Transcription
So, they're actually going to absorb the cost data for your products and present back to you a marginal return number. Then, ultimately, the final boss of incremental marginal return—and that's where this is all heading. And guess what? That frees us from having to use cost controls at that point, because now the optimization setting actually matches the business outcome. So we can remove needing to assert for ourselves that cost cap that's defining that result for the business.
This episode of the e-commerce Playbook is brought to you by OpenSend. Whether you're losing customers to untracked sessions or missing out on abandoned cart revenue, OpenSend has your back. OpenSend's identity resolution tech helps you identify anonymous, high-intent website visitors, reconnect with your customers across previously unrecognized devices and browsers, and replace bounced emails with active ones. 1,000 top DTC brands use OpenSend to drive conversions and retention with unmatched precision. Go to opensend.com/briowfc.
Content as it relates to—I'd say, of the things that we are most well-known for—I'd like to put like marketing and finance connection up there. But whether I like it or not, I would say that we have been deeply integrated into a very specific form of media buy, as it relates to cost caps in particular. A lot of my internet arguments over the last few years have been related to this topic, and we're not here to abandon that entirely. But I would say that we are undergoing the largest fundamental shift in the way that we think about and structure Meta ad accounts that we've had in a long, long time, and I'm excited to chat about it today.
Yeah, so the context for this is—I'm sure most of you are aware—Meta's rolling out Andromeda, which is sort of the next-gen of Advantage+ automation that incorporates AI into the algorithm, or sort of the general way it works. So I think what we want to talk through today is a kind of what this is, what it means, and then what the implications are, particularly for the way that we think about campaign structure, general media buying, all that type of thing. And I should note here upfront that we're putting out a video on Friday, another Sharpen Your Skills featuring your friend in mine, Taylor, about this specific topic that'll go a little bit more in depth. But we want to kind of tease that today and give a little bit of a summary about what's coming down the pike. So let's get into it. Taylor, talk to us a little bit about what Andromeda entails, sort of generally.
Yeah, so I'm gonna cover a few things, including Andromeda today, that I think are real change—changes to the way that Meta is designing their ad system to serve its advertising partners. And how then, in light of that, we are restructuring the way that we think about designing ad accounts to align with those changes. And unsurprisingly, the biggest piece of this to understand comes from changes as it relates to Meta's relentless focus on AI.
So, for many people, I think the important thing to understand about what AI offers is just insane amounts of computational power packed into individual applications of elements. So like, imagine when you sort of query a question on Google—that what is happening historically, if you search for something—is that you're getting an index presented back to you, right? This list of things that match your query based on contextual data elements contained within different websites on the internet. And Google did this monumental task of sort of like contextualizing the entire internet, indexing it, and providing it back to you in the, you know, the 10 blue links, right? The very famous thing. But what AI really is is just a massive step-function change of not just presenting you back the choices, but actually consuming all of the material of the world and synthesizing it down for you, right?
And so you can imagine just like a—this way to sort of think about computational power—sometimes I like to just think about like the illustration of—let's use Richard's books behind him, right? The amount of mental energy and time that would go into Richard reading all of those books, consuming the information and responding back to a question about which one of them is best would just take a monumental amount of time and brainpower, right? But what AI is functionally doing is compressing all of that down to an instant, right? And so when you hear these stories about these clusters of giant compute centers that all of the Magnificent Seven—the Metas and the Googles and the Xs and all the big businesses of the world—are building these giant data centers to house these Nvidia superclusters, it's all about this computational power to consume mass amounts of information and instantly, quickly synthesize it and deploy it on behalf of you, the user, in the case of ChatGPT, or in our case, you the advertiser.
And so what Meta is doing with this product—which is something that Meta is so good at—is that they're productizing technology in a way that allows us to enter into it. So Andromeda—think of it as a marketing PL—it's a description that allows advertisers to begin to understand what all this underlying technological infrastructure investment is going to do on their behalf. And so it's really important—I would say, if you're in e-commerce, you have to understand how this then affects how you should think about advertising in your account. And so we're going to go through today a little bit about Andromeda, which I think is a really important change, as well as then some of the specific things that it enables in that accounts that Meta is beginning to bring to life in different form.
