# S2 E27 T47 Moloco - Jon Flugstad

S2 E27 T47 Moloco - Jon Flugstad 

[00:00:00] Introduction to the Middlemen Podcast 

[00:00:00] ​ 

[00:00:00]  

[00:00:06] Tom: You are listening to the Middlemen Podcast with me, Tom Limongello, and my co-host Scott Messer . careers have put us in the middle between retail media and brands, but retail media is a closed loop. So we're here to open up the channel and let you listen in, join our discussions with the people who make retail media happen. 

[00:00:25] Speaker: You may be wondering why I'm sitting in front of a Telly, whoever it was at. Telly that. Put us to the top of the list. Thank you. The middleman has a a nice new prop and we're gonna feature this, , fairly often, especially when we have video clips. And boy do we have a video clip today to show you. Um, if you were at Marketecture Live, you saw Terry Kawaja, being interviewed by Ari Paparo. 

[00:00:47] Marketecture Live and Moloco 

[00:00:47] Speaker: And Ari asked him, which company should the trade desk buy? And after a long answer, which we're gonna show in a minute. He said Moloco just so happens that we interviewed Jon Flugstad of Moloco this week, [00:01:00] and this was great context. Terry is talking about how Moloco, is a great acquisition target and the conversation that we had with Jon was about how they're gonna be going it alone. 

[00:01:10] The Future of AI and Outcomes 

[00:01:10] Speaker: But all of it has to do with the future of ai. The future of the outcomes era. So have a listen and we're gonna get to the episode in a minute. 

[00:01:20] Terry Kawaja: what's the whole theme of this? Conference? Outcomes. Right? I believe AI is gonna accelerate the whole move to outcomes. 

[00:01:27] Terry Kawaja: I also think we need to reframe what we mean by outcomes. We do not necessarily mean, you know, direct response, bottom of the funnel. You know, final step before conversion. No, uh, outcomes is a full, as Lou Pascalis could tell you. It's a full funnel phenomena. Um, and I think that they. Uh, are missing a big opportunity in performance. 

[00:01:54] Terry Kawaja: So what company  

[00:01:55] Ari Paparo: should they buy? You get paid on a transaction, not on a word, right? 

[00:01:59] Terry Kawaja: [00:02:00] Moloco.  

[00:02:02] Ari Paparo: Moloco, all right. That is a, wow, that's a curve ball, but it, it resonates and, okay, now I want to hear why. 

[00:02:12] Tom: So what I love about what Terry k Quadra is saying here is that he's telling the trade desk to be a walled garden. And if you know the trade desk, that's really funny because they've been, they've made their whole identity, uh, as the anti walled garden. 

[00:02:30] Retail Media Networks and Performance 

[00:02:30] Tom: So for middlemen listeners, what I think is interesting is that. The retail media world, these retailers have been building walled gardens. They're mini walled gardens, but they're not walled gardens in the way that Facebook and, and Google are. They know the transaction, they have the consumer's data, but they don't really leverage behavioral data, and because of that, there's been this gap and it means that they don't drive as a, a performance as well as [00:03:00] say a Google or a Meta or an app lovin or potentially a moloco. 

[00:03:04] Tom: And so now you see Moloco moving into this retail media world powering, the onsite capability for Wayfair and their retail media network. I think what Terry is seeing here is like, wow, okay, if a real performance company that has that mobile DSP in their DNA, can go in there and and really power what the retail media network offering is, that could really open up. 

[00:03:31] Tom: , The full capability, not only the behavioral side, but also the first party data side, and that together could be the full funnel phenomena that he's talking about. And you're seeing it, the Trade Desk is moving into, with their partnership with Koddi and, and goPuff, uh, into the onsite world. 

[00:03:50] Tom: So it looks like the future of where retail media networks are going is that they need to become intelligent. They need to understand not just the transaction, but they need to be [00:04:00] able to use machine learning to understand the consumer journey. And that's what this discussion is gonna be about with Jon Flugstad from Moloco and my co-host Scott Messer. 

[00:04:11] Interview with Jon Flugstad from Moloco 

[00:04:11] Tom: Hey Scott, are you ready for another episode of The Middleman? 

[00:04:14] Scott Messer: You betcha I am Tom. Nice to see you. 

[00:04:17] Tom: So today, um, we're gonna be talking to Jon Flugstad , he's at a company called Moloco, but he started at, um, McKinsey under Quentin George. So one of the most, uh, interesting perspectives to the retail media world. When I had a chat with him before. I realized that both of us started in retail media in 2018, uh, me at Quotient him at McKinsey. 

[00:04:39] Tom: And it was really great to compare notes and see where things were back then and where they're going. So this was a pretty fun episode. And, for me, thinking back on that time 

[00:04:50] Tom: a lot of what we were trying to sell back then was like personalization. We were trying to basically make the creative somehow personalize to the the shopper, and it was totally wrong. [00:05:00] And the data that we had to use for that was really just transactions. We didn't have very much behaviorally about the consumer, and I think it was just kind of like the ad servers and the data infrastructure weren't really ready to handle consumer journeys. 

[00:05:15] Scott Messer: Yeah, they certainly weren't. Um, and we dive into this a little bit more with Jon of how the ad servers were really just dumb pipes and they were order takers and reservation systems and. Scaling, uh, sort of like a publisher monetization model where this all comes from, really required incremental salespeople, incremental, uh, systems managers, ad ops, support teams, and like, as he says, that's no way to scale a scaled business. 

[00:05:44] Scott Messer: Um, and then, you know, in steps, Moloco, where they come at it from a much different lens of. Um, enabling this personalization, really leaning into machine learning later into AI and how that [00:06:00] becomes the driving engine of growth and performance. Um, I really found it interesting when he said that, um, ad servers of old locked publishers into really bad habits of sales and maintaining, uh, guarantees and, um. 

[00:06:17] Scott Messer: Uh, early insertion order based reservations that prevented publishers from really driving outcomes or caring about outcomes in some ways. But that the new model of retail media has to focus on it. And that comes from the business relationships all the way down to, you know, the, if you don't have performance, the renewals don't come. 

[00:06:37] Scott Messer: It was neat. And I think the last piece that that ties into is where he was talking about self-serve. And that, um, he thinks retailers must really be leaning into self-serve, um, not just as a way to, scale budget because it is better than, you know, sort of what he calls the reservation based world, but also because it's the front end to a [00:07:00] performance engine. 

