# S2 E31 T51 - Ari Paparo - Ad Server Wars_

S2 E31 T51 - Ari Paparo - Ad Server Wars? 

[00:00:00] Ari Paparo: I guess what I would say is like the, the PLA market is so, um, dumb, right? I mean, like, think about it, it's like, like I have a list. I have a list of SKUs. Each SKU has a budget and then the page loads and has a SKU in it. That's a pretty easy technical problem. Right? Um, so why do we think that the same systems that are built around PLAs are gonna be good at personalized top of funnel advertising? 

[00:00:31] Tom: You're listening to the Middlemen Podcast this week with Scott Messer and Tom Limongello. 

[00:00:37]  

[00:00:37] Tom: We've been as the Middlemen podcast talking for like a year plus now about a new set of ad platforms, ad servers, that are very close to the publisher, in this case, the retailer. If this is the second wave of ad platforms. What was the first like, and, and how did that, that sort of story play out?  

[00:00:59] Tom: [00:01:00] And so Ari was, he's the author of Yield, um, a great and probably the only approachable book that really gets into the detail of ad tech. In this era where Google became a monopolist. 

[00:01:11] scott messer: And it was great to have him on because, as we were thinking through the mechanics of like, how does an ad platform or an ad server win in this war? It would be great to revisit history and see like what were some of the elements that caused Google to win the original ad server war. Um, and do some of those elements exist in today's current war? 

[00:01:31] scott messer: And then it was really fun when we sort of got into the other pieces of the ad servers and what does he think of certain business elements of them and are they interesting from a, a market entry opportunity? From a, a product development opportunity. 

[00:01:44] Tom: Yeah. I mean, I, I think it's great to hear his perspective both as someone who lived through this also as an ad tech investor, thinking about, all right, what, what is the real opportunity for product listing ads? What is the real opportunity? Um, in the fact that currently there's not a lot of [00:02:00] standards in retail media. 

[00:02:01] Tom: What is the opportunity, in areas where retail media hasn't really matured yet, like personalization of advertising. And so all of those things have been things I've been thinking about on and off for many, many years. So it's just great to be able to riff with him about that. 

[00:02:14] scott messer: Yeah. And then we also plugged into like the, the programmatic access of inventory and where the DSP role is in this and like, what is the ad server. I don't know if Ari like fully understood exactly the, the complexity that retail media networks are facing when they are running PLA campaigns and how hard it is to actually, um, operate those. 

[00:02:36] scott messer: And so, so take a listen to the episode and see if you agree with what Ari's saying about it and how your day job is going. And it led us into a really interesting conversation about how DSPs aren't really present in the retail media ecosystem for onsite advertising. And that, legacy ad servers like Google Ad Manager aren't really up to the challenge that publishers had. 

[00:02:58] scott messer: And so we saw, [00:03:00] um, some of the early iterations of retail media, black box ad servers, sort of encompass all this technology. And the newer generation is. Exposing all those endpoints and letting you bring in the decision engines and own the algorithm. Build a better optimization system inside your ad server. 

[00:03:19] scott messer: Sort of that same logic and decision engine that lives inside of A DSP that rules the display world. Retail media networks are starting with that engine inside of their own system. Um, and we didn't talk about it, but I think it would be an interesting challenge. Like if DSPs were to come into this world, which one would do better and how would it, um, how would they, they face off against each other. 

[00:03:43] Tom: Yeah, his lens of looking at who can be a monopolist and who can win this war. Um, you know, retail continues to look very fragmented. But unless, you know, if you just have some big, super auction winner like a Google, um, that's one way to sort of show [00:04:00] scale. But does it get you to the level of sophistication? 

[00:04:03] Tom: , In the opposite idea where you, you get into the publisher and you're fully connected to all their pipes and all their channels. And I think that idea of, you know, most of the, the companies that we've had on this podcast, the Zitchas and Vantages of the world, are trying to take things that are sometimes, you know, very simple digital products. 

