than 10.
- Signal quality is non-negotiable — Clean Pixel and CAPI data now directly drive Andromeda's learning speed.
In This Article
Meta's Andromeda update, fully deployed across most objectives and placements by late 2025 into 2026, replaced the audience-selection model advertisers relied on for a decade. Instead of picking interests, demographics, and lookalike seeds, the algorithm now reads your ad creative to decide who sees it.
That's the shift: your creative is your targeting now. And if your campaign structure, creative library, and data infrastructure haven't caught up to that reality, you're paying more and getting less.
We've been rebuilding Meta accounts under this new system since the rollout began. Here's what actually changed, what still works, and the playbook we're running across our portfolio.
What Andromeda Actually Is
Andromeda is Meta's AI-driven ads retrieval system. It operates at the stage before bidding: when Meta evaluates millions of potential ads to show a user, Andromeda is what narrows the field to viable candidates.
The old system started with your audience definition and worked forward. Andromeda works in reverse. It evaluates your creative first, predicts which users are most likely to engage based on behavioral signals, and serves accordingly. Advertiser-defined audiences are now suggestions, not gates.
According to Meta, Andromeda is 4x more efficient at driving ad performance gains compared to the previous recommendation ranking models. The advertiser-side implication: the algorithm is better than you are at finding your buyer. What you control is the creative it reads to make that determination.
Key Insight
Andromeda works alongside GEM (Global Earnings Model), which handles ranking and pricing after retrieval. Understanding both helps explain why two ads with identical targeting can produce wildly different CPMs: Andromeda determines relevance eligibility, GEM determines cost.
Before vs. After Andromeda
The End of Lookalike Audiences
This is the change that surprised the most advertisers, and it's worth being direct about it: lookalike audiences no longer drive the performance they once did.
We stopped using them as a primary targeting strategy across our Meta accounts. Not because they stopped working overnight, but because the data made a clear case. Broad targeting with differentiated creative was consistently outperforming our 1% lookalikes, especially on accounts doing meaningful volume.
The reason ties back to Andromeda. The system has access to behavioral signals that dwarf anything a lookalike seed audience can capture. When you feed it a 1% LAL, you're constraining a machine that's already better at finding your buyer than your seed data can define. You're not adding precision. You're adding friction.
Worth Noting
Lookalikes aren't completely useless at every budget level. For accounts under $5K/month with limited conversion history, a 2-3% LAL can still help anchor early delivery. But as soon as the algorithm has enough signal, broad targeting takes over. Plan for that transition rather than building a strategy that depends on LAL forever.
The brands that are struggling most right now are the ones still building campaigns the old way: multiple ad sets with 1%, 2%, and 5% lookalikes, stacked interest layers, and age/gender exclusions. All of that structure was designed to control a system that no longer needs that kind of guidance. It's slowing the algorithm down.
Creative as Targeting: The New Performance Driver
Meta's data science team has stated that creative quality now accounts for 56% of all campaign performance outcomes. More than targeting, budget, placement, and timing combined. That's a meaningful number, and it changes where you should be spending your time.
What Andromeda reads in your creative: visual format, messaging tone, subject matter, hook structure, product category signals. It uses all of this to predict audience match. If your ad looks like it's for a 35-year-old woman who runs and buys wellness supplements, it'll find her. You don't have to tell it to.
Pro Tip
Think of creative diversity as the new audience segmentation. Instead of building ad sets for different audiences, build different creative angles for different buyer motivations. Price-sensitive buyers respond to a different hook than aspirational buyers. Let the algorithm route them.
What "Creative Diversity" Actually Means
Creative diversity doesn't mean five versions of the same ad with different background colors. It means fundamentally different narratives, hooks, and formats. The minimum bar is 6 meaningfully different ads per ad set. The benchmark for accounts seeing strong results is 15-20 active creatives at any time.
A practical framework that's working across our portfolio:
New Concept (50% of creative budget)
Fundamentally different narratives: a social proof angle, a problem-first angle, a founder story. Each should be able to stand alone as a campaign.
Refined Variants (30% of creative budget)
Take your proven performers and improve them. Tighter hook, updated social proof, new format (static to video). These anchor your performance floor while you test.
Proven Performers (20% of creative budget)
Keep running what works until the data tells you to stop. Rising CPM without scale gains is the signal to refresh, not a fixed calendar date.