[00:07:01] Scott Messer: And if you think about he was comparing himself to Facebook and Meta and Google from a performance standpoint. Uh, and when you look at them like. You've logged into P max and you do those campaigns because they outperform and because they're really easy to scale for you. So when you see an answer or taking that position, I think it gets really interesting into sort of the scale they can achieve for their clients and the, how one of their clients can sort of open up their program without scaling headcount. 

[00:07:32] Tom: And I think from the arc of retail media, you know, what I had realized was from maybe 2018 to 20 22, 23, you had these turnkey ad servers that were there to sort of do everything, bring demand, do the campaign management and do the ad serving. And that was sort of seen as, okay, now it's time to bring everything in-house because we can potentially. 

[00:07:56] Evolution of Retail Media 

[00:07:56] Tom: Secure more margin. Um, what he's, what [00:08:00] Jon's gonna talk about in this episode kind of flips that and says, okay, the pendulum's swinging back and there's actually more work to do. You've got to spend nine figures on Google Cloud or wherever you're serving, uh, and hosting your data because there's a lot more that you can do now to understand the consumer journey. 

[00:08:21] Tom: I think we should just get into it and, and, and hear what he has to say. Um, yeah, there's, there's a lot here. 

[00:08:28] Scott Messer: Sounds great. Let's do it. 

[00:08:29] Tom: Welcome to the Middlemen Podcast. We are here with Jon Flugstad, who is head of Business Development and Commerce Media at Moloco. We've had a chance to talk with some of your teammates like Glen, um, at CES this year. And we had heard lots of really good things about, um, Moloco and their sponsored search capabilities, but when you reached out after the Ascendant network, um. Conference . You told me that you were actually expanding to, the rest of the [00:09:00] onsite suite of products. So thought it would make sense to, to get a chance to talk to you. But before we get into all of that, um, first of all, welcome and, um, tell us a little bit about yourself. 

[00:09:11] Tom: 'cause you have an interesting background. Um, you were McKinsey for a while, so tell us about how you got to where you are. 

[00:09:17] Moloco's Journey and AI Integration 

[00:09:17] Jon Flugstad: Perfect. Hey guys, thanks for having me on and uh, great, great to see you. Yeah. So, uh, I met MoCo. Uh, as you mentioned, we're an onsite ad server, at least that's one of our products. We also have a mobile DSP that's quite scaled. Um, and you know what? Been in this like commerce media space for, for a while now and ended up in it kind of by happenstance like, like many have who have been in the industry. 

[00:09:38] Jon Flugstad: Um, I joined Moloco about 18 months ago. Prior to that I was at McKinsey for a really long time, like 10 years. Um, you know, joined not knowing what to do with my life. Did kind of the early 

[00:09:48] Tom: look, you look good for 10 years at McKinsey. You're not, you're not, you don't, you don't look weathered. 

[00:09:52] Jon Flugstad: Yeah, it's the filters on these cameras, you know, they're amazing. It's like they take away all the grays and all the, all the wrinkles. Um, and, you know, kind of didn't know what to do [00:10:00] with my life, but ended up falling into a pocket of really great people. Kind of, you know, midway through the McKinsey tenure and working on some early retail media networks in kind of the 20 18, 20 19 timeframe when it was kind of being called retail media and folks were transitioning from these. 

[00:10:15] Jon Flugstad: Kind of tools and approaches, um, you know, the quotient of the world and the other sort of, 

[00:10:19] Tom: Yeah, that was my starting gun was 2018 too. So, yeah, that's, that, that's a, that was a fun time. Um, and it was also fun just because like Quotient didn't realize what we had and so they were like, let's rename it retail performance media or, and like I, now that I hear people are head of Commerce media and then Harvey Ma over on the CPG guys yesterday was saying it's retail experience media. 

[00:10:42] Tom: I'm like, you cannot kill this term. Retail media just doesn't, it's not gonna go away. 

[00:10:46] Jon Flugstad: Yeah. Just let it be the thing. Right. It's, it's fine. We all agree on that. Um, and then, you know, ended up working alongside, like I said, a great group of folks, uh, who were leading this practice at, at McKinsey, and, um, worked with a number of [00:11:00] retailers and an airline on kind of how to build these businesses. 

[00:11:03] Jon Flugstad: From the ground up. And so, you know, with when you've built like, you know, been involved in building like eight of these businesses, you have a kind of a unique perspective. You've been inside kinda the belly of these retailers. Know the team dynamics, know the tech, you know, tech challenges. Know kind of how to frame it from a strategy and capital planning perspective and what are the bottlenecks around, you know, ops around go to market, around kind of prod and eng approaches that exist within the organizations. 

[00:11:29] Jon Flugstad: And so it gives you kind of a unique survey on trying to build these businesses within a broader retailer, which is hard. Like I think, you know, the leaders of these businesses are kind of like heroes in our industry 'cause they're doing something really hard, like with aggressive targets. And, you know, you know, sometimes with one hand tied behind their back. 

[00:11:47] Ad Serving and Performance 

[00:11:47] Jon Flugstad: Um, but yeah, learned a lot in that timeframe and, uh, you know, we would partner for, you know, 12, 18 months with these businesses trying to get them from zero, like a strategy to scaled media revenue and a situation where they could operate [00:12:00] it on their own. So it's quite fun work. It's probably why. Stayed at McKinsey longer than I ever expected to, but, um, jumped over to Moloco. 

[00:12:08] Jon Flugstad: I thought that, you know, kind of knowing foundationally something about commerce media and then seeing kind of the AI and, you know, ML revolution happening, like who is bringing those things together? I didn't really see anyone else doing it in the same way. So made a hop and it's, it's been great. It's been a really great transition. 

[00:12:23] Tom: Yeah, we are very interested in trying to figure out where ad serving goes. So, you know, I'd love to hear a little more about, how Moloco got started. You said you have a DSP that scaled, um, like what was it that, I mean, the biggest question in my mind is what made you think I should leave McKinsey for an ad serving company? 

[00:12:44] Jon Flugstad: Well, I think my wife thought I should leave McKinsey. So that's probably for answer number one. And then you justify the rest of the reasons, kind of under the hood. But, um, no, she was very supportive. Uh, yeah, so like Moloco got its start, you know, 11 years ago. It's actually a pretty old company at this point. 

