Meta's Andromeda Update - Everything You Need To Know About Changes In Ad Delivery and What Works Best Now

If your Meta ad costs have been acting weird lately CPCs swinging wildly, cost per lead unpredictable, campaigns that used to crush suddenly struggling - you're not losing your mind. Something fundamental has changed.

In December 2024, Meta quietly rolled out Andromeda, a complete rebuild of how ads get matched to people. By July 2025, it was fully deployed to all advertisers. And here's the kicker: Meta never made a big announcement. No training for account reps. No splash page in Ads Manager. Just silence while advertisers watched their metrics turn from predictable to unpredictable overnight.

Even Meta reps were caught off guard. One advertiser reported their rep saying, "I've been trying to learn about this myself. They haven't given us any training on it."

So what is Andromeda, why does it matter, and how do you adapt before you get left behind? Let's break it down.

The Problem Meta Was Trying to Solve

To understand Andromeda, you need to understand how ad delivery works. There are three stages:

1. Retrieval – Narrow billions of possible ads down to a few thousand candidates

2. Ranking – Score those candidates

3. Auction – Pick the winner

For years, retrieval was the bottleneck. Meta's old system could only process a limited number of ads quickly enough to keep feeds moving. So it took shortcuts, leaned heavily on your targeting settings, and often missed better matches simply because it couldn't sift through enough options fast enough.

Advertisers compensated by building tight interest stacks, splitting audiences into micro-segments, and using generic hooks that cast a wide net. "Hey busy mom trying to lose 20 pounds" worked because Facebook needed you to do the heavy lifting of narrowing down who should see your ad.

That era is over.

What Andromeda Actually Does

Andromeda is Meta's next-generation retrieval engine. It's powered by new machine learning architectures, faster processors (NVIDIA Grace Hopper chips and Meta's own MTIA silicon), and smarter indexing systems.

The result? Three times higher throughput and a vastly larger candidate pool. Meta can now consider far more ads before ranking even begins, and it can do it faster and more precisely than ever before.

What this means in practice: The system got exponentially better at matching the right ad to the right person at the right moment. Your ad's actual content such as the message, visuals, format, and angle now does most of the targeting work. Not your carefully built interest stacks.

Why Your Old Strategy Stopped Working

When retrieval was limited, your targeting settings did the heavy lifting. You'd layer in interests, behaviors, and demographics because the system needed those constraints to function efficiently.

Now? Andromeda is sophisticated enough that tight targeting actually hurts performance. You're shrinking the candidate pool the system wants to explore.

The old "one perfect ad" mindset is dead too. If you only give the system one or two creative concepts to work with, retrieval has nothing to route. It's like giving a chess grandmaster only pawns to play with, you're handicapping the very thing you're trying to leverage.

This is why Meta keeps hammering "creative diversification." It's not marketing speak. It's the actual technical requirement of how the new system operates.

The New Strategy: Creative Diversification

Here's what Meta means by creative diversification and what they don't mean:

Don't do this:

• Take one video and change the first scene (hook)

• Change the CTA (Shop Now → Learn More)

• Swap out image text or background colour

Simply, do not create near-duplicates with minor tweaks.

Do this:

• Create meaningfully different concepts that address different pain points

• Use different hook types (problem-agitation, curiosity, contrarian, social proof)

• Produce multiple formats for each strong concept (9:16 video, 1:1 talking head, static, carousel)

• Test different angles (speed, simplicity, ROI, status, risk avoidance)

• Vary your CTAs and offers on the creative level

The goal is to give Andromeda a portfolio of genuinely different ads so it can map different audience segments to the creative they're most likely to engage with.

Think of it like this: You're no longer trying to find the winning ad. You're building a team of ads that work together, each appealing to different people within your target market.

The New Campaign Structure That Actually Works

This is the biggest shift, and it flies in the face of everything we used to teach.