Yeah. Okay, so let's talk then about—again, digging a little bit more into Andromeda. So you talk about the way that sort of AI functionality generally has sort of transformed, or will transform, the way that—I don't know—the internet is used or whatever. But there's one specific way that we talked about previous to hitting record here, which is around like the aperture or window of time that the algorithm can use to sort of track or judge your behavior and then serve you ads. So talk a little bit about how that kind of complicates things, or just makes them sort of more complex and useful.
Yeah, so if you think about you as a user of Meta ad products—of Facebook itself, of Instagram, of IG stories, of WhatsApp, of Threads, if you happen to be on there too—you have a long history of behavior. You have a long history of clicking around on websites that also have Facebook pixels on them that you could imagine in your head—if you close your eyes and think of like a big data visualization map of everything you've ever done on the internet—as scary as that might be for some of us to imagine being really released into the world that exists, right? And if you think about how that information could be used to decide the next ad to deliver to you, what you start to understand is why we often get this sensation or phenomenon that Meta's listening to us. It's actually just the sort of complexity of the map of information they have about us—from location to the apps that we use, the things we browse on the internet, to the things we've clicked on on Instagram—it really is true that they probably know more about you than your significant other or closest friend or whatever it might be.
But what has been true historically is that the amount of information that they would bring to bear in order to allow the ad algorithm to consider that context to make ad optimization decisions has had to be fairly small, because it requires an immense amount of compute power in order to have a really large context window, right? So if you've ever noticed in ChatGPT, like this idea of context windows or memory are often very limited because it actually—were they to store all the information about every user of the product—it would just overwhelm the system in terms of the amount of computational power that that requires. But this is why Meta is investing so much—is so that they can increase the personalization to every user based on larger context windows and more data. And that's what Andromeda enables. And one of the things that they tout specifically is that by leveraging advanced deep neural networks, unprecedented parallel processing, and real-time personalization, allows them to overcome traditional modes related to latency, memory bandwidth, and computational intensity. And the benefit is an 8% improvement in ad quality for the user and a 22% increase in ROAS for advertisers adopting Advantage+ tools.
Okay, so that—that's the benefit of all of that increased computational power on behalf of both the user and the advertiser, right? So I mean, ultimately, like the quick way to say that is that it'll just serve people better ads, and we can kind of expect a better return from that. I mean, I'm here in the—in the kind of article about Andromeda on the Facebook engineering site, and Ian, we can throw this up, Corey, when we produce or actually put this out. But there's a very difficult-to-understand graph here that I'm looking at. Do you—hey, I guess the question is, do you understand what this is trying to say, and is that useful to talk about? I guess?
Yeah, so so this gets to that visualization, and we should put it on the screen at this point, Corey, and we'll put it in the show notes as a reference to the second really important benefit. So think about what I just described as like Meta personalization—think about that as like further emphasis on the elimination of you specifying audience targeting and more allowing Meta to sort of continue to handle ad delivery and targeting for potential users for your product. Okay, that's—that's like really the end benefit of that, which is more of what Advantage+ shopping is pushing anyways. The second most important thing, which is something we've talked a lot about, which is the emphasis on creative volume. Okay, so what this is saying is that if you think about the idea of the ad corpus as being like all of the potential ads that might matter or be able to be referenced by Meta for the sake of then considering delivery into a user—if you go back to the old days of like ABO—they would tell you that there was a real limitation to the number of ads that you could put into consideration for the auction. And now with Andromeda, what Meta is saying is, no, no, no, we're actually going to give preference to brands that provide more creative options because you are actually giving us more leverage to create more pairing between user and ad. Okay, the more options you give us, the more likely it is that we can create a match between ad type and person. And so it's just even a further emphasis saying that Andromeda is enabling us to consider the possibility and delivery of more creative volume on behalf of the advertiser so that it can—this like new retrieval system—the way they phrase it is this: they say Andromeda employs hierarchical indexing to efficiently handle an explosion—their words, explosion—in ad creative volume driven by AI-generated content, enabling advertisers to scale campaigns without performance degradation. Which is them pushing to say, take these AI tools to create more variations of ads, different headlines, thousands of different backdrops, thousands of different headlines, thousands of different creative audio overlays, and allow us to find the perfect match between user and creator. And that is—this is like sort of a—I feel reinforced by this realization, which is to say that the dream scenario is that there's sort of like infinite potential variables to match to each unique person, and you could process through all of those options to find the perfect pairing, right? This idea that there's a soulmate of an ad for every person, and if you had infinite options you could find it. And that's sort of what they're getting at—is that we can now process more possible relationships between users and ads to find the right match to get you the best results.
