# S3 E20 T75 - MetaRouter - Jon Flugstad

S3 E20 T75 - MetaRouter - Jon Flugstad 

[00:00:00] Podcast Plug and Setup 

[00:00:00]  

[00:00:00]  

[00:00:05] Tom Limongello: There we go. Hey, Scott 

[00:00:09] Scott Messer: Hi, Tom 

[00:00:11] Tom Limongello: I, uh, I see that you have been busy. You have your own pod now 

[00:00:15] Scott Messer: I do. I'm co-hosting another podcast called Ad Tech Therapy with my co-host Janna Meron, where we're investigating sort of the woes and troubles of ad tech and what it means to live in that life. It's a really great show. I encourage you to check it out. You can head over to adtechtherapy.com and submit your questions into the patient portal, and we will answer those live on our recorded podcast. 

[00:00:38] Scott Messer: So I hope to see you over there, and thanks for listening 

[00:00:41] Tom Limongello: Yeah, I, uh, I appreciate the, uh, the patient portal concept that you've created. Um, and I-- the last episode that I listened to, you were talking specifically about the question you get throughout your career and the one that you always have to answer, which is, "Why can't I just buy this over programmatic pipes?" 

[00:00:58] Tom Limongello: And that kind of gets us [00:01:00] into the discussion that we're gonna have with Jon Flugstad from MetaRouter today 

[00:01:04] Scott Messer: It certainly will. And it was a great discussion that we had with John. I'm looking forward to our episode here. Some of the key things that we sort of get into, besides we talk about LiveRamp and how that's gonna change and what his, uh, fantastic opinion is on it, is we also talk about context and the new definition of context that, that I think the, the newer elements of retail media are trying to apply to it. 

[00:01:28] Scott Messer: And of course, I tried to bring him back to the roots of the original definitions of context. So I think our readers will have a good time listening to it, and, uh, let us know in the comments what you think, uh, context means to you 

[00:01:39] Tom Limongello: First of all, let's, let's thank John Flugstad for coming after, uh, a, a difficult week where he had to be in my home city of Philadelphia before being at Cannes. So he's been, he's been doing the yeoman's work of not only doing retail media, but dragging his family around the globe. 

[00:01:54] Tom Limongello: But, I'm happy that he got a chance to stay a little bit in Philadelphia.  

[00:01:58] Propensity and Retail Media Shift 

[00:02:02] Tom Limongello: I'm excited tonight f- to see if, uh, Kyle Schwarber wins the Home Run Derby but the time that I have now to think as, as the sort of summer is, is burning on, um, the really big issue that is coming up over and over again as we talk to all these retailers, Albertsons, Costco, Kroger, has been about how they are moving into a new area. 

[00:02:23] Tom Limongello: It's a little less stable. Um, you know, old school retail media was you had a, an audience that was there for six months to two years, um, and that was completely deterministic. But now we're moving in a world where the word propensity comes up, and you are potentially targeting someone with the goal in mind that is probabilistic. 

[00:02:44] Tom Limongello: So the, maybe the data is deterministic, but the goal, you're not sure if you know exactly what they're trying to do. Are they trying to buy yogurt? Are they trying-- Are they on a GLP-1? Are-- All those types of themes. What was your take? Because some of this stuff is, is old hat ad tech, um, and, you know, I wanted to [00:03:00] get your, your thoughts before we get into the episode. 

[00:03:02] Scott Messer: Sure. So it, it's definitely true that companies like MetaRouter are bringing a new approach to propensity scoring and modeling, uh, in the speed and volume of data that they can bring into the table and the way that it can be processed. So I think it's a little bit of the same method, like what is propensity modeling and predictive targeting? 

[00:03:26] Scott Messer: That's the same stuff that's been done, but now it's happening with the speed and volume of data that is incredibly modern, and they are seeing a lift in outcomes, and it is moving the needle for them. So when you ask me, how do we know or how does a, a retailer know or a brand know if this is working, you're selling more products and more things are moving off the shelf. 

[00:03:49] Scott Messer: And likely for the retailer, their net revenue is up in total from a yield perspective because you're not wasting impressions on things that [00:04:00] people won't buy. That's being-- That's making a more efficient system.  

[00:04:04] RPV Metrics and Yield 

[00:04:04] Scott Messer: I, I still will tell you, Tom, I don't think retailers are looking at RPD or RPM and understanding what, like, the yield of their page views are, 

[00:04:15] Tom Limongello: Oh, I never heard RPV once when I was working at Quotient, so y- you're the one who taught me about RPV. Tell us about RPV 

[00:04:22] Scott Messer: Revenue per thousand visits, right? So how much are they driving? It's sort of the same way you think about like average order value, but then also add in advertising with that. Or maybe just look at the advertising portion alone and how much revenue are you driving from them. It's about session depth and how many page views are created. 

[00:04:39] Scott Messer: It's about the ads that they see. It's about the actions that they take. And this is a great metric for your CFO because you can always point to RPV and say, "We're doing more with less than what we had before." Or even if traffic goes up or down, we're still driving revenue at a similar rate through our pipes, [00:05:00] and the traffic problem may be a different department's challenge, but the revenue team is keeping up 

[00:05:05] Tom Limongello: Well, I think that generally where we are in retail media or commerce media is that there's the question of whether or not it's the merchant's game and, and their sales velocity or s- their sales volume versus what's happening on your own operator property or with your owned audiences and, and the margin there. 