[00:12:58] Moloco's Founding and Growth 

[00:12:58] Jon Flugstad: Um, the founder's, a guy [00:13:00] named Ikkjin Ahn. He, uh. He was one of the early like machine learning engineers on YouTube. So you know, Google acquired YouTube. He was losing money. They were trying to figure out how to monetize the asset. And of course, like, you know, Google has invented a lot of the, the technology around machine learning over the last 10, 12 years. 

[00:13:17] Jon Flugstad: Like they invented the transformer model. They invented a lot of the. Things that make AI possible today. And he was in, in the belly of the beast. And now, you know, whatever YouTube is $30 billion, you know, a quarter or a year. I forget what it is, but it's, you know, super profitable. And it's all based on kind of AI based ad serving to find the right user with each ad. 

[00:13:34] AI and Machine Learning in Ad Tech 

[00:13:34] Jon Flugstad: And I think his, his mindset was, well, how come these assets need to be acquired by the big techs? Like why does Instagram need to be acquired to monetize properly? Why does YouTube need to be acquired to monetize properly? Why can't we build parity or even a more advanced infrastructure for other companies to monetize, um, so that they don't have to like sell the farm in order to do it in the right way and make it easier [00:14:00] for 'em to grow. 

[00:14:01] Jon Flugstad: And so, you know, he started out building, you know, at the time he called operational machine learning infrastructure. That's kind, you know, people call it AI now, but that's kind of the, the way it was framed. Looking for a use case to get started. They They started with, uh, kind of mobile gaming, mobile app install based performance, DSP, which, you know, is kind of pure raw performance. 

[00:14:21] Jon Flugstad: Like the ML can do a lot of damage and drive really high performance. And it's also like a way to scale quickly. And that's been a really awesome business. Like it's grown a ton. And then, you know, looking for other applications where can you really use first party data in a. In an environment where you can drive a lot of value, like commerce Media was kind the next bulkhead. 

[00:14:39] Jon Flugstad: And so, you know, three years ago launched Moloco Commerce Media and we are where we are today, kind of a holistic onsite ad server that really is the DNA and ethos is around performance and using AI for driving, you know, better, more efficient auctions and kind of operational simplicity. We buy and sell ads, you know, any way you want as a [00:15:00] retailer or, uh, on your site, like, you know, kind of traditional formats, et cetera. 

[00:15:04] Jon Flugstad: So, um, yeah, that's kind of the, the foundation of the company. The ethos is really around performance, driving the frontier on ai, and you know, how we use it across the whole stage of serving ads and bidding and pacing, and. Making all these things more efficient. So it's, it's kind of a fun place to be and to, to learn a lot, frankly, from all the amazing product, product engineering folks, like I've learned a tremendous amount in the organization just over the last year and. 

[00:15:28] Scott Messer: I absolutely love that evolution, and that's a great story. I'm, I wanna get back to this moment, but before that, like what was the first, what were those years like of Moloco between, uh, the founding and the retail media sort of boom there? What was, what was Moloco before retail media? 

[00:15:46] Jon Flugstad: It was that, just that really that mobile DSP piece, that was what they were building. And you know, there were, I think there were some really early years where there was like 10 people, right? Where they were trying to build, find the right use case and you know, know, like we know this operational machine learning thing is amazing. 

[00:15:58] Jon Flugstad: I mean, I can't speak on behalf of [00:16:00] iGen and DJ and some of the other folks who founded the company, but they found a use case and it really. Exploded. I mean, it's a really, I, you know, now if you look at Moloco, it's in the Financial Times Top 50 growers in the US across industries and number one scale and growth in ad tech. 

[00:16:16] Jon Flugstad: So it's like, it's a healthy business that they have built and that early kind of DSP piece is a big portion of that. But, you know, the, the second largest product is Commerce Media and it's growing really, really fast. So it's super exciting times. 

[00:16:28] Scott Messer: Talk to me about how, maybe like the, the ethos of A DSP, right? And that's like a performance buying engine then translates into being an ad server. Uh, I'll, I'll give a a, a moment here. Like in programmatic display, there's always been a lot of theories of. Why can't, uh, a publisher use a DSP to buy their own inventory on behalf of a client and make it a more performant campaign? 

[00:16:53] Scott Messer: Uh, 'cause we publishers, uh, who use Google Ad Manager say that ad [00:17:00] manager is a very dumb pipe and it just executes orders and it doesn't think about anything except really pacing. Um. With a couple exceptions. So how does like the, the ethos of what A DSP was move into, uh, sort of the, the founding of the ad server and how does that set the ad server up for performance retail media or whatever? 

[00:17:22] Scott Messer: Tom doesn't want it to be called, 

[00:17:24] Jon Flugstad: That's such a good question, and in fact, I think ad serving is actually a better use case even sometimes than the DSP for where you can use that kind of performance mindset to drive efficient outcomes. Why? Bunch of reasons. One, you, you're doing it on your, if you're like a retailer and you're using say, our ad server. 

[00:17:43] Jon Flugstad: You're doing it in your own and operated digital properties where you have like complete control over the auction. So you have all of the signals, you have all of the data, you've got your a, your longitudinal customer data. B, you've got all the data on your catalog and products, which you can like learn on, use, you know, [00:18:00] image embedding models, LLMs, and reviews. 

[00:18:01] Jon Flugstad: It all becomes data, right? And then C, you have all the real time browse events like, you know, just like Meta uses their Andromeda and GEM models for predicting sequences for like, you know, driving the most efficient a, you know, ad unit. Like you can do that too. And so you have amazing, like a. A corpus of data that's insane that you can, and, and wasn't really being used. 

[00:18:23] Jon Flugstad: Like you said, the, the GAM kind of dumb pipes. Like yes, there's some intelligent pacing. Yes, there's some decisioning across like exchange based demand, but exchange based demand sucks. If you're a commerce media network, it's just cheap. It's, you know, it's like you're foregoing value in a major way. 

[00:18:37] Jon Flugstad: Moreover, like the reliance on tools like GAM meant that you were cementing a whole bunch of bad habits around the manual way of selling. So you had to do like IO based, like, you know, upfront selling, very manual. And those ads are bad for shoppers too, at the end of the day because it's like you're, you're prioritizing an IO that has a commitment of, you know, impressions. 

