# Meta Ads Will Never Be The Same in 2027 (Andromeda 2.0)

**Channel:** Justin Lalonde
**Source:** https://www.youtube.com/watch?v=tDwzSLtBKqg
**Transcript page:** https://www.withtranscript.ai/video/tDwzSLtBKqg

## Chapters

- 0:00 — Meta's Andromeda 2.0 Announcement
- 2:13 — How Meta's Ad Algorithm Works
- 6:55 — What Andromeda 2.0 Changes
- 11:33 — Practical Strategy and Performance Signals
- 15:03 — Common Mistakes and Final Advice

## Transcript

**[0:00]:** Meta just released two new articles on the next version of Andromeda 2.0. Now, half of YouTube is already telling you that everything has changed. I read both and here's the truth. If you've been following the channel for a while, you're good. If you haven't, 2.0 is probably going to cost you more than 1.0 did. In this video, I'll show you exactly what was already true in Andromeda 1.0, what is now new and changed in Andromeda 2.0, and the exact setup and creative strategy I used to run along with it. Very quickly, I'm Justin Milan. I'm the founder at pitadvertising.com where we profitably spend and manage over $20 million in meta ad spend every single year on the behalf of our clients. And let's get straight into today's video. I'll actually leave the link to both articles in the pin comment down below if you guys want to read them yourselves which are both having posted on the 28th of October 2026.

**[0:47]:** One is as a business announcement and one is in their engineering division. So, quickly showing you what those are about before showing you essentially I prepared a bit of a deck to simplify a lot of that information for you today because a lot of that is technical jargon that is not super useful. But if we start with the engineering side and we scroll down, basically all they're saying here is that with time there's more and more advertisers and of course there's more and more ads and specifically with AI there are more and more ads now than ever at a faster rate which makes this harder on Meta's hardware and servers and their algorithm to be able to retrieve all those ads and show them all to the right people. And so they had to invest into a better backend, better tech, better GPUs, better hardware limits to basically make it so that they can more easily retrieve more ads at scale, which has then caused them to increase their

**[1:37]:** hardware capacity and therefore slightly change their algorithm to meet with the new hardware capacity they now have. And now they hit you with this quite complex graphic right here. And so I'll stop myself there on this article and then move on to this one right here. And now they're reintroducing you to what they deem to be their concierge, which is Meta Andromeda, and talking about how it retrieves AD and Meta Gem and all of which we already knew last year, but they're kind of talking how now this is better. So again, I made this very simple and easy to follow now with this kind of Mario/ video game themed infographics that you're going to see on screen right here.

**[2:13]:** Okay, so let's start here. Okay, how does meta as algorithm actually pick your ad? So there are four parts to their algorithm. There's metagam, there's metal lattice, meta andromeda and what they call their sequence learning. Okay, metajam they think of it and they explain it as the brain. It remembers everything people do on meta apps. Okay, so it knows the whole customer journey. Every video, every ad, every click, every save, every share someone has ever interacted with. Now metal lattice is the librarian. So this is essentially where every placement objective and all of that data about every user, it lives there. So this organizes all the ads that it could possibly show to an individual. Meta Andromeda is the concierge. It says this is essentially what actually pulls out the right books. In other words, the right ad and shows it to the right person.

**[2:59]:** And finally, we have sequence learning, which is essentially what's next. So the example right here is if as an example, you would have bought a ski pass for a lift for a day. Well, they're going to show you hotels nearby, maybe ski uh gear, also maybe some gloves that are heating up to keep you heated throughout your ski day, right? So, in other words, you know, we have brain, the librarian, the concierge, and then essentially the what's next part already handled. Okay? And this was new in 2025 when Andromeda came out. And I have it right here. I call Andromeda the concierge in October 2025 already because that's how Meta referred to it. Now to quickly show you if we now move a level down and we go to creatives okay and we go to how does that affect me at the creative level and what do I need to do with my