[00:05:23] Tom Limongello: So I think that's just a challenge where those are two different goals for the CFO to think about, and it's, you know, it hasn't been resolved how this new model is gonna work. So exciting 

[00:05:34] Scott Messer: it, it, it at least gives it a level playing field, right? Where 

[00:05:37] Tom Limongello: Yeah 

[00:05:38] Scott Messer: the CFO may have like average order value and orders per customer, but they don't have the advertising revenue and metric and a way to think about that other than just some like total, total sum at the end of the year. You need to be able to divide it by a similar measure, and I would do that by visits and by page views, right? 

[00:05:57] Scott Messer: Because if you're not driving enough page views [00:06:00] through your search experience or your advertising experience, you're just not gonna be able to generate the impressions that ultimately generate revenue. And as I, as I say that, there must be a little bit of an inverse relationship because you're still in a CPC basis for most of these models. So what's the point of showing an ad if nobody clicks on it? This isn't like traditional display advertising that's sold on a CPM basis. When it's at a CPC, you need that propensity to drive the click to generate revenue, and the better you are at predicting what somebody will click on and ultimately lead to a purchase makes you more efficient in your advertising stack. And your goal is to show one ad and get one click, and that's 100% yield. That'll never-- It sort of doesn't happen 

[00:06:49] Tom Limongello: Yeah, that never happens, but that's okay. 

[00:06:51] Scott Messer: but that's what these people should be striving for, to the point where maybe the, the merchant team is complaining that you're not selling enough [00:07:00] organic stuff because you're so good at predicting what's sponsored. 

[00:07:03] Scott Messer: But I bet you if that happens, you're making a lot of money, and that's a champagne problem. 

[00:07:08] Ad Tech Home Run Theme 

[00:07:08] Tom Limongello: Well, I, I think I finally figured out what it is about this episode that made me, wear my Schwarbaum shirt. Um, and it's because I'm trying to see can retailers hit the ad tech home run? Can they predict what behavior's gonna be? Not just can they measure outcomes, but can they predict them? 

[00:07:26] Tom Limongello: So I think that's what, uh, that's what this episode's about, and that's what the trend is, I think, going into the second half of this year. Um, so I think that's where we are. 

[00:07:36] Scott Messer: Great. Let's take a listen  

[00:07:40] Meet Jon Flugstad MetaRouter 

[00:07:40] Tom Limongello: Hello. We are here with Jon Flugstad of MetaRouter. Um, he just got back from Cannes and so did we, and heard that you had some travel difficulties 

[00:07:52] jon flugstad: Yeah. I actually got back from Cannes a little bit late because I, I got there a little late. I had, uh, an opportunity to spend [00:08:00] five extra days in your fair city of Philadelphia, Tom, uh, after losing one of my family's passports along the way. So, you know, added trip. The kids thought of it as extra trip, uh, to see all the historical sites in Philadelphia, but all good. 

[00:08:12] jon flugstad: Stronger for it,  

[00:08:13] Tom Limongello: It's a good year for you to be doing it. It is the 250. There-- it-- we, we've cleaned up, uh, we opened up, uh, the fir-first bank in the United States for you. All the... There were lots of things to see. I hope you saw some of that  

[00:08:23] jon flugstad: We did. And World Cup timing, it was very, a lot of energy and so all, all good. Uh, we, you know, we got through five days without our suitcases in Philadelphia with our kids, but, you know, the, the, these things make you stronger, so  

[00:08:34] Tom Limongello: Amazing. Yeah, hopefully you brought home some Wawa gear or something. We were excited to have you back on the podcast. This is your second time with us. Um, previously you talked to us about how a sponsored search ad server is much more than just something that hoovers up keywords. 

[00:08:50] Tom Limongello: It's something that handles behavioral, data in new and interesting ways. And it was a really interesting conversation. For anybody who hasn't listened to it, I think, um, you [00:09:00] should go back to it.  

[00:09:01] LiveRamp and Rented Identity 

[00:09:03] Tom Limongello: Um, but we're gonna jump off from there and talk about where you are now at Metarouter we're in an interesting time. 

[00:09:09] Tom Limongello: LiveRamp was just acquired by, Publicis Epsilon, and I wanted to hear more about your thoughts, and you've been talking about this on your LinkedIn, about how rented identity is no longer a good idea. What does the end of rented identity mean for you? 

[00:09:29] jon flugstad: Yeah. Well, thanks for having me on again. I can't believe after the first time you're going through this, uh, again, so ap- appreciate it. I don't know if you guys are gluttons for punishment or, or, or, or what. But, um, yep, I lead the commercial team at MetaRouter. Been at MetaRouter about seven months, at Moloco before that, McKinsey before that. for those of you, especially, you know, you guys, you've been around commerce media quite some time now, kind of in the early innings of, of launching retail media networks, there really wasn't a viable alternative. It was like, you know, yes, the, the setup was pretty similar across retail media networks. 

[00:09:58] jon flugstad: I use LiveRamp. I'm gonna use [00:10:00] them for onboarding. I'll use them for some degree of measurement. Um, you know, maybe I use them for some degree of data collaboration, maybe. Um, and then, you know, the similar set of DSPs and similar set of on-site, you know, ad servers. And the setups were pretty similar, and they didn't change for, for quite some time. 