[00:18:59] Jon Flugstad: Like [00:19:00] that's not relevant for a shopper, it's just an ad that's in their face and you kind of hope it performs. If you're a, you know, a retail media network, you have all this data to balance the interest of the advertiser and the shopper, and it needs to be like, performance is like the beautiful  

[00:19:14] Retail Media and User Journey 

[00:19:14] Jon Flugstad: medium where you actually like have to serve something relevant for the shopper, or it doesn't get clicked and you don't get paid like that. 

[00:19:20] Jon Flugstad: And so it's actually a beautiful use case to use the kind of AI based, you know, auction time prediction. Ad serving because of that safe environment to use the data and a set. And you have all the signals around sales and conversion too that help you kind of train the models in, in the right way and lead to like really high fidelity inferences. 

[00:19:39] Jon Flugstad: So, um, I think it's a really amazing use case for that kind of performance ethos. And frankly, it's where the, it's the direction in which, you know, we, we need to move versus some of those legacy models. 

[00:19:50] Scott Messer: Absolutely they, it was just kind of like exploding my brain a little bit in a really good way of listening to, um. Sort of how [00:20:00] like, uh, the bad habits being cemented, the bad habits of selling, being cemented inside gam, I think really resonates with a lot of publishers. Um, you get into like the journey of the user a little bit. 

[00:20:14] Scott Messer: How is it different inside retail, both from like. What the consumer is doing. And then also the responsibility of the retail media network from like a traditional publisher or things. I mean, this is gonna set us up for later, like where are we going in programmatic and like what happens with networks and demands later? 

[00:20:31] Scott Messer: So it's like, how is the user journey different inside of a retail website versus anywhere else? 

[00:20:39] Jon Flugstad: Yeah. Yeah. Broadening the perspective on where everything goes, like, I don't know, nobody knows, but within the world of like com, you know, retail media, how like. You have basically high intent shoppers that exist in your ecosystem. Um, and you have, you know, privacy, safe ways of knowing [00:21:00] what they're up to. 

[00:21:01] Jon Flugstad: And you have, like I mentioned before, the data around the catalog and their history. And the ability to basically use all that data in ways to make predictions around, you know, efficient auctions and what to serve each shopper. That's the, that's the way the retailer needs to think of, you know, serving that customer and elevating like the need of, of the, of the shopper. 

[00:21:24] Jon Flugstad: They have the data to do it. And so what and what's not really been happening, which I think is really critical is, and this is a bit of like what, what we're doing and kind of our theory on what makes us different is. Like, every thing you do on the site represents a sequence. You know, like Scott, you like go to product description page and you go back to a search results page, you go back to the homepage, you go to your cart. 

[00:21:49] Jon Flugstad: Those are all signals and you'll end up looking like other shoppers that have done that sequence in the past. And like the way it kind of works under the hood for, for us [00:22:00] is, you know, think of chatGPT They've read every sentence that ever exists. And they predict tokens and each word is a token and like the, the, you know, the model fires and they've got, you know, weight, you know, weighted vectors that basically predict what word should come next and it comes outta some kind of coherent sentence. 

[00:22:16] Jon Flugstad: The same should happen for ads. Like you have a sequence based on your behavior that looks like other sequences that have existed that lead to outcomes that are desirable for everybody. That prediction should have an ad impression, like you should create a bespoke impre, you know, prediction. Not just click through, but also will you convert for that user? 

[00:22:35] Operational Machine Learning Models 

[00:22:35] Jon Flugstad: And that means you show a much better, it's actually like a one-to-one personalized ad at scale versus, you know, to use jargon that's been in the industry for a really, really long time, but it's actually true. Which is, which is convenient, right? Um, and so you're serving the right ad relevant for that user and that like retailers should be trying to serve that kind of shopper because they're high intent. 

[00:22:51] Jon Flugstad: They're already on your property. Like that's, you wanna give them what's most relevant. You have to use everything that's in your arsenal to do it, like all the firepower to, to do that. So. Answers your [00:23:00] question, but it's kind of like within that context of what's right for the shopper and the journey, like that's the way we are thinking about it. 

[00:23:04] Jon Flugstad: Um. 

[00:23:06] Scott Messer: It. It, it builds together a lot of the, the theories of like sell side decisioning, right. And building more complex, more. That, uh, A DSP could never see because they're further away from sort of the buying decision. How does Moloco, I'll ask you like a, a quick product question as my dog and baby walk in here. 

[00:23:27] Scott Messer: Um, how does, uh, Moloco sort of build those models and like, how many models can an individual publisher or an individual retailer like have. 

[00:23:39] Jon Flugstad: Yeah, that's a great question. Um, there are like numerous models in the kind of like stage of operational machine learning. So you're using, you know, LLMs to read text. You're using image embedding models to understand imagery. And you need, stages of these things to like retrieve the right ad candidates, score them [00:24:00] across like very complex catalogs. 

[00:24:01] Jon Flugstad: So it, that's kind of like some of the secret sauce is like that process and like the models that are used under the hood. And to do that in a way where it's like, you know, very low latency. You see some folks out there that are like, you know, 10 milliseconds on ad serving and it's like you're not using machine learning and ai. 

[00:24:18] Jon Flugstad: If you're  

[00:24:18] Technical Insights and Retail Media Evolution 

[00:24:18] Jon Flugstad: doing that, you know, it's you, you have to be able to create these predictions. 60 milliseconds or something like that, you know, really quickly to provide the ad response to serve it properly. And you know that that's kind of the secret sauce around it. And the way we approach it is actually these are bespoke models for every customer siloed. 

[00:24:38] Jon Flugstad: So if we're working with a retailer, we're not using that cross model learning or inference or any of their data anywhere else. It's like its own instance in the cloud where it is trained only on their data because it's their proprietary data. And so we're not gonna like. Use their data for our own ends and go in some, you know, external call and say, this is how many customer profiles we have, let us go sell 'em to you. 

[00:24:58] Fragmentation in Ad Serving and In-Housing Trends 

[00:24:58] Jon Flugstad: Or this is, , something [00:25:00] we learned from another retailer. It's like very siloed and it uses their data for their models because their users are, are unique. Um, and so, you know, I, it doesn't really matter like how many models you have, but like how you use the data and like the flow of models to be able to get to predictions pretty quickly is, is really important because. 

[00:25:17] Jon Flugstad: Every millisecond counts. Like that's how, you know, Google in its early days would give you the time it took to search result because they know time was paramount. That's why they invented a lot of the tech they did because they were trying to do predictions fast. Um, and so, you know, again, like visiting how many models, I don't know. 