The Old Way (2019-2022)

• One campaign

• Multiple ad sets (each targeting different audiences)

• 1-3 ads per ad set

• Goal: Find the best audience for each ad

The Old-ish Way (2023-mid 2025)

• One campaign

• Multiple ad sets (each testing different concepts)

• 3-6 ads per ad set

• Campaign Budget Optimization at campaign level

• Goal: Let Facebook distribute budget to winning concepts

The New Way (Post-Andromeda)

• One campaign per goal

• One broad ad set (Advantage+ audience on, minimal constraints)

• 10-50 ads inside that single ad set

• Goal: Give retrieval maximum creative diversity to match people to ads

That's right. One ad set with up to 50 ads running simultaneously.

How to Structure Your One Broad Ad Set

Here's what "broad" actually means:

Targeting:

• Use Advantage+ audience (let Meta find your people)

• Set minimum age if relevant (e.g., 25+)

• Use value rules instead of hard exclusions (decrease bid by 90% for demographics that don't convert, rather than excluding them entirely)

• No interest targeting

• No lookalikes

• No saved audiences

Why this works: Every hard constraint you add shrinks the candidate pool Andromeda wants to explore. Value rules let you de-emphasize low-value segments without eliminating them completely, keeping your exploration space wide.

Inside That One Ad Set: The Portfolio Approach

Your 10-50 ads should cover different funnel stages, all working together:

Top of Funnel (30-40% of ads):

• Personal story / founder's journey

• "Why now" educational content

• Value-driven content with soft CTAs

• Quick demos or transformations

Mid Funnel (40-50% of ads):

• Direct problem-solution messaging

• Clear CTA for lead magnet

• Specific outcome promises

• Product proof and mechanisms

Bottom of Funnel (10-20% of ads):

• Testimonials and case studies

• FAQ-style objection handlers

• Offer-focused content

• Strong urgency and scarcity

The magic is that Andromeda coordinates these. Someone cold might see your personal story first. Someone who's been researching solutions for weeks might see a testimonial. The system figures out the right match but only if you give it the raw material to work with.

The Monthly Creative Drop System

Here's where it gets tactical. You can't just dribble new ads in daily anymore that constantly resets the learning phase at the ad set level.

Instead, move to a monthly (or twice-monthly) drop schedule:

Week 1: Launch 30-50 new ads all at once Week 2-4: Monitor daily and prune ads that get zero spend

Your 50 ads at launch might look like this by week 4:

• Week 1: 50 ads live

• Week 2: 40 ads still getting spend (kill the 10 non-spenders)

• Week 3: 30 ads performing (kill 10 more)

• Week 4: 20-25 ads proven winners

Then for your next drop, create:

• 50-60% creative refreshes of winners (same angle, new edit, different format)

• 30-40% adjacent angles (same product, different pain point or benefit)

• 10-20% completely new big swings

Format Diversity Is Non-Negotiable

For each strong script or concept, produce at least:

• 9:16 short-form video (Reels/Stories)

• 1:1 talking head or UGC-style video (Feed)

• 1200×1500 static image

• 3-5 card carousel

Why? Because different people prefer consuming content differently. The 35-year-old scrolling Instagram during lunch might engage with a carousel. The 28-year-old watching Reels at night wants short-form video. The 45-year-old on Facebook in the morning might click a static image.

Same message, different delivery vehicles. That's creative diversification.

What About Learning Phase?

Yes, adding new ads to an existing ad set triggers learning phase. This is why the monthly drop system matters.

When you add 50 ads once per month, you enter learning phase once and exit. When you add 2-3 ads twice per week, you're constantly bouncing in and out of learning, never stabilizing.

Batch your launches. Give the system time to optimize. Prune the losers. Repeat monthly.

Measurement & When to Kill Ads

Your primary KPI is now at the ad level, not the ad set level.

Kill ads that:

• Never get spend after 5-7 days (Andromeda is telling you they're mismatched)

• Spend but deliver terrible cost per result consistently

Keep and scale ads that:

• Get consistent spend (even if modest)

• Meet or beat your target CPA/ROAS

• Show engagement signals (comments, shares, saves)

Don't be afraid to pause 30-40% of your initial drop. That's the system working, it's telling you which creative resonates and which doesn't.

