If you're running Meta ads the same way you did in 2020, you're probably underperforming and you might not even know it. The platform has changed significantly over the past couple of years, and most advertisers are still using strategies that made sense back then but actively work against them now.
This isn't a small shift. It's a fundamental change in how Meta decides who sees your ads, and the advertisers who adapt are pulling away from the ones who haven't.
The Old Way: Interest And Demographic Targeting
For a long time, the Meta ads playbook looked like this. You'd build tightly defined audiences. Women aged 25 to 45 who like yoga, follow specific wellness pages, and live in major cities. Then you'd run separate ad sets for each audience, carefully testing which one converted best.
The logic was simple. You knew your customer better than Meta did, so you'd tell the algorithm exactly who to show your ads to. Control the inputs, control the outputs.
That approach worked when Meta's targeting algorithm was weaker. The platform genuinely needed your help to figure out who was a good match. Detailed targeting compensated for limited machine learning, and advertisers who got the targeting right had a real edge.
It doesn't work that way anymore.
What Changed: Meta Andromeda
Over the past two years, Meta rebuilt its ad delivery system around a model called Andromeda. It's a machine learning system that processes thousands of signals per user to predict who will convert, and it's dramatically more sophisticated than what came before.
The shift is philosophical as much as technical. The old system was built around the idea that you, the advertiser, would tell Meta who to target. Andromeda flips that. Meta now figures out who to target, but only if you give it enough signal to work with.
Here's what that means in practice. When you stack tight interest filters on a campaign today, you're not helping the algorithm. You're constraining it. You're telling a system that can analyse 52,000+ data points per user that it should ignore most of them and only show your ad to people who match a handful of categories you guessed at.
That's like hiring a world-class analyst and then refusing to let them look at your data.
The Modern Meta Ads Playbook
The advertisers seeing the best results right now have stopped fighting the algorithm and started feeding it. The new playbook looks very different from what worked in 2020.
Use Broad Or Advantage+ Audiences
Instead of tight interest stacks, go broad. Run open targeting (just a country and age range) or use Advantage+ Shopping campaigns where Meta handles audience selection entirely. This sounds reckless if you're used to the old way, but Andromeda performs best when it has room to find buyers you wouldn't have thought to target.
Run Fewer Campaigns With More Creative Variations
The old approach was to split campaigns by audience. The new approach is to consolidate campaigns and split by creative. Run one campaign with five to ten creative variations inside it, and let the algorithm figure out which combinations of creative and audience work.
This also solves a problem most advertisers don't realise they have. When you fragment your budget across many small ad sets, none of them gets enough conversion volume for the algorithm to learn properly. Consolidation gives Meta the data density it needs to optimise.
Concentrate Your Budget
A related point. Andromeda needs roughly 50 conversions per ad set per week to exit the learning phase and optimise effectively. If you're spreading $50 a day across six ad sets, none of them is getting close to that threshold.
Fewer ad sets with bigger budgets almost always outperforms a fragmented setup at the same total spend.
Let Meta Find The Buyers
This is the mindset shift that ties everything together. Your job isn't to define your customer for Meta. Your job is to give Meta clean signals about who converts, then get out of the way.
The advertisers who resist this are usually the ones with the strongest pre-existing opinions about their audience. The advertisers who lean into it are usually the ones seeing 30 to 50% better return on ad spend than they did under the old playbook.
Where Tracking Comes In
None of this works without clean, complete conversion data. Andromeda's optimisation is only as good as the signal you send it. And here's where most advertisers have a much bigger problem than they realise.
If you're relying on the standard Meta Pixel alone, you're probably missing 30 to 40% of your conversions. iOS 14.5+ restrictions, ad blockers, third-party cookie blocking in Safari and Firefox, and browser-level privacy features all chip away at what the pixel can capture. Some industries see signal loss above 50%.
That's catastrophic for Andromeda. The algorithm is trying to learn who your buyers are, but you're showing it an incomplete and biased sample. The conversions it does see skew toward Chrome users on Android with no privacy extensions, which probably isn't representative of your full customer base.
The result is an algorithm that thinks it knows your customer but actually doesn't. Your broad audiences won't perform the way they should, your Advantage+ campaigns will struggle, and you'll likely conclude that the modern playbook doesn't work, when actually you just haven't given it the data it needs.
This is why server-side tracking through the Meta Conversions API (CAPI) has gone from a nice-to-have to essential. CAPI sends conversion data directly from your server to Meta, bypassing the browser entirely. Ad blockers can't block it. iOS restrictions don't apply. You recover most of the signal the pixel misses.
When advertisers add proper CAPI implementation to a modern campaign structure, the difference is usually obvious within two weeks. The algorithm finally has enough clean data to optimise properly, and broad audiences start performing the way the case studies promise.
The Bottom Line
Meta ads in 2026 reward a different set of skills than Meta ads in 2020. The advertisers winning right now aren't the ones with the cleverest targeting logic. They're the ones who feed the algorithm clean data and let it do what it's built to do.
If your performance has been sliding and you can't quite figure out why, this is almost always the answer. Old playbook plus incomplete data equals algorithm starvation. Modern playbook plus clean tracking equals algorithm doing the heavy lifting for you.