[00:25:33] Challenges and Strategies in Retail Media 

[00:25:33] Jon Flugstad: I'm not a, I'm not an engineer, um, say that first and foremost. But you know, that it's that ensemble of models to, to balance speed and, uh, fidelity is really, is kind of the, the sweet spot. 

[00:25:44] Tom: Your ability to talk about how the models work, I think got us a little to the point where we're like, oh, maybe he knows all the technical details. Um, but I, you know, but I, I do think it would be interesting to pull back a little bit to the retail media. Construct that you talked about in, say, [00:26:00] 2018, where it was very much you, you picked an ad server and a sales team and a customer service team and you know, it was end, end, end, end. 

[00:26:08] Tom: So that it was a turnkey solution. When we had our discussion, I think it was probably a month ago, um, you know, we talked about that post that Kathryn Lundstrom wrote, uh, in Adweek about how. The traditional go-to partners is fragmenting in terms of ad serving in the retail media space. 

[00:26:28] Tom: Um, and what you were just talking about with Scott is starting to give me an understanding of what the new sell is because what I understood as I was leaving the retail media operator business was that everybody was gonna in-house everything, um, or at least for a little amount of time. That was the, the trend. 

[00:26:46] Tom: And you're out there trying to sell an ad server. Where you're probably not suggesting to in-house because you're doing all this stuff with ai. So can you talk to me about what the market is like today? 

[00:26:56] Jon Flugstad: Yeah. I think the, like there was selective in [00:27:00] housing that had to happen from that era, there was the early vendors that were doing everything would just cut you a check and say, look at this free money. And then, you know, the retailers realized that they were actually losing half the margin they could have made from those vendors. 

[00:27:12] Jon Flugstad: And so. There was like the kind of the early wave. Yeah. Not to point fingers. Um, there was an early wave where it was like, well, what do we in-house? Some said we're gonna in-house everything. Like we should build an ad server. We should build all the underlying, you know, audience management, tech. We should build, um, the measurement. 

[00:27:32] Jon Flugstad: Uh, we should, you know, we should be doing 

[00:27:34] Tom: mean, Bobby Watts sat me down at at grocery shop and showed me how he was, he was gonna do all that in-house. I was kind of shocked. 

[00:27:41] Jon Flugstad: Yeah. Yeah. And they, you know, they've acquired some tech, they've partnered with companies that have helped them kind of build, operate, transfer. It hasn't been all alone. I think they're trying to do some, you know, Bobby's awesome. He's a really, really good 

[00:27:52] Tom: Yeah, Bobby runs the ah, Ahold-Delhaize, uh, ad retail media, USA business. So, 

[00:27:57] Jon Flugstad: Yeah, so other and others have tried to build [00:28:00] it in-house and um, others have said, we're gonna in-house selectively say we're gonna try to build a measurement layer across all these different pieces. We're gonna try to build an audience layer across all these pieces. Maybe we'll try to do some decisioning across all the different like logos we're using. 

[00:28:13] Jon Flugstad: That's also been really hard. If you think about where a lot of the media networks are now is. Either they've tried to build everything and they're kind of hitting a wall because the number of like expensive AI engineers, you need to hit this next wave. Like what ad serving looks like is actually very challenging. 

[00:28:30] Tom: many do you need? Uh, do you need engineers? I thought you just, you just put things in the model and you get what you need. 

[00:28:34] Demand Models and Market Positioning 

[00:28:34] Jon Flugstad: Yeah, yeah. No, you do It turns out, yeah, you need, you need lots of engineers and they all, it turns out engineers are expensive. 

[00:28:40] Tom: But I, I think you're giving us the answer that the tech is there, but do you still have to do the selling right. 

[00:28:47] Jon Flugstad: um. Yes, because of the related special relationship you have with that vendor. If you're a retailer, there's going to be high touch for some top tier of, you know, your, your vendors. Like if you're Dick's, you, [00:29:00] you have a special relationship with Nike. You know, it's just, it's just the nature of it. Like there's so much like retail media is one component in a broader merchandising and strategy together where you're trying to drive top line growth. 

[00:29:10] Jon Flugstad: And you know, Brian Monahan talks about this, like the beauty is the shared incentive across both to drive top line sales. Where I think media networks have gone wrong is they've over-indexed on that model for too broad of a set of vendors, not just the strategic ones, and it's not scalable. So then it's like, I need to add another sales head to with a hundred K quota to get my media sales up. 

[00:29:31] Jon Flugstad: And you're like literally just counting the people you need to like grow your, your media network. That's not gonna work. That's not what like Meta's doing. No scale digital media company has done that. Like in terms of this is the way I get to sales. They, they're self native. They make it really easy to buy. 

[00:29:47] Jon Flugstad: Is it a hundred percent that way? Absolutely not. But the, there needs to be a shift in terms of the amount of which is being bought self-serve, or in ways that advertisers wanna buy to make it a lot easier to act. Didn't mean [00:30:00] you like lose the controls as a retailer, but it means that that like IO based upfront sales model. 

[00:30:06] Jon Flugstad: Needs to be less of the, like crutch because it's just too hard. It's just like, it's just way, way too much work in our, in our opinion. So, you know, I think it like changes the nature of what you means by demand in, in that regard. And you know, it means that there's new approaches that can be tried that I think, 

[00:30:22] Tom: Well, what, what, what does Moloco consider? What's your positioning around demand? Do you support some sort of, , demand model that, that you help your retailers with? 

[00:30:32] Jon Flugstad: Yeah. Yeah. Such a, such a good, such a good question. Um, because demand is like a, such a hot topic, and I think. I think we frame the problem a little bit differently, um, just to, the problem statement is really like, how do we increase bids and budgets to drive auction density? Like, and, 

[00:30:49] Self-Serve Dynamics and Machine Learning 

[00:30:49] Jon Flugstad: and if that's like the problem framing it, it takes away some of the like expectation or the way we've conflated demand, I think, in the market, which is. 

[00:30:58] Jon Flugstad: Do I have like a boiler room of [00:31:00] sellers that are like dialing for dollars? You know, that can be a piece of it, no doubt about it. But that's not it by any, by any measure. And that has all kinds of problems. Like if your ad tech is the one that's dialing for dollars on behalf of your retail media network, they're also dialing like across retail media networks. 

[00:31:15] Jon Flugstad: They're not yield optimizing for you. They're actually quite conflicted. And so I think as a model, it's like pretty murky. Is that being like the path for growth? In reality, there's probably three like structural things that we, in the way we think about demand. Like one is tech, like that's the answer to a lot of it. 