[00:10:15] jon flugstad: And what's happened is, like, these retail media networks have, have scaled and that now, now they have, like, reasonable volume flowing through their pipes. And when you do that and you have, what do you call it, what we kinda call rented identity, that becomes, like, a real limitation both to drive financial and operational leverage for the retail media network, but also just in terms of capability, especially kinda now as we move into an AI era where things need to be more, you know, near real time, more acted on behavioral signals, uh, more, more contextual. 

[00:10:46] jon flugstad: We'll get into that word a little bit more. And so this, this concept of I have, you know, a customer that I know, I'm gonna ship all that data to a third party, I'm gonna use their lookup table to be [00:11:00] able to match it to the IDs that I need, in that case it's the Ramp ID, and then that Ramp ID will map to the different channels that I'm working with. 

[00:11:06] jon flugstad: This made sense, and it was kinda the only path for a long time, particularly when there were a lot more meaningful publishers out there. Now it's like, you know, there's only so many channels that matter quite so much. They all have kinda some direct sync capabilities, and that kinda limitation on timing and dollars mean that it's a new discussion point across many retail media networks that I, I can't, I can't rent this anymore, and I can't have the dependency on this one third party, which all kinda came into relief as soon as they were, you know, when they were acquired by Publicis as well. 

[00:11:34] jon flugstad: It's like now there's, you know, even, even-- it's not, not Switzerland anymore. It looks like something else, so can I rent this anymore has kinda become the strategic question, uh, du jour, I think.  

[00:11:43]  

[00:11:43] Customer Context Explained 

[00:11:43] Tom Limongello: I think, um, what this new world that you're talking about where you don't rent the identity, it offers up some new opportunities, and I've heard a buzzword, um, customer context that's come up from that. And I wanna understand what that [00:12:00] is, because I think if you think of the old ad tech world of context, contextual just meant what was on the page. 

[00:12:05] Tom Limongello: Um, can you explain sort of what you guys mean when you talk about, uh, customer context? 

[00:12:09] jon flugstad: Sure. Yeah. 

[00:12:10] jon flugstad: What do I need to know about my consumers? Let's take the context of retail media. Like my on-site ads need, uh, need data or context to make the right decisions over what ads to show to that individual. Moreover, I actually have signals that I can share with off-site platforms that drive their algorithms in a more effective way. 

[00:12:30] jon flugstad: And what-- So what is this context? It's not so much like the contextual page or contextual ad targeting that you're thinking of in the past. It's related to an individual user. It's I have the consent for that user, I know that what I, what I'm able to do with them or what I can track for them. It's I have the identity for this user, and by the way, I have this identity not just in my own environment, but I have the id- identifiers across like third-party marketing platforms, and I have the behavior. 

[00:12:55] jon flugstad: Because what's becoming more and more the case is what happens next, [00:13:00] like the action to take, is based on a sequence of events or activities that are called context. And this is like, the kinda crown jewels of what commerce and retail media networks own. Like especially in the world of AI, where, you know, they're not gonna make the infrastructure investments to compete kind of on a frontier model perspective. 

[00:13:18] jon flugstad: They have the data or the gasoline, the context that can help make decisions in real time to serve kinda toward commercial outcomes for customers. And so, you know, at the, at the end of the day, like that's kinda what we're focused on, is like, you know, you ha- you... 

[00:13:33] jon flugstad: It's, it's, it's boring. It's kind of infrastructure, it's data, it's glue, but it's how do you take events in near real time, so, you know, very, very low latency, with identity appended to it, that can be used kind of across any use case, whether it's ads or something else. But, you know, creating that kind of contextual infrastructure that helps power that decisioning capability. 

[00:13:55] jon flugstad: That's kind of what we're thinking about and the way we think about context  

[00:13:57] jon flugstad: in general.  

[00:13:58] Scott Messer: It's a different way of [00:14:00] thinking about context. When you come from a more, like, traditional publisher standpoint, context was very immediate of exactly like what's on the page, um, and, you know, what article are we reading, and then later that would turn into an audience behavior, right? 

[00:14:16] Scott Messer: It was people who were interested in sports. If you weren't on the sports page, that was the moment of the context. Later you become an audience who is people that are interested in sports. How does... How do you think about the separation between, like, what would be the consumer journey being important to the decisions that are being made versus, like, this exact moment and the, and the exact elements of this static moment in time being the context? 

[00:14:43] Scott Messer: Is it about their history or is it about exactly where they are? 

[00:14:49] Real Time Scoring and AI Journeys 

[00:14:49] jon flugstad: Both matter. So to your point, Scott, it's not against one versus the other. It's basically like taking all the context you had before and introducing the real-time behavioral context on top of that, and now this is like what some of [00:15:00] the models can actually handle around inference. They, they couldn't really do that before. 

[00:15:03] jon flugstad: It was all batch, right? You'd have like all these... You'd have like a crawled, you know, set of context from a page, and you'd have longitudinal customer data, and you'd have like a kind of batch approach like, you know, offline, uh, you know, scoring that would create propensities or something like that that would live in this table that, you know, you assign it to a customer segment or an individual customer. 

[00:15:26] jon flugstad: That's all changing. The, the-- that calculation can actually be done on the fly in like, you know, 20, 30, 40 milliseconds us- using all that data together. If you think of like, say how, you know, ChatGPT works, like we're chatting with it, like you give it a prompt, it's read everything, it'll give you the next words in a row sequentially that like make you give you a coherent answer. The same thing is true with consumer behaviors or actions. 