95% of your web traffic leaves without a trace—no purchase, no opt-in, no opportunity to re-engage. Meet OpenSend, the platform that helps you identify unknown site visitors and re-engage returning customers, even after they switch browsers or devices. With OpenSend, you can identify high-intent unknown site visitors in real time, automatically recover bounced emails, retarget via email, SMS, postcards, or ads, and plug directly into Klaviyo, Braze, and your favorite ESPs. It's like giving your marketing team superpowers. Try it for just $1 at opensend.com. Book a demo, right?
So I—it kind of sounds like—so every time like something like this rolls out, it feels like the upshot tends to be, creative is the most important thing—make more creative. And it sounds like this is, in some ways, just like the most intense version of that so far, or maybe the kind of like apex of that movement now. So what's different about this other than just an order of magnitude, like… okay, yeah, go ahead. Let's get into…
So, so very practically, if you think about our media buying structure at CTC, there's a video on the internet that you could go look up, which is a reference to what has really driven our account structure historically, which is called the Pipes methodology from CTC. It's this idea that we think of every campaign as its own unique pipe that you're trying to get delivery through or flow through, and that the cost control—control—is this like lever that controls a constraint on how tight the expectation is and how much flow can get through. And the way that you increase flow through an account or dollars spent is by launching more campaigns. And that was historically true because there was a limitation to new creative existing within the same campaign or ad set. But what now they're saying with an ASC campaign is that that's no longer true—you can sort of add an endless amount of creative into this campaign, and Meta can process through it and find the delivery. So it's changing for us; it's bringing a more consolidation at the campaign level and more ad creative volume per each campaign. We used to have this relationship where it was like three ads, six variations in each campaign, and that was like this defined constraint that's now gone—it's out of our system. We don't even use the language, concept anymore, which referred to a campaign with three ad variations. So we're now unbounding the relationship between campaign and creative. And in fact, if you have a high-performing campaign around a core offer that's going to exist for a long time—maybe one of your best SKUs—there's actually going to be a repeated process of consistently adding new creative to that campaign for the sake of elongating its ability to continue to scale and spend. So the basic—I mean, to kind of recap a little bit of what you're talking about or contextualize it—our creative sort of structure has been, for a little while—at least a year and a half, maybe two years—this idea that there's offer, angle, and audience—any combination of those three equals a concept. A concept is a new campaign; each campaign contains six ads. But it sounds like essentially what this is—it sounds like this is doing away with ABO, is that basically…
Yes, exactly. Everything is going to move to default ASC, so there's ABO going away—as a promise—that's… yeah, that's the end tale of this sequence of actions that Meta's taking, and really that's just about getting AI enabled into all of the ad product delivery more than anything else.