[00:04:23] Tom: You know, dumb PLAs as Ari would call it. Um, but also things are analog. , And also things as complicated as CTV. And so I think from that perspective, super fun conversation and I hope you guys all enjoy it. Welcome to the Middlemen Podcast. We are very excited today. Our guest is Ari Paparo Of Marketecture fame. And if you don't know him from Marketecture fame, we will introduce him a little bit. This may be the only podcast out there where he needs an introduction, but in the retail world, maybe you need to say that he came from Google, from [00:05:00] AppNexus. 

[00:05:00] Tom: He was the founder of Beeswax, which was acquired by FreeWheel, but he now has created Marketecture and made me wanna be a podcaster. So thank you Ari, and thank you for joining. 

[00:05:11] Ari Paparo: Well, thanks for having me. I apologize for making you wanna be a podcaster, but other than that, I'll take, I'll take the, the props. Happy to be here. Thanks for all your support and it strange coincidence. My, uh, first grade teacher was named Miss Middleman, so I feel like I am, uh, at home here. 

[00:05:29] Tom: Very good. Some people who read your book yield, uh, said that there were some negative connotations for, for middlemen. We totally own that. We, we love it. So, no hard feelings there. But, we wanted to get started and, the reason we wanted to have you on more than just to aggrandize our podcast, get traffic, get somebody who's an ad tech influencer on it was. 

[00:05:49] Tom: To talk about what we are calling the ad Server wars. So Scott, do you want to talk about what we, we think of as the ad server wars? 

[00:05:56] scott messer: I'd be happy to Tom and nice to see you again, Ari. The [00:06:00] ad Server Wars in short are, something that we're seeing inside Commerce Media. Retail Media start to play out where, ad servers in particular are shooting for control over a large part of the system of commerce media. When you think about onsite, offsite in store and all the various places where you can activate media, the ad server becomes incredibly important. 

[00:06:25] scott messer: That wasn't necessarily, as complex as it was for digital publishers perhaps until like CTV really came out. And what's interesting about Commerce media is that, ad servers that are coming to the rescue, because Google Ad Manager really isn't a consideration for, uh, retail media or commerce media network ad servers. 

[00:06:46] scott messer: They're coming outta the woodwork from a bunch of different directions. And it gave us this thought that these platforms are going to start. Colliding with each other in sales, which we're seeing happening and they're fighting over the same [00:07:00] customers. Their feature sets are becoming very similar, but they have different, initial capacities, whether it's traditional ad serving, machine learning, decision science, and then how do they fit into product catalogs and SKUs and things that were traditionally handled by Algolia and. BloomReach and other platforms. And then you'd put like this orchestration layer on top of it that we've talked about how do you coordinate across ad servers and bring businesses together. So as we see this war happening, Tom, you and I were chatting about like who's gonna win it 

[00:07:34] scott messer: and you know, like all great things in ad tech history repeats itself. And so we thought, who better to have us to give us a little bit of the history of ad servers. And platform wars than Mr. Ari Paparo himself. 

[00:07:46] Ari Paparo: Right. Well, it's a pretty complicated subject. Um, so you know, to start with, your typical retailer or commerce company might have three different technology stacks. You have the standard ad server for, display like units or video units. You [00:08:00] might have the, um, ask SKU oriented, , bidding system for, uh, PLAs, and you might have the offsite DSP. 

[00:08:07] Ari Paparo: Um, so that's a challenging problem if you're, you have a Salesforce that wants to sell a single io. That has all three of those and deliver it on time. And they're, they're even priced in totally different ways. You know, uh, the PLAs may be performance, the banners might be CPM and then with a guarantee. 

[00:08:25] Ari Paparo: And the DSP offsite stuff might be, um, CPM with no guarantee or, or variable amounts of volume. So it's a pretty tricky business problem. So that, that's kind of the scope of the technical problem. But then you have to segment the market because, um, there, when you say commerce media, it's just like so many different kinds of companies. 

[00:08:43] Ari Paparo: A typical retailer that sells products for multiple manufacturers or brands, right? Like the Dick's Sporting Goods kind of company. Um, and that's, you could, you could have that in your head and kind of understand their needs. Um, but they're really different from a marketplace company, like an Uber or, um, [00:09:00] or, um, any of the food delivery companies. 