[00:31:31] Jon Flugstad: Two is like, yes, operational, like sales, like, you know, how do you scale that? And then three is partners. So on the tech piece, like you can make, you know, start a campaign with three clicks with like the best targeting possible with dynamic bidding. Like if you're using Moloco, which means you can expand self-serve and that, you know, if you're driving better performance, you get existing budgets up higher. 

[00:31:52] Jon Flugstad: With like real nudges and realtime nudges about like what you should change in your campaigns, but also expanding the portfolio of self-serve in a way that's, you know, easy [00:32:00] for whether you're marketplace or, you know, you're meeting long term, long tail sellers and you just can't treat with people like that, that that's the way to like get those folks into the ecosystem and drive density. 

[00:32:10] Jon Flugstad: Yes. Like there should be sellers as a piece of it. Like we could, we will do that for large partners and we will also train your teams on how to sell. Then the last piece is like, where do people wanna buy? Like, you know, we're partnered with Sky, that's public. We have other API demand sources that are, you know, going to be public quite soon. 

[00:32:28] Jon Flugstad: Like philosophically buy where you wanna buy, I guess is like, sort of the other approach and across those levers, like it's not just one monolithic thing on what it means to drive demand, or do you bring demand. It's like, it's a really, a holistic approach, I think is, you know, kind of what needs to be considered. 

[00:32:42] Jon Flugstad: And tech is a bit of the foundation of that, at least in our view. And others will disagree, but. Um, you know, people take different paths and that's kind of the one we're, we're thinking about. 

[00:32:51] Scott Messer: It's, really interesting, the self-serve dynamic verse, the iOS, I think you really crystallized something for me is that if you wanna [00:33:00] get off of. The, the IO business, you, you go self-serve and that's the obvious part, but how much performance you need inside self-serve to let clients sort of just say what their budget is and some basic targeting and parameters and some outcomes. 

[00:33:19] Case Study: Wayfair's Partnership with Moloco 

[00:33:19] Scott Messer: And then like the, the machine is doing an enormous amount of heavy lifting that is. Finding those outcomes for you, which is what you were talking about, the Andromeda models and all the things that like Facebook and YouTube have built to make that automatic so that their self-serve can scale, which is I think, probably a large disconnect for many retailers who think we'll just open up a, a self-serve website and they'll just traffic in insertion orders and then those will perform. 

[00:33:50] Scott Messer: And like honestly, if, if I was a self-serve advertiser and I just saw like. Basic ad server campaign delivery, I wouldn't be very [00:34:00] happy. But if you knew that you had this enormous machine learning model that was delivering outcomes and it kept getting better and better and better, every single month, you would continue dropping dollars into it. 

[00:34:11] Scott Messer: Um. So it's really, it's a really good highlight there that self-serve isn't automatic and it isn't a replacement for insertion orders. It's an entirely different optimization service that people can, that is so good that advertisers can just do it themselves. 

[00:34:29] Jon Flugstad: That's the idea. You know, they, they require some like retraining and, you know, there are sophisticated advertisers who still want knobs, but like. Kind of paper out after paper in the academic world says that people can't do it the way the machines can. You know, there are, or folks who are inside Google who are testing P max early, they went to some of their biggest customers and said, who had reams of data scientists who had, you know, very sophisticated bidding models and said, you know, can you, can you beat, can we test against like our dynamic bidding, our outcome-based, bidding based on the [00:35:00] inferences we can create? 

[00:35:01] Jon Flugstad: And the results weren't even close. It's just, it's so much more efficient. And like you said, if you can just say, this is my budget, you know, I am gonna set a 400% roas, target it just completely, or, or if you're like a food delivery app, like I can set a price per order because I have AOV predictions, I have conversion predictions, I have click through predictions that are really, really good. 

[00:35:26] Jon Flugstad: It just completely de-risks it for the advertiser and makes it so, so simple. And so, you know, some will hold on and wanna do the knobs and like, we believe in choice. It's not like, you know, you're gonna be like dogmatic about, like, you must automate everything. But it should be an answer to a lot of the, a lot of the things, and it should make it easier for advertisers in, in, in general is just kind of philosophically where I think things are moving. 

[00:35:45] AI Use Cases in Commerce Media 

[00:35:45] Tom: I wanted to, to get you to talk a little bit about, I mean, we've been very theoretical here, so, um, I'm assuming there are some clients, there's some customers that Moloco works with. So can we talk about, let's say Wayfair? 

[00:35:57] Jon Flugstad: Yeah. 

[00:35:58] Tom: like what, 

[00:35:59] Jon Flugstad: Large. 

[00:35:59] Tom: [00:36:00] like, 'cause I think that would bring it home for me. I mean. 

[00:36:02] Tom: Especially because I worked like mostly on grocery and maybe dollar. Um, I am super interested in hearing sort of what, you know, if, if you're looking at really like high ticket items like furniture, how does that change the game or does it. 

[00:36:16] Jon Flugstad: Yeah. So I mean, Wayfair is really, really interesting. They have built a lot of their own tech and they're a, they're like a customer we love, like they're very sophisticated. They have a lot of ad engineers who are, you know, really sharp and. The idea was like, once we kind of, you know, started talking with 'em, it was like we can accelerate the things they, that they want to do, and they're, you know, they're, they're a retailer. 

[00:36:40] Jon Flugstad: Like they have, you know, they're not just like marketplace, so they, you know, are holding inventory and, and that kind of bear risk around that. They also have like a very complex and large catalog. And so, and they're high ticket items. So you need to be able to like, predict with some fidelity, the purchase of, of, you know, high ticket items. 

[00:36:58] Jon Flugstad: So, you know, we [00:37:00] power some of their ad formats, not all of them, um, but they're a, a really deep partner of ours. And, you know, like the performance has been outstanding together. Like, we love the partnership. We've, we've spoken at a few kind of conferences together, um, because it's just really a, like a better together story like. 

[00:37:18] Jon Flugstad: Pretty quickly, you know, we did like a proof of concept with them and the, the improvement of click through rates and like performance was, was really marketing. Like, you know, I'm not gonna like share specific numbers. There's a lot of, you know, I think egregious number sharing in our industry of like, we did this, you know, and we 

[00:37:34] Tom: I always tune those out anyway. That's 

[00:37:36] Jon Flugstad: 20 x their ad revenue or whatever. 