[00:15:48] jon flugstad: Like you can train it on what, ad to serve it next, what item to serve it next, the price to give it next toward the outcome you want. And you've seen millions of journeys that look just like that one that have the behavioral elements [00:16:00] of it. And so, you know, having read all those journeys then, you have, um, the ability to like score it and know what to serve it, and it takes into the, into account the context that you talked about. 

[00:16:09] jon flugstad: There's-- So it's a, it's almost like an, an additional context layer that you can use on top of what you thought about it in the past, but it just the real differentiator is this like kind of real-time decisioning and real-time scoring, bespoke predictions for that individual at that exact moment. And that's what's really powerful right now. 

[00:16:29] Real Time Scoring Wins 

[00:16:29] jon flugstad: And if you're not doing that with, you know, with ads, like you're, you're, you're behind, right?  

[00:16:33] Scott Messer: Yeah, it's a great approach to it. And, on the targeting side of media, that's always been a challenge of how fast can you process these, and should you be using like pre-populated segments or really just letting the machines discover the segments and, and then create targeting. 

[00:16:48] Scott Messer: I'm still gonna push on the fact that you're calling past audience behavior context, which I will allow as an English major, but I think as an ad tech practitioner, I have to say [00:17:00] that, that, that context is the exact moment in time, but you are pulling in audience behavior and customer journey in, which sounds like a reasonable thing to do for a company like MetaRouter that is building an identity spine and is using audience behaviors, right? 

[00:17:18] Scott Messer: You need a way to actually remember what an individual user did. So I think we agree here, John, but I'm still gonna always say that context is context and audience behavior is audience behavior 

[00:17:32] jon flugstad: Yeah. Yeah, that's fair. And the, I mean, the context becomes the model weights, right? The outcomes are dramatic. I mean, if you are using a real-time scoring per onsite, we did some tests when we were at Moloco, like, you know, if you did the real-time scoring that used the browse events for that individual session to, do bespoke predictions, the performance improvements were like 50% on the ads. 

[00:17:55] jon flugstad: Like it wasn't like 10%, it was like 50% on the ads based on them being better targeted. [00:18:00] Similarly, like, you know, a customer we're working with where that event stream is powering inference for their conversion events. We, like, we improved their conversion rate by 11%. So again, it's like in the world of like e-com, you know, a 2% improvement is like massive. 

[00:18:17] jon flugstad: Like 11% is like, holy crap. Like this is something like fundamentally different. So it turns out this like new form of context, it matters a lot and like it's, you know, from, from ground up, it's just like, it's the infrastructure plus the context together that lets you like, make decisions that, you know, drive the commercial outcomes in a really fundamentally kind of different way. 

[00:18:36] Scott Messer: And how long does it take, before the machine starts like understanding that? Like how many journeys do you really need to see before it has like meaningful context? 

[00:18:50] jon flugstad: It'll depend on the tool and the volume for sure. I mean, it's a... Again, I'm not a machine learning engineer. I pretend, I pretend to, to play one on podcasts, I guess. Um, but you [00:19:00] know, uh, you can see real value in two, two, three weeks training the models, like observing customer behavior. Like that's the, gonna be the most, you know, the, the most relevant set of journeys that help you predict things and then, you know, kind of change the policies of what you serve on a site, um, or change the way you're scoring predictions for click-through on an ad or something like that. 

[00:19:18] jon flugstad: So pretty fast. Pretty  

[00:19:20] Scott Messer: and do you see-- And since you're not a machine learning engineer, I'll just ask you more questions about it. Do you see, do you see customers like tuning the, the look back period? And like, is the, is, is there anything about the relevance of like how far back you're looking and what actions were taken sooner rather than later? 

[00:19:38] jon flugstad: Yeah, that's a good question. I, I don't know. I mean, I think it depends on the, like, the type of retail, the type of site, like how long buying cycles are, degrees of seasonality. I think it's a business choice. Um, so it's a g- reasonable question though, and probably something that you could, you know, like if you're con- consistently testing, like, you know, testing how long your look-back window is for, you know, how long you're training your [00:20:00] models or, you know, those kinds of things. 

[00:20:01] jon flugstad: But, you know, that's, that's a question for the actual machine learning engineers  

[00:20:06] Scott Messer: We'll have them on next 

[00:20:08] Tom Limongello: yeah.  

[00:20:09] Wh ole Page Optimization Debate 

[00:20:09] Tom Limongello: I w-- I wanted to, um, steer the ship in, in a direction where you talk a little bit about what I see happening in the overall marketplace. And, you know, you mentioned your previous history at Moloco and how, you know, you were setting up the ability to improve the site performance or potentially the, you know, the ad performance. 

[00:20:30] Tom Limongello: But now we're seeing a lot of companies out there, uh, specifically I'm gonna play this clip from Criteo at Cannes, um, where Ari Paparo on "Marketecture," uh, talks to Michael Komasinski and says, you know, like, "What are, what are the things that you're doing in the retail space?" And what he says is, is, you know- 

[00:20:51] Scott Messer: [00:21:00] We can't hear it if you're playing it  

[00:21:24] Tom Limongello: So what we just heard was Michael Komasinski saying that they are looking to, uh, manage not only the ads, but also the search relevance on the page. Understanding, you know, we've heard of this from the Pentaleap episode that we had, um, fluidity of how many ads you put on the page, uh, is important. But I just fundamentally want to understand, do you see the architecture of this, should the DSPs be controlling everything that's happening on the page? 

[00:21:53] Tom Limongello: Or, you know, like, should supply and demand be managed all through one system now? Like, what are you seeing as sort of the [00:22:00] way that, way forward in this world of commerce media? 