[00:09:02] Ari Paparo: Yeah. DoorDash. Right? So, so those, those marketplace companies, um, are a closed loop system. They're, their advertisers are only gonna be on their system and they already have all the data. Right. And it's just a. Bidding and optimization system, you don't really need ad serving. It's ad serving as you think about. 

[00:09:17] Ari Paparo: It's kind of antiquated. Um, and then, um, and then the third thing to think about, and this also kind of parallels the segmentation, is where's the demand come from? In the multi-product retailer world, you might have demand coming from an external, you know, agency, uh, a technology like Skai, a technology like Criteo, um, et cetera. 

[00:09:40] Ari Paparo: And getting demand is kind of the hard part. In a, uh, marketplace company, you already have the demand. Like there are already retailers and restaurants on your system, uh, or brands on your system, and they just have to log in and start spending money. Um, and so the integration, I think is a lot less important. 

[00:09:58] Ari Paparo: Um, so that's a [00:10:00] big complex sort of, uh, way I see the landscape. Before I start saying, well, who's gonna win? I start saying, well, what are the, what are the different constituencies need? Um. It. If I was gonna take an example, what happened in Banner ad serving is that, uh, and this is what I write about my book, and it should be familiar to probably most listeners, which is, the publisher ad server, double click, DFP, which then turned into Google gam, um, got a monopoly on that business because the publishers were, um. 

[00:10:34] Ari Paparo: Sort of desperate for demand and there was a enormous source of demand that no one else has access to, which was Google AdWords. So all those millions and millions of small customers, uh, logging into AdWords and putting in their credit cards, were never gonna buy ads on, you know, the Washington Post unless Google enabled it. 

[00:10:54] Ari Paparo: And so when the. The game plan for the go that turned into the Google monopoly [00:11:00] that I write about in my book is that they kind of hard bundled those two things, or as the legal side said, they tied them and as a result, you couldn't leave the ad server without losing a really big chunk of your revenue, um, because it wasn't gonna come to you through a different channel. 

[00:11:15] Ari Paparo: It was just going to disappear. Um. So that's the lens that I would look at. Your question. Uh, is there a similar play, um, that would, that would make a retailer, um, you know, unable to make proper decisions, or at least just preference one tech stack enormously because it came with revenue as opposed to just coming with technology? 

[00:11:39] scott messer: Yeah, and I think the answer today is no. But Tom, do you wanna to explain a little bit about. The different demand sources, what? We'll, we'll take direct out where you're getting it from your JBPs just coming in direct. But what's out there for like programmatic revenue for these commerce brands. 

[00:11:55] Tom: On the middlemen. We mostly looking into sort of the multi-product retailers or the multi-brand [00:12:00] retailers and in their situation, you know, it's interesting, some of the things that you said there, um, sort of rang true, like, oh. Maybe the retailer could somehow, uh, if they own their own DSP, could be this sort of monopolist, uh, if they wanted to, because they own this measurement capability for the transaction layer that they have at their point of sale. 

[00:12:19] Tom: And then, you know, when I was working in this world at Quotient, which was originally coupons.com, the whole point of it was to join that transaction log. To, an identity spine, which no retailer had done yet. Um, and so that was a world where, you know, the targeting in retail media has always been sort of behind. 

[00:12:38] Tom: It was always just the measurement that was something that you couldn't get at otherwise. And so that's sort of a, where I would sort of level set in terms of the commerce slash retail media world. But what is out there in terms of demand, I think, you know, historically you see things like the rise originally. 

[00:12:55] Tom: Of, uh, triad, which was all about, you know, just supporting the Walmart demand. [00:13:00] Then you had somebody like Criteo come in and say, you know, once, once retargeting became sort of a, a, an albatross for them. We're gonna really lean into our hook logic acquisition and do, performance based CPC auction model for. 

[00:13:14] Tom: Uh, sponsored listings or product, uh, listing ads, as you mentioned. And so that was the beginning of it, but they were very much a network model. And, that's where some of the retailers now are, are moving off into these newer ad servers that, unbundle the demand and the, the supply. 