Yeah, so that—that's the piece of it. The other—the other things that this does—AI also has allowed Meta to begin to introduce more complex optimization settings. Okay, and this is really where what I get excited about for the future of advertising is—is that if you think about why we have fought so much around cost controls, there's a very simple reason: is that they represented the ability for an advertiser to assert their desired business outcome into the algorithm in a way that did not exist before. When you were bidding and your only choices were lowest cost of acquisition, okay, or highest value of purchase value—those are the only two options—do not—do not include the ability for you to optimize around the business outcome I know people actually care about, which is the efficiency of those results. So if it's lowest cost of acquisition, there's always an inferred or necessary relationship to AOV, which is AOV divided by CAC equals ROAS—so the efficiency of that acquisition. And if it's value, then there's also the inferred missing CAC, or the cost of that value, and it's—it's actually the relationship between those things that the cost—the cost controller, the target ROAS, was always inserting. But what Meta is doing now—they started by saying, hey, we will allow you to optimize for third-party metrics first. They said you can optimize for a last-click result if you want; you could optimize for a North Beam ROAS; you could optimize for whatever third-party metric you care about. That was the first additional optimization result they cared about, but they also recognized that that's not actually the solution either—those are still just proxy metrics. And so now the next set of things—we're in betas for both of these—we have a large-scale study going live for one of them—around one incremental revenue creation. So this idea that where they take this process of geo holdout and measurement allows you to optimize for incremental results, okay, is one. And then the final stage—the Godfather of this all—is profit optimization. Okay, so so they're actually going to absorb the cost data for your products and present back to you a marginal return number, and then ultimately the final boss of incremental marginal return, and and that's where this is all heading. And guess what? That frees us from—we don't have to use cost controls at that point, because now the optimization setting actually matches the business outcome. And so we can remove needing to assert for ourselves that that cost cap that's defining that result for the business.
Right, so in other words, the cost cap is essentially a stop-gap measure in a lot of ways—is it's like—it's always going to push towards one of those two objectives, and the cost cap artificially inserts this restraint in order for us to kind of jerry-rig it to get to the profit goal that we want. But at a certain point, that's just not going to be the case anymore.
Well, well, you know, it—assuming like—if you think about the idea of incremental marginal return, there's still going to be this need to define what term you would accept. And so maybe—maybe we still have to say greater than zero, you know, like or whatever it might be, but because there's some point in which people are always willing to spend to a loss at different scenarios. So there's probably going to be some version of it still, but the idea is it's moving closer into alignment, and this is all enabled by increased computational power, where the idea of connecting the client's data and additional information combined with the ad delivery and optimization is enabled by the power of the additional compute through AI.
So so all of this is really exciting and opportunistic, but it just changes the way that we bid a lot. The other big change for us is that historically we have been really rigorous around the optimization setting that we would choose in terms of Meta. Right now, offers one-day click, 7-day click, 7-day click, one-day view, one-day click, one-day view—those are the only optimization settings. It used to be 28-day click, one-day view—that went away. There's now only these four settings. And so we used to push really hard to move everything to 7-day click only, and now what incrementality and measurement as a baseline does is it allows you to say, hey, I—I don't actually necessarily care which one we're using—we're going to use an incrementality study in either case to help build the relationship back to the actual business outcome that you care about, and then we can factor a 7-day click, one-day view, or we could factor a 7-day click. And in more cases, we're actually getting comfortable with factoring against 7-day click, one-day view, because that adds the most signal back to Meta. Now there's all sorts of reasons—you gotta be careful, and there's different cases—but it's just something we're holding looser than we used to, I would say. So we're fewer campaigns, more consolidated, more creative per campaign, still broad targeting because we believe that's the way that uh Andromeda and their—their understanding of context windows—is better than mine, and so we're going to allow for that breadth of delivery. And then we're going to have—ideally, we're going to use incremental optimization or profit optimization as those things become available and are tested and validated for their efficacy. But if not, we're going to use maybe 7-day click, one-day view, but we're going to pair it with a measurement test and then deploy an iROAS goal into that system. So these are big changes in terms of the way we think about the structure.