**[3:41]:** creatives some of you probably already know of awareness levels if you don't very quickly your creative picks the awareness level to how aware the prospect is and meta essentially through going back to sequence learning it'll help you show different ads to people based on where they are in their awareness level journey because Andromeda can pick out those concepts steps from the library because they know what you should see next. Okay, so going back right here, it's very important that you make ads in your campaign of different awareness levels. What people consider top of funnel, it is unaware, problem aware, solution aware ads. Middle of funnel would be problem product aare/ sometimes solution aware and then most aware would be bottom of funnel ads. So examples of that is you know that 3pm crash isn't normal. That's a hook that is unaware now always tired even after eight hours.

**[4:30]:** They know they're tired, right? So that's problem aware. Solution aware. Why electrolytes beat a third coffee? They know of other solutions. So like a coffee but not of electrolytes. Product aware. You wanted them to convince. You want the audience actually to be convinced to buy your own product. So reviews, testimonials, right, is what you would do here. And then finally, most aware they know of you, your brand. They just haven't found the right time to buy. So you hit them with an offer. And typically the more top offunnel you are, the bigger the audience, the lower your CPMs, the less the more aware you are, the smaller the audience, the higher the CPM. Okay. Now, I personally typically do and I still recommend no retargeting campaign, one simple campaign. I'll get back to that later, but essentially one campaign that is brought targeting and your awareness level in the creatives does the targeting because once again, if we go back, the concierge is able to find the right ad

**[5:20]:** that has the right awareness level for the right person because of this sequence right here, right? The brain knows that the user has already seen an ad before that was problem aware. And through sequence learning, it knows that the next ad it should see should probably be solutionaware. And so it'll tell the librarian, hey, find me a bunch of solution aware ads and organize them all together and then Andromeda will go into that section specifically, find the right solution aware ads and show them to the user. And so if we keep going, the campaign structure I use is one campaign per offer. So as an example, we have campaign A for offer A for landing page A. It is broad. You only select whatever country you want to advertise to. You leave placements to automatic. Anyways, most ad accounts right now don't even have the option anymore to switch uh to manual placements.

**[6:05]:** You pick leads if that's what you want. So, leads, calls, etc. schedules, right? Or sales if you're an ecom brand, a low ticket funnel, etc. And the way I usually have it is this. I have one ad set per concept. I have one ad concept per adset. And I typically have three variations of this ad concept. Could be three different formats, three different content creators. basically one variable per concept that would change but they're overall similar concepts just having essentially one variable change between them. Okay, daily budget I typically do as an example here the math is daily budget divided by my target CPA is the maximum live adset that I have under this campaign. So as an example I have $100 a day in ad spend divided by a $10 CPA target. I could have up to 10 concepts live inside of my campaign.

**[6:49]:** And I only do one new campaign if I have a new offer or product, a new landing page or a totally new audience.

**[6:55]:** So if I keep going now, what's actually new in Andromeda 2.0 version is they kind of go from depth to breath. So in the 1.0 version, the concier if you want checks the popular shelves. So if I go back to earlier, the library would kind of function as helping Andromeda the concierge just look into the most popular sections. So it would go, hey, a lot of users typically come into this section. But the darker sections of the library, the ones where the light is flickering and whatnot, there's nobody going there to read the books. And that was the issue simply because Meta didn't have enough hardware power to actually have all the time their AI, their algorithm, Andromeda to go out and look at every possible book, analyze every possible possibilities of a combination of an ad that could show you on your feed. They would only look into what's like trending if you want and serve that to your users.