[00:13:29] Tom: Pipes. Um, but we're seeing, you know, from Scott's discussion that there's orchestrators coming in to try to tie it all together again. And so, , there are many fronts to this 

[00:13:38] Ari Paparo: There are a couple of interesting other dimensions here, which is, uh, first the lack of standardization. So, um, at the SKU level, the SKUs on Dick's Sporting goods aren't at Kroger's, like most of them are not. So there's no network effect of having those two retailers in the network together. 

[00:13:54] Ari Paparo: There's a little, obviously they, but they both sell batteries. They both sell like stuff. Um,[00:14:00]  

[00:14:00] Tom: I mean, I think Criteo was big on like a unified product catalog and the idea of looking at it across some of the retailer and they had like 225 retailers. But yes, you're right. There's certain, there's definitely different, the Venn diagrams are not 

[00:14:12] Ari Paparo: No, I mean, but they're really far from circles, right? So you go by vertical and, um, and there's no commonality other than a couple of commodities. So like the only category that has scale is grocery, in my opinion. Any other category has like a very limited number of head-to-head competitors who have no interest in, in cooperating, um, and then groceries regional and has a. 

[00:14:35] Ari Paparo: Lot of SKUs. So grocery is kind of like the perfect example where it should be able to have some network effects and there would be a, a, a value add both to the buyers and the sellers of having a network, um, in other areas. Um, you may be talking about just like a net negative, where like, once again, if we do Dick's Sporting Goods, I just, they came to mind like, you have a sporting good manufacturer. 

[00:14:58] Ari Paparo: Um, probably, [00:15:00] um, Dick's is a pretty important retailer to them. They're willing to do what in integrate into whatever sporting, um, API they have. Um, there's not much benefit to having the same system for that and other retailers, like there aren't the many retailers, you know, Walmart, a couple of others. 

[00:15:17] Ari Paparo: As a buyer, why pay, you know, some significant percentage of your spend to an optimizer? Um, the other question, and I think this is very much what I would be thinking about if I was in this business 'cause I'm not really is, um, you know, how much optimization is there? 'cause it doesn't feel like there's a lot of optimization. 

[00:15:35] Ari Paparo: It feels like match the SKU and you show the PLA and if your budget runs out, you stop. 

[00:15:40] Tom: Well, I mean, I think this is. Other big question the omnichannel area is very different in this world where, , and I think this ties very well to the discussions in ad tech, CTV is starting to become a really big part of retail media. 

[00:15:54] Tom: And it wasn't really for the, the, you know, the sort of the publishers of the world. There, there was, there were [00:16:00] missing pieces of the ad tech there where people weren't logged in, in both systems. And it seems like CTV is a, you know, chocolate and peanut butter in retail media. Um, but that plus the in-store activation plus, you know, the fact that you have the user's identity, there's so many more variables in retail media, whether or not they're being, you know, used effectively. 

[00:16:21] Tom: Doesn't that create more opportunity or is it, you're saying you're still in a sort of low optimization. 

[00:16:26] Ari Paparo: It's a good question. Um, so, um, so I'll try to address your ques your points one at a time. So the CTV to retail opportunity. Question is, is it just gonna belong to Amazon and Walmart? Like, is there a play for, you know, once again, Dick's Sporting Goods in CTV or they, they would, it would basically have to be audience extension, and then who's taking the risk on buying that CTV ad that features the product that's on there. 

[00:16:49] Ari Paparo: Whereas Amazon with their media properties. 

[00:16:52] Tom: Yeah. 

[00:16:53] Ari Paparo: Could take that risk and has the measurement. Um, second point is, um, personalization there, to [00:17:00] my understanding, there's virtually no personalization going on in the entire retail media world. It's almost entirely SKU based and context based. I, I, I think that's, I talked to a couple of the buy side tools. 

[00:17:11] Ari Paparo: I won't say who. Um, and, and they don't support cookies or IDs at all. Like, it's just not part of the stack, 

[00:17:16] Tom: I think, I think that they haven't had to, to date. And it, it was sort of the dirty secret that personalization wasn't necessary because you were getting the money anyway. Um, but it is interesting talking to the retailers now, um, guys like Bobby Watts ahold who were saying, you know, and they just took their publisher ad server in-house and they were saying, we're looking at doing predictive audiences. 

[00:17:36] Tom: And I was like, good, good job guys. Like, I, you know, I thought you would've been afraid of that. 