Yeah, and so one thing I don't think that you mentioned that it's actually going to stay the same, or rather most important holdover, is the idea that each campaign is built around one expected AOV, right? Or one expected CAC, or whatever. So whereas previously we would change out or build a new campaign for a new angle or a new audience or whatever the case may be, at this point is basically dumping in creative into one campaign, one ad set, and then having—but making sure all of them are pushing towards the exact same offer, so—or the same things that are related to the same margin profile. So because then you could use a TRAS, auto-value optimized, and you could get to the same marginal result, right? So the key is—yeah, as long as—and the other reason we would separate it out is if there are inventory needs based—so in other words, I want to make sure we have some clients that are like, we need a certain amount of budget every month going into women's versus men's. Okay, cool, well let's make sure that we're getting budget into both of those areas separate and not just consolidating that allocation to Meta in those cases. So it's either inventory demanded or it's separate offers or designs are reasons why that we would be changing them out. And then the last thing is is related to exclusions—like in an ASC, we are setting up our campaigns where you're now defining the audience exclusions at the—at the account level. So you're defining what an existing customer is, you're defining what an engaged audience is, and you're defining what a new customer audience is at the account level. With ASC, we're doing that in a unique way, which is we're using new customers to mean what you would expect it to mean, but we're using engaged audience—a lot of people do this as like remarketing—I don't really care about the distinction in remarketing; what I care about is the distinction in active versus lapsed customers. So in engaged audience, we're going to define as your active customer base, and that's who we're going to want to exclude. And then your existing customer base are going to be your lapsed customers—the ones that are not engaged functionally. So those are distinctions where we're going to allow for those people back into the funnel in most cases in order to inform the delivery, because for a lot of—especially larger businesses—we're going to want to make sure that those people who very much are not responding to you via your free marketing channels are getting—are making their way back into the educational funnel. The other thing we feel a lot more freedom to do now with incrementality and measurement is to bring in other campaign objectives you want to try—add to cart, you want to try reach, you want to try video views—those things are easily enabled by pairing it with a holdout study that allows us to look at, okay, what is the impact of some of these more higher-funnel actions? People have this phrase, so I'm just going to use it—video views, add to carts, page views, etc.—where we can run that, and we can pair with the holdout study, and we can see the incremental impact. So that's actually usually very easy to do because it's a true holdout where you're actually just running it rather than turning it off in some regions—you're actually turning it on in just a very narrow, select set of regions; you can run the test and see that result, and then we have a measured way to apply a result against that, which has historically been very difficult to do with a traffic study or a reach campaign to really understand what the result or impact is. Well, I—this really is revolutionary.
If I'm hearing you say that you might build an ad-to-card campaign, or a video view campaign, or whatever, yeah. Now, to be clear, I think that I would be very hesitant to tell you that I am very confident in the result of that impact. But I, I don't actually—I want to be less and less—I want to only lean on the evidence that I have available to me as much as possible. So, if I've run 400 studies of XYZ, I can let you know. But if there are cases where people have seen add-to-carts be incremental, then I, I'll hold those as appropriately, and as many as I have, and as repeatably consistent the results are, I want to speak through the lens of that confidence.
I think we're in a very new era of Meta advertising, and so a lot of the norms and heuristics we have to allow to evolve into the present and to understand that what might have been true on Meta in 2019 may not be true on Meta in 2025. And we should do our best to update our priors and to reconsider the present as best as we can and make the best decisions for today. Yeah.
All right, so let's talk about then, real quick, the what the timeline looks like on this. So, as this is—will be coming out on Tuesday, the 25th of March—what will be happening already as we're listening to it? What's coming down like—what do we know is coming in the near future? And then what's a little bit more vague, and what sense do we have on the timeline for those things?
Yeah, it's a good—it's a good question. So many of these changes have already rolled out for Meta, in that the—as this changes to Advantage+ shopping—are already underway. It's now the default way that you would build a new campaign; it's going to opt you right into that from the start. So, on that core underlying Andromeda algorithm change, that's sort of in place, some of these other new optimization settings, like I mentioned, we are right now running a 10-client, agency-wide incremental attribution study across a series of clients to understand the incremental impact of that optimization setting. Same thing with profit optimization; we have that in a couple of places where we're trying to be diligent to work very closely with the Meta data science team to release those things and then go from there.