[00:22:04] jon flugstad: How advanced was that system of what the ad server did to look up which ads to show? Not advanced. The like retail media  

[00:22:10] jon flugstad: networks of the world were lagging real pioneers in search by a decade, right? Google invented transformers in 2016. It was applying it to search pretty quickly thereafter. 

[00:22:21] jon flugstad: And, you know, it was like then it was operating on consumer intent versus keywords, and those were deprecated like a long, long time ago, and still retailers were kind of using those kinds of signals. So first and foremost, even if it's still siloed, even if you don't have like all the alignment required across the different teams in a retailer to say, "We're gonna do whole page optimization," there is a lot you can do just in the interim to improve the, the ad results that come back even into those punch slots on the page. 

[00:22:47] jon flugstad: I actually agree with Like ads should be supportive to a whole consumer outcome because the goals are the same from the advertiser and the retailer. They're trying to drive sales. And so what drives sales overall is like the right [00:23:00] ad load, the right products shown overall, um, and the like the way that you actually blend those things together though is Uh, you know, k-kind of an art. 

[00:23:10] jon flugstad: Like, it's not that you should just have, like, one model that scores organic, and then you look up what's being campaigned, and it turns those ones into paid. They're very different. Like the-- So you need a system that's smart enough to know how to run ML on ads and how to do organic because their goals are different. 

[00:23:27] jon flugstad: One is clicks, the other is ad monetization and has constraints around budgets and pacing and, uh, like, you know, the, the-- it's... They're not the same problem to solve. So that said, like, should folks come together toward a whole page optimization? Absolutely. Is it like one system like what Pentaleap does to like take organic results and flip them into paid? 

[00:23:46] jon flugstad: Probably not. It's, you know, like n-no advanced ad system does that. Like Google has separate ML for its paid versus, uh, organic. Meta has separate ML for its paid ads versus what ends up organic because they're different [00:24:00] problems and, and they're-- they train differently and they think differently. And so, you know, if whole page optimization means unifying around like one model, it probably means you're leaving monetization on the table at the end of the day. 

[00:24:13] jon flugstad: But if you, if you are able to combine sort of different approaches in ranking and then have something that says this is the right ad load for a consumer experience, it's a good idea overall. So there's-- Uh, hopefully, that's not like too nuanced or, but, you know. 

[00:24:26] Scott Messer: Oh, we are, we are the podcast of nuance, so you're, you're in the right place here. And congratulations on saying what, what is the, the right ad and saying right almost three times without saying right place, right time, right person. 

[00:24:39] Tom Limongello: Yes. Thank you. 

[00:24:39] Scott Messer: But what I  

[00:24:42] jon flugstad: Just narrowly hope  

[00:24:43] Scott Messer: though is it-- this seems like it works very well when a retailer is running their own ads and is in charge of it to be able to match the ad with the context through the program. 

[00:24:57] DSPs and Programmatic Risks 

[00:24:57] Scott Messer: And what I wanna take a moment [00:25:00] to think about is how does this extend out into a DSP universe where external buyers are bidding into these, let's just call them walled gardens effectively? And, um, does this, does... I, I don't think the context, the raw context translates out very well. Like, you're on a page for Pepsi, like who knows what that exactly means. 

[00:25:24] Scott Messer: Um, and all of, you know, how do the... I wanna ask it both from the MetaRouter perspective, but maybe first from the broader perspective of like what is going out to the DSPs right now from retail pages that are connected to them? And then how do ads come back and sure, how does MetaRouter solve that problem, if it does? 

[00:25:46] jon flugstad: Yeah. Let me make sure I understand the question. So there are c- there's, there's the DSPs, right? Which are helping you buy kind of on the open web, And if it's in the retail media context, they're using audience, you know, audience data to do it. There's also the, like, walled gardens or direct marketing channels that [00:26:00] are making decisions in their own environments using some of your data as well. 

[00:26:03] jon flugstad: Is it kind of thinking through both of  

[00:26:05] Scott Messer: yeah, I wanna talk specifically about DSPs that are trying to buy into retail media networks. 

[00:26:11] jon flugstad: Ah, okay  

[00:26:12] Scott Messer: like it could be Skai, it could be PacVue, it could be The Trade Desk, Koddi and Gopuff that was announced previously. 'Cause th- that's where I think there's a massive opportunity for DSPs is to be able to untap all of this. 

[00:26:25] Scott Messer: And, uh, I also think it's a treacherous and slippery slope for retail media networks to head into a programmatic universe. Um, but when my li- my life's work is answering the question, why shouldn't we just buy you in the open auction? And data is always the number one answer. So if retailers can protect their data and it's structurally difficult to put out to DSPs, then DSPs don't have a strong right to win for why you should use a DSP to buy rather than going through Target's front door, Best Buy's front door to place [00:27:00] campaigns directly with them  

[00:27:02] jon flugstad: Yeah. Yeah, So those DSPs are kinda buying doors you mentioned. They, they, in my understanding, they, some of them do different things. So for example, like Skai, like it's not exactly a DSP 'cause there's no decision. It's like a-- It connects to ad servers and basically submits bids and budgets to an ad server directly without doing any like decisioning in the traditional way you'd think of a DSP. 

[00:27:28] jon flugstad: So like the traditional way you think of a DSP is like The Trade Desk or DV360, where there's all this inventory, and it's making decisions on y- you know, to, to, on where to target based on the data you've provided it. That's not really what Skai is doing. Skai is like integrated directly into Criteo or integrated directly into Moloco or Pentaleap or Kevel, all of them. 