**[7:46]:** Now, the 2.0 version, if we go back to the analogy, reads every label on every shelf and routes each ad to specialist experts. So, in other words, right here, this is a great example of it. It's like okay before as an example you have all these ads in your account but you have as an example like five ads that are you know getting decent stats decent soft metrics they're deemed as popular and so Meta will like just go crazy at those spend a little more on those and that's it. Okay remember before Andromeda you used to typically find one two maybe three winners in an account. What Andromeda unlocked was more computer power to find you more winners. So that's why you were now in a position where Meta had a few more popular shelves. But now apparently from what they say at least in their articles with this 2.0 model is they can look at every ad, every possibility which means every unique ad of yours has now the same fighting chance at potentially

**[8:35]:** being delivered on someone's feed. Okay? And so this reemphasizes the importance of having different creative concepts, different formats, different personas, different angles because the more differentiation you have, the more possibilities the algorithm has at finding someone that might find one of your ad interesting. So here's an example, right? What happens is lots of ads never even get looked at or considered by the algorithm. And what is said apparently in Meta's articles is that before they were able to analyze and consider about 10 million ads per second roughly as a potential possibility to show to your audience. Now they're apparently able to consider over a billion ads per second as a possibility to show on your customers feeds. Okay, so they've more than 100xed the output or the throughput of ads they could potentially show to a customer. Now, continuing alongside that, they have a new recommendation model.

**[9:27]:** So, they call that Genrex right here, which this is basically an AI model that reads your entire ad. If it's a video ad, it will analyze every frame of that video ad, analyze the transcript of it, the sound, if there's a music in it. And so as an example, if you have this, you know, this is again an AI generated just like emoji based ad. You know, you have an ad. It's about, you know, babies and it says moms that 3 p.m. crash um isn't normal. And you have, you know, some like baby stuff in there. Okay. Now, the AI reads the text in the image and goes busy mom, all the energy. This is a problem aware person. They have an afternoon crash angle. This is a lowfi UGC hypothetically speaking. Uh it's in the kitchen. You have the kids around. And so it matches similar ads that already have data.

**[10:07]:** So in other words, like you and so what will happen there is they'll match your ad with either other ads in your account that have had similar data before or other ads of competitors that share similar uh variables if you want and then it'll be easier to match this ad that you've produced with a customer. Okay, which emphasizes the importance of making different concepts once again because if you have near duplicates, you only change one very very small variable on the ad as an example, you'll basically like AI will spend a lot of time and a lot of tokens on their side. it won't cost you anything, but just to say they'll spend a lot of time to analyze very similar ad and therefore be like, oh well this guy, they all belong in the same bucket. So like, well, I'm just going to pick whichever one is the best out of the bunch and then the other ones I won't really show them because they're all the same thing.

**[10:48]:** So why would I have them fight together? Whereas if you have different unique concepts once again and they're different pockets, Meta can match you to in the algorithm and show those essentially to uh to users. And a little reminder I put here at the bottom, if it's not visible in the ad, Meta can't use it. if it's not also said out loud in a video ad, meta has a hard time using it. So again, you really have to think, will the person I'm trying to attract recognize themselves in the ad? If they will, the AI will also catch that and therefore be able to match you with the right person. And now if you want to build yourself a system to help you create a stream of different ad creative concepts that all have different variables like different personas, desires, awareness levels, angles, and have your own version of a

**[11:29]:** testing matrix that fits with now being rewarded with Andromeda 2.0.

**[11:33]:** Check the link in the description below to join my ad engine school community and will help you implement our ad engine creative system inside of your ad account. Would you be an ecom or lead genen business? So anybody talking to you about an Andromeda 2.0 campaign structure is like honestly trying to sell you some This is not what I want you to even retain from this video where I had a very small portion about a campaign structure which the only advice was just go broad and focus on creatives. Now, what actually moves the needle outside of creatives is of course always going to be the outcome you optimize for. Whether you pick leads or purchases or whatever conversion event you pick will help also meta have more data because if you pick purchases, they'll also put you in the bucket of people who have taken actions before on the

**[12:12]:** platform that has made a purchase or going back to this last step right here of sequence learning. They'll then look at who has taken as an example the step before. So like an add to cart or checkout has yet to purchase and then you come out as an example with a product ad which makes sense to show that to someone who has added to cart and and or initiated checkout but hasn't purchased and since they know you want to purchase they know that's the next step you would want because again of sequence learning and therefore the event you're optimizing for matters and so does the quality of the data you also share and how much you let it optimize. So again, typically we let it broad automatic placements, budget allocated by meta. So CBL campaigns is what we typically prefer, especially on mass market, B toC or e-commerce offers.