One thing we've seen consistently: founder-led content formats (authentic talking-head videos, behind-the-scenes, direct-to-camera testimonials from the founder) tend to outperform polished brand creative under Andromeda. We're regularly seeing 2-3x ROAS on founder content compared to standard brand video. The algorithm seems to weight authenticity signals strongly.
Creative Format Performance Ranking (2026)
How to Structure Meta Campaigns in 2026
The most common structural mistake we see: over-segmentation. Multiple campaigns, multiple ad sets, narrow audience definitions, age/gender splits. That structure was logical when the advertiser controlled targeting. Under Andromeda, it fragments the signal the algorithm needs to optimize.
Simpler structures outperform complex ones, not because the algorithm is lazy, but because budget fragmentation slows learning. Each ad set needs conversion data to optimize. Spread the same budget across 12 ad sets and none of them get enough data to exit the learning phase.
The 2026 Campaign Structure That Works
ASC now represents 62% of ecommerce Meta spend, and for good reason: it delivers 17% lower CPA than manual campaigns and outperforms manual structures by 15-25% on ROAS. For most DTC brands, it should be the core of the account, not an experiment running alongside your manual campaigns.
What Works at Different Budget Levels
ASC requires a minimum daily budget of about $100 to gather enough conversion data. Below that, the algorithm can't learn fast enough to be useful. If you're under $100/day, run one broad campaign with 6+ creatives and focus all your energy on creative quality.
Once you're at $300+/day, ASC is the move. At $1K+/day, run ASC as the primary and use a separate broad campaign as a dedicated creative testing sandbox before promoting winners into ASC.
Signal Quality and Conversion Data
Andromeda validates its predictions through outcome data. Every time someone it routed to your ad converts (or doesn't), that result feeds back into its model. Clean, accurate conversion signals directly affect how fast the algorithm learns and how well it optimizes.
This matters more in 2026 than it ever has. Bad signal quality isn't just an attribution problem. It actively slows Andromeda's learning loop and inflates your CPMs.
Signal Quality Checklist
Pixel + CAPI Running Simultaneously
Pixel-only is no longer sufficient. Conversions API (CAPI) sends server-side events that survive browser privacy restrictions. Run both with deduplication enabled.
One Prioritized Conversion Event
Pick one event (Purchase, Lead, Initiate Checkout) that maps to your business outcome and optimize for it consistently. Switching optimization events between campaigns confuses the algorithm.
Attribution Window Aligned to Purchase Cycle
7-day click is standard. For impulse-purchase products, 1-day click can reduce the attribution inflation that inflates your reported ROAS. Know your actual purchase cycle and choose accordingly.
Event Match Quality Score Above 7
Find this in Events Manager. Below 7 means Meta is having trouble matching your conversion events to real users, which degrades Andromeda's learning. Common fix: pass email and phone via CAPI with proper hashing.
Common Signal Quality Mistake
We see accounts where CAPI was set up once and never audited. Shopify app integrations in particular can have deduplication issues that cause event double-counting. This inflates your reported conversion numbers and confuses Andromeda's feedback loop. Audit your Events Manager monthly, not quarterly.
2026 Meta Ads Benchmarks
Here's where the numbers stand for ecommerce advertisers running Meta in 2026. Use these as context for your own performance, not as hard targets. Your category, average order value, and brand maturity all move these significantly.
CPMs are up 20% year over year. That's not a surprise given the ad load and the broader advertiser competition on Meta. What it means practically: your creative efficiency has to improve to maintain margin. The brands absorbing CPM increases without it are the ones seeing ROAS compression.
January is still the best month to scale. CPMs run 22% below average in January. November runs 41% above average during BFCM competition. If you're building your annual Media plan, weight acquisition spend toward Q1 and protect Q4 margin by running retention and loyalty campaigns rather than cold acquisition at inflated CPMs.
Instagram Feed continues to lead on ROAS for ecommerce (4.1x average), followed by Facebook Feed (3.8x). Reels placements are improving but remain inconsistent depending on the creative format. For most DTC brands, we let ASC handle placement optimization rather than manually favoring specific placements.
The Rebuild Playbook: How We Adapt Accounts
When we take over a Meta account that was built for the pre-Andromeda world, here's the pattern we almost always find: too many ad sets, too many campaigns, lookalike audiences doing the work that creative should be doing, and a CAPI setup that was installed and forgotten.