[00:27:50] jon flugstad: And what it does is it just submits via API the, the bids and budgets that that ad server will optimize all on its own. And so, you know, that's like it kind of solves the [00:28:00] fragmentation problem of like, I wanna buy in lots of different places. It is not deciding across media networks at all. It's basically just an instance of the ad server for that individual, uh, retail media network that allows you to do it in one spot and kinda buy in one place. 

[00:28:15] jon flugstad: So it'll give you lots of insights or like keyword recommendations, things like that but it's not decisioning. So it, it's actually a pretty fundamental difference on like what... Now I, I don't know as well like the Koddi plus Trade Desk setup and what they're enabling. I don't know if it's like the same thing where it's like you basically get like an ability to submit campaigns via The Trade Desk into Koddi for Gopuff. 

[00:28:37] jon flugstad: I don't know it as well, but the folks who control the data, who are the retailers, have been pretty hesitant to say, "I want my inventory floated programmatically across a bunch of other media networks." 'Cause, you know, it's like kind of a prisoner's dilemma. You end up kind of in a suboptimal norm, and they've all had-- by the way, had prior to like really closing things down on their sites, [00:29:00] they had GAM running open. 

[00:29:02] jon flugstad: You know, they were getting 

[00:29:03] Tom Limongello: That's 

[00:29:03] jon flugstad: $2.50 CPMs, right? It sucked. Like it was like, you know, too, way too low. And so that's the experience that they're kinda coming from. 

[00:29:11] Scott Messer: Yeah  

[00:29:11] Tom Limongello: of what I got dropped into in 2018 was, "Okay, here's Albertsons', you know, 17 banners. Um, can you make this site work better?" And we're running with GAM. And it wasn't until sponsored search became the norm that the, that those sites started to really make money. But going back to GAM didn't really make sense either. 

[00:29:31] Tom Limongello: Um, and so when you, when you see news like Koddi plus goPuff plus The Trade Desk, people like me are like, "Oh my God, they let the fox in the henhouse." Like, what, what are we gonna do here? 

[00:29:44] Scott Messer: yeah, I, I think that d- if I could speak on behalf of those two companies, I think that's more of like a programmatic guaranteed type scenario, similar to what Skai was doing in your example, where they're pushing a campaign into an ad server rather than [00:30:00] having a more traditional DSP, SSP relationship. 

[00:30:04] Scott Messer: But the nice thing is for The Trade Desk in that scenario, you're still running your media through a programmatic pipe, and it's getting reporting back. So when you wanna lay your performance next to other elements that The Trade Desk is already measuring, that is happening. And potentially, I guess you're also running The Trade Desk, uh, creative and the wrapped pixels that go with it, so you're getting a different type of measurement potentially that's coming out of it. 

[00:30:30] Scott Messer: But certainly, retailers, take this word here, do not let your data get floated across into the open auction, or you will be commoditized further than you have ever seen anything before. Let me just 

[00:30:47] Scaling Offsite with Platforms 

[00:30:47] Tom Limongello: that's where, I think where we were initially gonna go, and I think, John, you were thinking that we were asking this question, and this is what I wanna ask as a follow-up is flip it around.  

[00:30:56] Scaling Offsite Activation 

[00:30:56] Tom Limongello: And now, you know, we've had Mark Williamson on the podcast talking about how he's moved, uh, Costco into a Google universe where they're running, you know, the normal SEM ads there. 

[00:31:07] Tom Limongello: Um, we just had a discussion with Christine Foster from Kroger about how they are moving, uh, their audiences to be, you know, usable in the TikTok platform. They've already done it with Instagram. And leveraging your technology, why are they now excited about moving outside of their own inventory, um, in a way that they weren't before? 

[00:31:31] Tom Limongello: And sort of what allowed that to happen? 

[00:31:36] jon flugstad: Yeah. Yeah. And it's interest-- I mean, I like some, some retailers went that way much earlier. You know, like they, they-- think about Dollar General. They, you know, they, they built a pretty large media network, you know, in the scope of hundreds of millions, mostly offsite. That's kinda where they went first because they have, you know, 20,000 stores. 

[00:31:56] jon flugstad: They have very little e-commerce. They kinda had to, they had to do it that way. And so, [00:32:00] you know, like whereas the other playbook, the one that's like the starting playbook for most retail media networks has been onsite, you know. Then when growth gets hard, go, you know, prioritize offsite more and try to do both in parallel. 

[00:32:12] jon flugstad: I think there's, I think there's a lot of opportunity in both still. You've probably heard that like on onsite, like there's so much low-hanging fruit still to change to the in-- I think Andrew Lipsman talks about this a lot, like inventory levels are still really low. the technology powering most of it's still pretty low. 

[00:32:25] jon flugstad: So it's, it's not like it's or, it's like and. And the mandate is growth. Like it's, you know, it's gotten, quote unquote, "harder" for growth. You hear people say easy dollars are over, uh, blah, like, you know, maybe. But so I think it's more like, okay, in order to grow, I need to prosecute both of these paths in earnest. 

[00:32:43] jon flugstad: And what has been my gap to basically scaling offsite in a major way, it's been that all the campaigns are managed service. And so that's like, that's really hard. You know, you need like large teams planning and selling, you know, campaigns that [00:33:00] don't really scale. And so if you think about like what Kroger is super smart about is like they are picking distinct platforms where they wanna scale. 