[00:37:38] Jon Flugstad: That's silly. It's a very sophisticated team and we help them go farther faster and it's a really awesome deep partnership and one we're really proud of. And you know, now we're. It's gonna be onboarding some other like, major, major retailers here in the next quarter where, um, you know, I think folks feel like our tech and our approach is a little bit different and they want to build something that's pretty differentiated and kind of that Pmax [00:38:00] style offering within the context of retail media. 

[00:38:02] Jon Flugstad: And you know, that's one of the things that Wayfair really did like is the sophistication of our bidding algorithms and allowing for outcomes. Making it simple across some of their ad formats like that was very attractive to them. And yeah, we, we, we love that partner and wanna keep serving them really well. 

[00:38:18] Tom: So, um, we don't really do like lightning rounds. But you actually brought your own visual aids, uh, this time 'cause you had a, you had a presentation you shared with us about AI and commerce. Media. Um, and you had a whole bunch of, uh, use cases and I wanted to sort of, you know, for, for those who are watching on YouTube or on Spotify, um, you can see this on your screen, but if not, you know, for those, uh, who are on Apple, um, let's go through some of these and talk about what we have because, um, I think it's interesting the way that you've set this up in creator use cases, optimize our use cases and operator use cases. 

[00:38:59] Tom: So. [00:39:00] Like, give us the, the, the, the, the voiceover for this slide and tell us sort of why you think MoCo is, is cares about these use cases? 

[00:39:09] Jon Flugstad: Yeah, I, a former consultant than me, I couldn't help myself. Like I just, there was so much discussion on AI in the market and agent and kind of went to these certain endpoints that I didn't know, like where they fit in terms of context. Uh, but yeah, because there are kind of three broad roles right there. The way I think about is like creating new customer experiences and surfaces for commerce and ads. That's where a lot of the discussion goes, I think. 

[00:39:35] Jon Flugstad: But, you know, changing the fundamental way customers will interact with ads and AI enabling that is a, is a key piece, I think very, very transformative. And one, we should talk about the piece where the pieces where, you know, there's really impact happening right now, and where there's real juice to squeeze is on optimizer and operator. 

[00:39:54] Jon Flugstad: Optimizer, meaning how are you decisioning on ads and using data to personalize ads [00:40:00] for your customers. Then the last one is operator. Like, how are you making your, you know, ops more efficient? Whether it's some really exciting startups and tools around creative generation, you know, gen, AI based, creative, whether it's, you know, ad op, you know, ad ops or rev ops or measurement or, you know, like these different pieces that agents can do pretty well because they're kind of processes that, you know, are, are pretty predictable. 

[00:40:25] Jon Flugstad: Or, you know, underlying pieces like audiencing or generating insights from campaigns or new intelligence on, you know, these are things that these models can do really effectively. Um, and so this is just like a broad kind of structuring of the market. I, I think, you know, like 1, 2, 3 is where a lot of the discussion has been. 

[00:40:46] Jon Flugstad: And you know, again, these are maybe jar. 

[00:40:49] Tom: before you go into what you think of it, so just so the audiences, the one, two, and three, you have as either onsite, uh, LLM based conversational ads or offsite. [00:41:00] Conversational ads. You, you, you characterize those as basically what the, the Sparkies and the Rufuss is like you're on a retailer and you're using their, uh, site versus offsite. 

[00:41:10] Tom: Is that like chat, GBT or offsite meaning like an ad has a chat in it? Like, just tell us what, what those mean first. 

[00:41:16] Jon Flugstad: Yeah. Yeah. You're, no, you're, you're, you're pretty much spot on. So like, you  

[00:41:20] AdCP Standards and Industry Impact 

[00:41:20] Jon Flugstad: know, onsite, LM based conversation lines would be the Rufuss or the Sparky, or, you know, home Depot has one that they're trying and. You know, will there be sponsored recommendations or answers or, you know, the way, however, you know, like ads will end up being embedded in that kind of experience. 

[00:41:35] Jon Flugstad: As, you know, consumers bring intent. Not just a search, not, not just a search bar to try and find and discover new products. I think it's like a real evolution around how people are finding products. It's really, really critical. The second one, as you mentioned, is offsite, all on based conversational lines. 

[00:41:49] Jon Flugstad: This is like in, in the experience of the frontier models. So you know, if you're using Claude or chatGPT or Gemini or you know, perplexity or whatever [00:42:00] tool, um, they're gonna be ads based, like you just saw that chatGPT launched a browser like that gives them. All kinds of data and like a, you know, a conversational native experience for finding and discovering things like there will be advertising and if you're a retail media network, thinking about offsite customer acquisition or where your customers are going to be, not just the intent based marketing to those are already on your properties. 

[00:42:25] Jon Flugstad: You are gonna have to really think about that and test into it and, you know. How those ads will work and, and how you're gonna kind of partner with your suppliers to, to draw folks back to. 

[00:42:37] Tom: But based on what you, you have on this slide here, you're, you're saying that these are really the most complex and difficult, and so are you more excited about, you know, the ability for Moloco to do the optimizer and operator in the, in the short term? Is that sort of what this slide's saying? 

[00:42:54] Jon Flugstad: We can do those now. So that's, it's, it's a much clearer use case and frankly, where [00:43:00] there's, you know, I would call it kind of low hanging fruit, like if you are ad serving in kind of a legacy manner, if it's, you know, buying keywords or not using realtime browse behaviors or not having inferences that are generated at auction. 

[00:43:14] Jon Flugstad: You have a ton of upside and then you can do it tomorrow. You know, it's, that's, that's the thing that I think is kind of lost sometimes in the AI conversation is that you look at Meta's earnings, it was like, you know, 22%, you know, growth year over year. And it's, they're attributing it to these models like that, that's what they're doing. 

[00:43:31] Jon Flugstad: Like they're, you know, maybe they're a little late to the game versus Google on the kind of inference and, you know, the model. The impact is crazy for them at scale. Like a massive, you know, a a a ma, massive, massive player. Like the same can be true. And so there are a lot of use cases where, yeah, that middle bucket optimizer is driving a lot of impact. 

[00:43:49] Jon Flugstad: Um, and you can do it now. I think on the top one, I just think nobody knows. That's the, that's the like, simple kind of dumb answer is like. We all know there're gonna be [00:44:00] something like adoption will happen in pockets. Maybe it's really fast and like, you know, all your onsite ad revenue goes away because everything is happening in, I dunno, maybe, but No, but nobody really knows. 