The Pixel & Optimization Event Matter More Than Ever

Here's the catch with giving Advantage+ this much freedom: You must optimize for qualified outcomes.

If you optimize for "leads" but accept any junk submission, the system will happily deliver volume. It's learning that any lead is a good lead.

Fix this by:

• Adding friction to forms (more qualifying questions)

• Using server-side conversion tracking for qualified actions

• Optimizing for downstream events (calls booked, purchases, not just form submits)

• Cleaning your pixel data regularly

The cleaner your signal about what "quality" looks like, the better Andromeda performs.

What Died With Andromeda

Let's be clear about what best practices you can throw out:

• The 6-ad limit: Gone. Meta now advocates for 10-50 ads per ad set.

• Tight interest targeting: Counterproductive. The ad content does the targeting now.

• Multiple ad sets by funnel stage: Unnecessary. Your ad portfolio covers all stages within one set.

• 1 campaign → many ad sets → few ads each: Doesn't leverage how retrieval now works.

• Finding "the" winning ad: You're building a portfolio, not seeking a unicorn.

The Biggest Mindset Shift

Stop trying to outsmart the algorithm. You won't win.

Meta makes tens of billions of dollars from ads. They have more data on user behavior than any platform in history. Their algorithm has millions of signals you don't have access to.

Your job isn't to constrain the system. Your job is to feed it excellent, diverse creative and let it do what it's built to do: match the right message to the right person at the right time.

The advertisers winning right now aren't the ones micromanaging ad sets and tweaking audiences daily. They're the ones shipping 40 genuinely different ads per month and letting Andromeda conduct the orchestra.

Quick Start Action Plan

If you're ready to adapt, here's your 30-day plan:

Week 1:

• Audit your current campaigns. Are you running the old ad set structure?

• Create one new campaign with one broad ad set (Advantage+ on)

• Set minimum age and value rules if needed, but keep it broad

Week 2: 

• Audit your existing creative. Identify your 5-10 best performers 

• Create 30-50 new ads using the 50/30/20 rule (refreshes/adjacent/new) 

• Ensure format diversity (video, static, carousel for each concept)

Week 3: 

• Launch all ads simultaneously 

• Monitor daily for non-spenders 

• Prune ads getting zero impression after 5-7 days

Week 4: 

• Analyze survivors and understand what's working 

• Plan your next drop (due in 3-4 weeks) 

• Document your creative themes and angles

The Bottom Line

Andromeda isn't a minor tweak. It's a fundamental infrastructure change that requires a strategic response.

The advertisers who adapt, who embrace creative diversification, broad targeting, and portfolio thinking are seeing efficiency and scale that felt impossible six months ago.

The ones still running 2023 playbooks are watching their costs rise and their ROAS decline, wondering what went wrong.

Meta shipped this change in December 2024. Deployed it fully by July 2025. And said almost nothing about it.

Now, you have deeper insights into how the Andromeda update has changed ad delivery on Meta - with active steps and strategy that you can implement to scale your ad performance. 

Need Help Fixing Your Ad Account Health?

Andromeda also brings stricter controls and measures against advertisers who fail to follow Meta ads best practices or are in violation of Meta’s advertising policies. 

These can lead to hidden penalties and silent restrictions on the backend that throttle ad delivery of an account as a mechanism to reduce risk and exposure for users on Meta’s platform. 

These penalties and restrictions can hurt your ad strategy and performance. You need to work with a trusted and reliable partner that can identify these issues that are impacting your accounts and assets on Meta - and implement solutions that deliver real and scalable results. 

Scoreify is the preferred growth partner to many of today’s top ecom brands, advertisers and agencies who rely on its portfolio of Meta ad services and solutions to scale their business and maintain high trust rankings and reputation across Meta platforms.

Get in touch and book a call to learn more about Scoreify's Meta ad services and how it can help you stay ahead of everyone else.