[00:33:08] jon flugstad: They're setting up direct connections for their audiences so it's efficient for them, and they're enabling brands to be able to buy the way they wanna buy so they're not in the way. There's not like a, you know, a, a, a contract, an insertion order in the way. There's not like a team planning it on their side where they need to have hundreds of planners. 

[00:33:26] jon flugstad: Like, so I think they're being really smart, to be honest, to be able to, you know, control what data they share, to own kind of the match table and identifiers that go to those pro- those environments, and then being flexible in the way they activate, in the way that brands want. Like it's very collaborative. 

[00:33:43] jon flugstad: So, you know, I, I'm, I'm really like, honestly like im- uh, impressed and inspired by what, you know, folks like Mark and people like Christine are, are, are doing. I think they're, they're kinda changing the game and, you know, humbly getting out of the way of what brands wanna do in a way with their [00:34:00] audiences, so. 

[00:34:01] Tom Limongello: I think for the listeners, it would be interesting to know, and we sort of talked about this in the beginning with customer context being faster and, you know, having your own data spine, um, that to me, and this is sort of what I've been seeing over the last, say, six months this year, uh, the first half of this year basically has been it's okay now to let the ad channel that I experience the most, Instagram, uh, actually play a part in retail media. 

[00:34:30] Tom Limongello: Um, I think for a while we're always like, "Oh, we gotta go upper funnel." And, you know, CTV maybe, you know, at some point we'll, we'll have really good examples of that. But I think the be- best examples that are happening now are Instagram, TikTok, um, discovery happening, and now the retailer being able to use that data. 

[00:34:48] Tom Limongello: And I think there was something missing in terms of how they could use it before, potentially the speed. Um, I mean, what, what are you seeing from that perspective? Like you-- what are your customers talking about that they can do [00:35:00]  

[00:35:00] jon flugstad: Yeah, Yeah. I mean, the, the, um, h- how do you grow offsite? Simple question. Like, you either need higher velocity of campaigns, you either need more audience data, or you need to be able to charge a higher price for it, and, and, or the, you know, the, the results need to be good enough that justify more spend. 

[00:35:20] jon flugstad: Like there's, there's only so many levers for it. And so the things that retailers are doing to try to capture that are, number one, data connections with these providers that help improve like their match scores or match rates so that their audiences are more highly addressable on these platforms. 

[00:35:37] jon flugstad: Number two, thinking about signal quality and volume. Like what are you sending them? Is it just that, you know, they're pixeled on your page and you're sending them all the behaviors, or are you actually controlling the data you s- you know, the data that you've collected in a first-party environment? 

[00:35:51] jon flugstad: You're sending it server side so it's not blocked by ad blockers. You're incorporating in-store sales and i- identity matched related in-store sales [00:36:00] into your feeds to them. These are the kind of signals that now these companies are providing to, you know, Meta, to Google, et cetera. And it's improving the, the match scores and fidelity of the audience and the signal quality for those algorithms to basically do what they wanna do, which is target, you know, a customer or find lookalike audiences and get them to click to come to your environment. 

[00:36:23] jon flugstad: And so the better you can feed them, uh 

[00:36:26] jon flugstad: with, with that in, you know, in a s- you know, privacy controlled way, in a way that's consented, but the better you can feed them, the better they can operate. Moreover, the, the, you know, the more you can sort of get out of the way of running campaigns on behalf of your advertisers, the faster the velocity. 

[00:36:42] jon flugstad: So I think that's the kinda mindset is like, you know, bigger audience, better signals, easier to buy. And if you kind of watch those principles in nature, like you're gonna be able to predict a lot of the movements that what retail media networks are gonna do in general. Is that 

[00:36:56] Tom Limongello: Yeah, I 

[00:36:56] jon flugstad: conversation with you guys? 

[00:36:57] jon flugstad: Or, you know, would you push back?  

[00:36:59] Propensity Versus Determinism 

[00:36:59] Tom Limongello: Well, you [00:37:00] know, it's funny. When you have conversations about this, where people get a little bit stuck is, okay, there's the retailer data that originally lived in audiences, and those audiences were brand buyers, you know, uh, shoppers of a certain category, things like that. 

[00:37:15] Tom Limongello: You're now moving, as we talked about earlier in this discussion, to more propensities. And if you talk to old-school retail media people, they're like: Oh, that's just propensity. That's not typical, like, retail media, and it's getting into behavioral, sort of what, like, what Scott was talking about before. Um, and so is there a, a, a risk here where you're seen as not as deterministic as you need to be? 

[00:37:39] Tom Limongello: I think that's the only thing that comes up when I, when I have these discussions, it's like, "Oh, propensity is interesting, but it's not, it's not as real as, you know, just a brand buyer." And like, like how do you deal with that discussion? 

[00:37:53] jon flugstad: Yeah, that's a good question. And by propensity, do you mean like a, like a probabilistic audience match? Is that kinda what you mean? Because  

[00:37:59] Tom Limongello: I think [00:38:00] that's sort of what-- when, when the word propensity is used, I think that, that it's thought of as, as probabilistic. Is that really what's going on with a, when a customer uses MetaRouter if they have their own data spine now? 

[00:38:11] MetaRouter Origin Story 

[00:38:11] jon flugstad: No, it's deterministic all the way. So, I mean, the way that MetaRouter works, which is... I mean, I don't know if you guys have kinda heard the MetaRouter story. So the founding customer is the Home Depot. They had a third party pixel on their page take down their entire e-commerce site for like 24 hours. They like missed all the revenue. 