[00:44:09] Jon Flugstad: And so you have to like really pay attention to those ones. But you can't let the conversation only be on the unknowns. I think in the context of AI for commerce media, like if you're an operator, what do you do now? And that's, that's the mindset was trying to, kind of, trying to take. 

[00:44:24] Scott Messer: This is great. We use different words sometimes, like I see operator a lot as like workflow Optimizer is more like creative and decisioning. Um, and then creator is, um, AI rendering. Uh. It's, it's a really fascinating thing. But on that sort of unknown piece, like last week AdCP was announced. 

[00:44:47] Scott Messer: How do you think like, you know, this sort of burgeoning standards that are coming out, um, might impact the building strategy of a Moloco or somebody that has like an agentic [00:45:00] future in their, in their hands. Like, how do you react to something like AdCP. 

[00:45:04] Walled Gardens and Merchant Relationships 

[00:45:04] Jon Flugstad: Yeah, the short answer is I don't know yet. Um, I don't know who, who knows yet. Um, I think. Generally, I mean at, you know, if you look at AdCP, it's, you know, kind of being developed by a lot of the folks who are involved in the programmatic revolution. So smart folks who have shown real impact and built serious companies. 

[00:45:25] Jon Flugstad: So I think that means like should have a lot of credibility around it as a, as a starting point, I also think it like, moves the ball forward in kind of an open way on use case three around agentic. Like, you know, how will publishers, selling agents and buying agents. Communicate, um, you know, if you're doing it in a thousand different ways across different sites, like that's, that's really a tax on the industry. 

[00:45:50] Jon Flugstad: And so finding a way to do that in a more common language, I think. Foundationally in principle is probably a really good thing. The details I'm not an expert on and also like how it affects [00:46:00] Moloco, I don't know yet. I'm not sure if it impacts like our onsite ad serving for now. I think it probably has a earlier use case on more like open web type publishing and kind of, you know, broader buying versus where you have like very clear direct endemic relationships. 

[00:46:15] Walmart's Trade Marketing Shift 

[00:46:15] Jon Flugstad: That, you know, so I have a general high level principle of like, step in the right direction, trust some of the people. It's quite credible. Um, but how it'll impact, like, I, I, I don't know. I don't know yet. Uh, maybe you guys know. I hope, hope you do. 

[00:46:29] Future of Ad Servers and Data Collaboration 

[00:46:29] Scott Messer: I mean, that's, that's about as good a summary as I think, as, as anyone can land on. I think the important part is if you have an existing business, uh, you need to continue running it and you can't pivot to this thing entirely. If you have a new business, you may want to consider it. Um, the hard part to understand is like, will these standards specs sort of really dictate the future of how you build something and then how, how fragile is your business, whether it's [00:47:00] written on your own proprietary set of standards or if you, and to move it into an open standard, like you think way back to like the ORTB days when we went from like pure ad networks into an auction model. And you know, how many people were like destroyed by that? Like none really. They were able to transition their code. Um, and I think anyone who's running an existing business probably doesn't have to be the early adopter of AdCP, but somebody in the basement should be thinking about it and figuring out how these pieces are gonna come together. 

[00:47:35] Conclusion and Final Thoughts 

[00:47:35] Scott Messer: Um, because, uh, I've heard this phrase more in the last. Three months than I have ever. And it's one of my phrases favorites, which is, how did you go bankrupt? Um, and it's in two ways, slowly and then all of a sudden, and that is the same way that sort of the agentic revolution is happening, which is slowly and then all of a sudden.[00:48:00]  

[00:48:00] Jon Flugstad: yeah, yeah. I agree. And you know, I think it'll hit differently for like broader based publishers and commerce media networks. You know, I think. The term walled garden is often a pejorative, like people use it in a dirty way and they kind of, you know, hate the non sharing of data and proprietary buying and things. 

[00:48:19] Jon Flugstad: But like commerce, media networks are kind of walled gardens and they should kind of act like it in, in my opinion. They have this like really unique value and they are not just a general open web publisher. They have transaction data, they have consumer relationships, they have unparalleled data and like programmatic auction based. 

[00:48:37] Jon Flugstad: A lot of publishers, it was pretty tough, you know, like transition for, for kind of their monetization models and you know, it kind of made it maybe efficient on the buy side. And again, I'm not an expert in that like wave of things, but I just think like to go that route and be kind of fully programmatic and opening things up as a commerce media network, I'm very hesitant because I think you have something a lot more special. 

[00:48:59] Jon Flugstad: Um, [00:49:00] and. Keeping some wall around that garden is like, you know, should, should you be fully, I. 

[00:49:07] Tom: Well, it's also what you said earlier, there are merchant relationships. So I think that those, those merchant relationships, uh, require, and this is some of the things that we just talked about, which is, you know, we were talking about this on LinkedIn. Then Drew Cashmore is gonna be on, uh, soon on the pod. 

[00:49:24] Tom: It was sort of like Walmart decided to one day get rid of trade marketing or trade, uh, trade spend basically, and say, let's put it all into ads. And that was a, a, a mistake. And they, they copped that now, but back then the idea was, oh, this could be simpler, this could be automated, this could be one budget. 

[00:49:41] Tom: This, you know, and what we're seeing is that those, those relationships die hard because there are purposes for it. Um, you know, how many cases are you putting on the shelf is an important question that media doesn't answer. Um, but I am, I am happy that you came with us on the ride to, to talk about this because I, I [00:50:00] feel like the newer ad servers like Moloco have to live in that new world and, and change the game a little bit. 

[00:50:07] Tom: Um, and so yes, there are walled gardens, but um, you know, how much is data collaboration? How much is what you were talking about in terms of. Sequencing the buyer behaviors. Um, so those are the types of things. Honestly, that's the reason why we talk at the middleman and why Scott's on is to, to make sure that we're talking about things like AdCP and others that are at the bleeding edge. 

[00:50:28] Tom: So I want to thank you for, for joining the middleman and, uh, hope to keep in touch. 

[00:50:33] Jon Flugstad: Hey, thank you guys. Hope it was a useful conversation. Uh, yeah. Feel grateful to be here. 

[00:50:38] Tom: All right. See you. 

[00:50:41] Jon Flugstad: Bye guys.  

[00:50:42] ​