And then, as is Meta's case, there's also a lot of other exciting things coming. There's WhatsApp ad placements that are down—coming down the road. Very—they actually just announced this past week—uh, notification ads. So there's a new ad placement now in Meta. If you go into Facebook and look at your notifications, you're now going to get ads in there. That's a new—a brand new, net new placement that'll be coming from Meta. They've—they're launching Omni-Channel ads, which is the ability to deliver ads if you have first-party store ownership. So let's say you're Travis Mathew and you have your dotcom, but you also have retail stores where the ad can both allow the user to click to the nearest map location, and it will track and report on those purchases, or they can click to the website. So that's—there's some new Creator content-related ads that allow you to run collection ads alongside Creator content. So continued evolution of the ad product for sure, but it's exciting. There's a—there's more changes to Meta, and—and I think in an era where AI is so rapidly evolving, that's what you would hope to see out of an ad product. And now we can continue to lean in and figure out how to deploy these new tools to the best result.
All right, folks, so this episode of—of Sharpen Your Skills—is coming out on Friday; this episode on Andromeda, so you can dig a little bit more—he'll go deeper on this particular topic. And of course, as always, if you guys are out there, you want to work with us, you want us to put some of these changes into place for you, comtho.com, hit that highest button; we would love to chat. Taylor, anything else you want to hit on this?
I—I would just—the last thing I would say is I know I am a broken record on this creative volume thing, but I really think that nobody has really figured out how to do this well yet. Like, it has reframed their brain around like—we're not talking about 50; we're talking about 5,000. And the way that you enable that is with—it—Meta is offering you the tools, and so many brand leaders are hesitant around like, "Oh, what about if I don't like the music over the top of my ad?" And I just think, if—especially let's say you're a brand that might be struggling a little bit—if you're willing to just deploy these things absent that, and allow the process of the user's response to the creative to be the end-all be-all, there's going to be gain in this new system for people who lean into what Meta wants out of it and what they're telling you it values. So I just think that continuing to think about your supply chain of ad creative and how you could 10x, 100x, is really important, and then to allow for the least human intervention is possible.
To—what—here's—I'm gonna say a thing—I've said this before on—think—is that I believe that 99.7% of all media buying, including media buying that happens at CTC in many cases, is a net negative on the results of the business. What do I mean by that? I mean that humans, in the way that we interact with a system like that, it is so hard for our brains to map computationally what's happening with—what's happening inside of this—this system of complexity that is Meta—and to make decisions that we think are affecting things in ways that they just aren't. And I really believe that this is all going the way of hedge fund and stock trading—is that this is like fly boys is coming for our industry because we're just not good at it. Like, we—we—right now, the studies will someday show that just like most portfolio managers don't outperform the index, I think that's true of most media buyers too, and that—that we have to allow for some of this automation to continue to press forward—ward—and we're going to get better results as we do that. And so your job is to be clear on your business objectives, to be clear on your inventory, to be clear on the offer design, and then to fuel this machine with as much quality creative as possible.
If you reframe like a very simple media buying structure that says we're not going to spend all our time and energy worrying about this structural design, we're going to build ASC campaigns, we're going to do it to broad audiences, we're going to do it around the right offer design with a clear target, and then we're going to build the—of the system—around deploying as much creative volume and thought into that as we can, and then go build the marketing moments and branding exercises that we talk a lot about—those businesses, I think, are going to drive immense value off this platform in the next 12 to 24. Right.
All right, cool. Well, we'll have to do a—a pod on—on creative volume enablement as it becomes even more important moving forward. But all right, folks, well, good chatting, and we will talk to you again next week, everyone. Goodbye. [Music]