The rebuild isn't a one-day project. It takes about 4-6 weeks to run it cleanly without destabilizing active delivery. Here's the sequence:
Audit Signal Quality First (Week 1)
Before touching campaign structure, fix the foundation. Audit Events Manager for duplicate events, check CAPI redundancy, and verify your Event Match Quality score. A structural rebuild on top of bad signal data just amplifies the problem.
Consolidate Campaigns (Week 1-2)
Collapse multiple LAL ad sets into broad targeting. Combine campaigns targeting the same audience with different creatives. The goal: fewer campaigns with more budget per campaign and more creatives per ad set.
Launch Creative Testing Campaign (Week 2)
Separate broad-targeting campaign at 20-25% of budget. Minimum 6 new creative concepts per test cycle. Anything hitting your target CPA in the first 7-10 days gets promoted to ASC. Anything not meeting threshold gets paused by day 14.
Migrate to ASC as Primary (Week 3-4)
Launch Advantage+ Shopping as your primary campaign with your top 6-10 proven creatives. Start at 50% of budget and scale up as the learning phase completes (typically 7-14 days). Keep your retargeting campaign running but at reduced budget.
Build the Creative Pipeline (Ongoing)
This is the part most brands underinvest in. You need a repeatable process for producing new creative, not a one-time sprint. Brands testing 20+ new ads monthly see 65% higher ROAS than those testing fewer than 10. Build the pipeline, not just the campaign.
When I used to buy media, I would be calling ads within 48 to 72 hours and rotating new ones. The accounts that kept winning had a constant supply of fresh creative, not just a library of winners they kept rerunning. That principle matters even more now.
— Edwin Choi, Jetfuel Agency
Frequently Asked Questions
What is Meta's Andromeda update and how does it affect my ads?
Andromeda is Meta's AI-driven ad retrieval system that determines which ads are eligible to show to which users. Instead of relying on advertiser-defined audiences (interests, demographics, lookalikes), it reads your ad creative to predict user match. The practical effect: your creative now functions as your targeting. Broad audience settings combined with high-quality, diverse creative outperform narrow audience targeting in most account types.
Are lookalike audiences still worth using in 2026?
For most accounts, no. Lookalike audiences are no longer a primary strategy under Andromeda. The algorithm's behavioral signals already exceed what a seed audience can define, so adding a LAL constraint limits delivery without improving targeting quality. The exception: accounts under $5K/month with limited conversion history may still see value in a 2-3% LAL as early-stage scaffolding. But it should be treated as a temporary tool, not a long-term strategy.
Should I switch to Advantage+ Shopping Campaigns?
Yes, for most ecommerce brands at meaningful budget scale ($100+/day). ASC delivers 17% lower CPA than manual campaigns on average and handles placement optimization, creative delivery, and budget allocation automatically. The main requirement: you need at least 6 strong creatives to give the algorithm enough material to work with. Start ASC at 50-60% of your budget and keep a separate broad campaign for creative testing.
How many creatives do I actually need running at once?
The minimum is 6 meaningfully different creatives per ad set. For accounts doing $500+/day, aim for 15-20 active ads at any time, with 6+ new concepts tested monthly. Brands testing 20+ new ads per month see 65% higher ROAS than those testing fewer than 10. The key is "meaningfully different" -- different hooks, formats, and narratives, not just the same ad with different thumbnail crops.
What is a good ROAS for Meta ads in 2026?
The median ROAS across all ecommerce industries on Meta is 1.93x, with an average of 2.98x according to Triple Whale data. However, ROAS targets should be set based on your blended margin and customer acquisition economics, not industry averages. A brand with 60% gross margin can be profitable at 2x ROAS; a brand with 30% gross margin needs 4x or more. Instagram Feed delivers the strongest placement ROAS at 4.1x on average.
Where to Go From Here
The brands that are struggling with Meta right now are the ones running the same playbook they had in 2023. Multiple LAL ad sets, interest stacks, polished brand videos as the primary creative. That approach worked when the advertiser controlled targeting. It doesn't work when the algorithm does.
The shift is real but the path forward isn't complicated: simplify your campaign structure, invest heavily in creative diversity, fix your signal quality, and let ASC do what it's built to do. The accounts we've rebuilt under Andromeda are performing better than the old structures did, once the learning phase clears.
The hard part isn't the strategy. It's building the creative pipeline that sustains it.
Want an expert review of your Meta account structure?
We audit Meta accounts and identify exactly what needs to change for Andromeda. No generic audit -- we look at your structure, signal quality, creative library, and ASC readiness and tell you specifically what's holding performance back.