**[12:56]:** And so Metaone numbers in their article says that each part of the system actually helps you get better results. So Jam as the brain apparently helps you get 18% better conversions since 2025. And Jamba 2.0 itself has only been improved by about 2.4%. It's funny, I use this in the title. I speak of Andromeda, but it's such a small part of the entire system. Genrex, you know, their new kind of like reading your ad system gives you a 2% boost, but the whole system combined together apparently gets you a 15.7% better visibility of your ads to again your potential customers. In my experience, in terms of what matters most, number one is the offer. A strong offer will still be the one thing that will get you the most amount of conversions. Number two is as simple as possible of a campaign structure that does not overwhelm your account.

**[13:44]:** Number three is going to be creatives. And number five is going to be of course your tracking setup. So the feedback loop you've set up within your account. And then how do you make good ads? Five pillars. Very simple again. Desire, awareness, sophistication, persona, angle. Okay. Desires. So as an example, someone who wants faster recovery, awareness, their problem aware, sophistication, what level are they at? Persona, what kind of persona? Overwhelmed mom, frequent traveler, etc. Angle, you know, the workout unlock the social armor. Just by using those variables, every single account can have over 3,000 different ad ideas to test, which is probably for most of you going to take you weeks, if not months, to get through all of those before having to start to reuse some combinations and iterate on winners. And this is just a general reminder of what we typically do when we have our ads up and running.

**[14:29]:** We let them live for at least 7 days. We have them spend roughly two times our target CPA. And if they're green, so they're beating our target CPA, we'll scale the entire campaign if it's a CBO by 20%. If the entire campaign again is in the green, if we're in an campaign, which we only typically use for B2B lead genen accounts nowadays, more than any other businesses, then again, we just do the same thing at the asset level. Yellow, so around target, profitable. I would just kind of leave it coasting for now. Then red, so way over target, no conversions, turn it off. And if it was green, but it's starting to decrease in profitability, then you can descale by 20%.

**[15:03]:** And very quickly, some ways that you could lose with Andromeda 2.0 is first of all trying to rebuild your entire account structure forward. You don't need to. Okay, this is a creative play more than anything else. Of course, forget interest in look alikes. Go broad. Third, very very important point and I think the most important point right here. A lot of people now with AI try to make so much volume of ads, but they make super similar ads. It doesn't matter. This will only overwhelm the algorithm further. You need differentiation of concepts. How do you do that? what I just showed you right here. You can use different pillars, different variables and interchange those to make over 3,000 unique ad ideas before you have to start going back into iterations and similar concepts. Now, people who kill ads too quick. In my experience, I have seen so many winners.

**[15:50]:** I've did that test many times on the channel, but to give you some example, most accounts we have done this test with, if we've killed an ad within 48 hours, keep in mind that having add the same ad live longer, about 50% of ads that do not perform in the first two days start performing in the following 5 days leading up to a full 7-day period. Which in other words, what I'm saying is by simply waiting a full seven days to have your ads run and get enough data, you could actually get 25% more winners inside of your ad accounts than had you just kill them too quick. And then of course, if you just mindlessly copy other competitors winners and just pump out the same stuff, you won't ever win because again, you're not different. And in this AI game, both as an advertiser, which most advertisers now use AI, and in the algorithm, which also uses AI to rank your ads, they will prioritize them.

**[16:40]:** And I cannot stress this enough, different creatives. And so make sure you focus on creating very different concepts from each other. And if you want to learn more about producing a high volume of different net new creative concepts using AI, check out this next video of mine called a new era of static ads is just begun where I show you exactly how to use Hicksfield and Claude's MCP to produce way more ads for your business.