[00:38:28] jon flugstad: So,  

[00:38:28] Tom Limongello: I haven't heard this story. This is good 

[00:38:30] jon flugstad: burning platform, right? It's like We missed all this e-commerce revenue, you know, these third party tags be damned. We're determined to get them all off our page, you know, in some fixed amount of time, 10 months, whatever it is. Like we're gonna figure this out. Anyone done this before? No. So the guys who were, uh, who founded MetaRouter, they worked at a company called Astronomer. 

[00:38:49] jon flugstad: Oh, you, you know that name 

[00:38:50] Tom Limongello: We only know of that because of, uh, the Coldplay concert. 

[00:38:55] jon flugstad: Yeah, exactly. So notorious, but, you know, pretty good company actually. But, you know, that, that was a, [00:39:00] a poor nega- negative outcome for them. Um, but the guys who founded MetaRouter were part of Astronomer. They worked with the Home Depot. 

[00:39:07] jon flugstad: They said, "Hey, can you solve this problem for us?" They said yes, and so they, like, were trying to figure out how to get these tags off the page. Like, how do, how do I do that? They read a paper that Nikhil and the team at Walmart had written for, uh, WMX back in the day because they were anonymous. They had no loyalty program. 

[00:39:24] jon flugstad: They were trying to identify people who were anonymous, you know, to use them as audiences and deterministically, you know, use them for measurement based on trying to build an ads business out of loyalty program. So these guys founded the white paper, built MetaRouter with a bunch of additional features, and what that meant was that, number one, they could accomplish the mission of getting all the pixels off the page. 

[00:39:45] jon flugstad: There's control and compliance of the data. You're not just sending it directly to the partners. You can decide server side how you wanna send it. But number two, this piece around identity is deterministic, where if there's anonymous user on your site or if they're known, you can n- you know, whether they're known or anonymous, what happens [00:40:00] directly from the browser is it communicates with each of the marketing partners and pulls back their identifier into the browser, and then for every event that happens in the browser, it gets streamed into the cloud of the customer appended with all those identifiers. 

[00:40:14] jon flugstad: So, you know, like an add to, add to cart event happens, that event gets streamed into the cloud. It has the Pinterest ID, it has the TikTok ID, it has the, the, you know, the T- The Trade Desk ID, it has the, the Meta ID appended to it. So then all of a sudden you've got this person that you didn't know, but they're addressable because you have their IDs on, you know, in your environment for direct activation in those marketing partners. 

[00:40:39] jon flugstad: Or if I know them, if I have PII on them, right, they're part of loyalty or part of an existing kinda group I know, you just enrich that profile and you have all these... So that's, that's the thing that they built and I think kind of  

[00:40:49] Tom Limongello: Yeah, it's sort of like, uh, medieval her-heraldry. Instead of telling you your name, you got all your badges and your, your, your insignias. 

[00:40:56] jon flugstad: Yeah, yeah, yeah. Like w- all these activation pathways for this [00:41:00] individual, but now it's like a key ring that lives inside the environment. Like, that's, that's the difference, and that's, I think, what was so exciting about, like, coming, coming to MetaRouter. It's like, holy crap, this is like a viable alternative to be able to, a- you know, own the identifiers on your own without just shipping them out to this, like, third party and just go directly to those channels that matter, matter the most. 

[00:41:19] jon flugstad: And so that's kinda what they built, and that's the, the way, you know, we, we kinda think about is like own it in your cloud, like have it. They use your, your identity graph to do whatever you wanna do with, right?  

[00:41:29] Tom Limongello: yeah,  

[00:41:30] Wrap Up and Next Stops 

[00:41:30] Tom Limongello: I think, for anybody who has listened this far, um, and hasn't seen the original episode with Nikhil, uh, from MetaRouter, we should definitely do that because, um, I think what, what that brings up for me, we just actually had a great conversation with the IAB, uh, Colin Colburn, and he's been putting out research about how retailers can't all be Amazon, and I think it's a great thing to say. 

[00:41:49] Tom Limongello: But what I'm hearing a lot is that a lot of retailers could benefit by being a little bit more like Walmart, which was also part of the story of, of Nikhil's Walmart Labs', history and [00:42:00] now, you know, as part of MetaRouter bringing this idea of identifying people with their behavior more so than, uh, identifying them on their PII. 

[00:42:09] Tom Limongello: But it's very hard for somebody who comes from the retail media world, which was-- it was all about the PII discussion. And so I think, I think that that, that is a tension that will remain, um, as we, as we move into this place. If retail media wants to be all media or commerce media is saying that it's a, it's this bear hug ar-around the entire media industry, you now have these two things that can... 

[00:42:29] Tom Limongello: The context of, you know, what, what is this person doing right now versus what have they done historically? That tension is not resolved and, it'll be interesting to see how does sellers deal with those, those types of discussions with the client? Um, we wanna hear more about that, so. 

[00:42:43] Tom Limongello: But wanted to say thank you. Um, we went a little long here, but this is, this is-- We expected that because you, uh, have a lot to say. So thank you for joining "The Middlemen," and, uh, where will we see you next? 

[00:42:57] jon flugstad: Oh, I don't know. Um, [00:43:00] Ascendant Adweek, probably a few things in the fall. Hopefully you guys will have some overlap at some of those places. Um, but yeah, otherwise really appreciate you guys having me on. ​