[00:17:39] Ari Paparo: I guess what I would say is like the, the PLA market is so, um, dumb, right? I mean, like, think about it, it's like, like I have a list. I have a list of SKUs. Each SKU has a budget and then the page loads and has a SKU in it. That's a pretty easy technical problem. [00:18:00] Right? Um, so why do we think that the same systems that are built around PLAs are gonna be good at personalized top of funnel advertising? 

[00:18:11] Ari Paparo: Anywhere, CTV website, mobile push notifications. The answer is, is not, they have nothing in common. 

[00:18:17] Tom: Yeah, I think the ad servers who are gonna do well in this world are ones that can go beyond just the simple PLA ads. 

[00:18:24] Ari Paparo: Well, if there is such a thing as an ad server in any of these worlds, right? 

[00:18:28] scott messer: it, it's an interesting question, like, I don't know if DSPs are up to the job that, uh, onsite ad servers could do. Um, because as dumb as that PLA ad is like looking up the SKU and understanding the contextual nature of it is actually like a very personal thing to that website and is best served by the ad server, which is plugged into the, the publisher page and sort of runs its own optimization algorithms. 

[00:18:56] scott messer: Um, and like a DSP. [00:19:00] Isn't really ready to do that. There's, they can say like, yes, I know that SKU and I want to place an ad for it, but how, how long will it take them to get to the point where they're buying PLAs in the same way they're thinking about buying display ads across the open web? And I think that's really far away. 

[00:19:16] Ari Paparo: Yeah, I, I have to agree with you. I, I think that none of the legacy ad server or ad tech stack, including DSPs, are really oriented towards the SKU use case. Um, you know, ad servers are used to contextual targeting, where you have, you have, you know, millions, sort of an infinite number of web pages, but they're boiled down to like 10,000. 

[00:19:39] Ari Paparo: Key values. And there's no such thing in the ad server world as like a similarity key value. Like, hey, this is about football, so I should also serve rugby ads. That doesn't exist. You can't do that easily. Um, and DSPs are also kind of, DSPs have a better ability to sort of have an infinite catalog. 

[00:19:59] Ari Paparo: Um, [00:20:00] but they require a lot of. Um, custom algos to do anything like similarity scores. 

[00:20:06] Tom: You, you actually built a company on custom algos, isn't that right? 

[00:20:09] Ari Paparo: Yeah, so I built beeswax when the unique selling prop of beeswax as A DSP was, uh, among other things that you could have very customized algos that you could write in code, uh, and put in your own server. So in our world, we would say, well, you know, pass in a variable, which you. 

[00:20:26] Ari Paparo: The web is usually the URL and then you can do whatever lookups and calculation and give us back either a score or matching creative or whatever it is. Um, but still, you know, at the scale of a large retailer, um, some of that breaks down and doesn't work very well, and it's not really suited for purpose. 

[00:20:44] Ari Paparo: You'd probably have to build some new tech to really make that work. 

[00:20:48] scott messer: And that's what I think the new ad servers are. Um, they've moved that sort of decision logic and the algorithm into the ad server. Like it doesn't really exist in Google Ad Manager. Google Ad [00:21:00] Manager just runs one big auction and it does some campaign flighting. These new, this new crop of ad servers that are competing have taken that decision engine and they store the algorithms and the models inside of their own systems that they're doing exactly that for the 

[00:21:17] Ari Paparo: Who are we talking about here? Are we talking about like Kevel, Moloco? Like what? What we talking about here? We. 

[00:21:22] scott messer: We promised not to, not to name names in this episode, but you are more than welcome to continue repeating 

[00:21:28] Ari Paparo: I don't have, I don't have a long list. Um, 

[00:21:30] Ari Paparo: but.  

[00:21:30] scott messer: It'll be top sort, Koddi. 

[00:21:34] Ari Paparo: Mm-hmm. 

[00:21:35] scott messer: Zitcha, Moloco as you've said, Kevel, it's a, it's a small list, but it's growing and I think we're seeing more companies with their, their product in one corner start to move into becoming an ad server because the fighting part is really easy campaign management stuff. 

[00:21:53] scott messer: And then it's about do you have a secret sauce that helps power these retail media networks? 

[00:21:57] Ari Paparo: Mm-hmm. 

[00:21:58] scott messer: And they all kind of think that they [00:22:00] do. Um. 

[00:22:01] Ari Paparo: Mm. Yeah. So uh, my understanding is, and maybe you guys, you guys certainly know better than I do, is like, Moloco's approach is like a network model. They're like, you know, ad sense ad words for retail, um, where you just like put it all in one system in the best ad shows. Uh, whereas like Kevel like more like a DFP approach where they have APIs for building an ad server. 

[00:22:21] Tom: I think that's right. I think, you know, from the Moloco episode that we just had a few weeks ago, John Flugstad talked us through the idea of actually looking at consumer journeys. Um, and so taking some of those types of behaviors, um, I don't know if it also includes behavioral stuff that people are doing elsewhere. 

[00:22:39] Tom: Um, but yeah, generally trying to do things that would potentially enable personalization. So directly to your point, like, you know, hey, retail media isn't doing what it needs to do. And honestly, I've known that from the beginning. It was, we were always doing, we had such an easy job because it was captive, uh, audiences and captive brand budgets.[00:23:00]  

[00:23:00] Tom: Um, but moving, you know, the, if, if the pressure is on to make retail media, all media, they're gonna have to go through all this same stuff. Um, and you know, I think the questions that we would have for you did Google get to a place where, they innovated enough to be a monopolist, but the publishers weren't requesting anything more from them to innovate anymore? 

[00:23:25] Ari Paparo: Well, yeah. So Google hit a point in its journey where, um. It became impossible to switch off of the ad server. And uh, this was found in, in court. Uh, there was evidence they had a churn rate of under 5%, um, which is like crazy for a software company. Uh, yeah, it was, it was, yeah. 

[00:23:47] scott messer: left. 

[00:23:49] Ari Paparo: Exactly. 

[00:23:49] Ari Paparo: The quality of GAM as an ad server versus the needs of the, of the marketplace is an unclear question because if you talk to publishers, of course they love complaining about it. [00:24:00] Um, but it's not clear that if, if there was a competitive environment, that there would be better options. 

[00:24:05] Ari Paparo: Probably cheaper options, especially in the ads. At exchange side, they took a pretty high take rate. Um, but uh, on a feature for feature basis, GAM is pretty comprehensive. Um, so, so there's a lot of sour grapes out there. So it's unclear really where we ended up in that environment. Um, but I think it is, um. 

[00:24:23] Ari Paparo: It is very clear that the retail media opportunity passed by Google. Um, I mean, Google Shopping is obviously a big business. I don't understand it. I don't know who uses it, but it exists and is big. Uh, but uh, they're not, they're not really, they're not in the business we're talking about. Um, they're not buying feed ads on other retailers using AdWords, which would seem like a huge opportunity for them. 

[00:24:46] Ari Paparo: Uh, they're not powering feed ads on retailers. 

[00:24:50] Tom: Yeah, I mean, they sat next to me, they came into our meetings at Albertsons 'cause GAM existed at Albertsons and they, they would try to unseat us as quotient. Um, and they just, they, they showed [00:25:00] Albertsons their PLA ads, but that was, it wasn't enough. 

[00:25:03] Ari Paparo: and their PLA ads, it was not really a SaaS offering, right? It would be like just a. A way to make money. Um, you know, so that's interesting. I mean, Facebook's also not in this business, despite having the feeds and the catalogs in their system and using them for ads, they don't work on third party sites, to my knowledge. 

[00:25:22] Tom: Yeah. I mean, the only way I see this happening is somebody buys a transaction engine, so it would have to be like Facebook buying Shopify or something like that. I don't know. 

[00:25:30] Ari Paparo: It's, uh, Shopify is obviously a huge opportunity here. So does PayPal, um, so does Amazon, um, in all three cases though, they, they have, what, what I, I'll step back in one, say, you know, why was Google able to run the play and display ads then Facebook didn't. Um, and that, I talk about that in the book, but the main reason, the easy reason is that Facebook had a ton of o, O&O and they didn't need the network. 

[00:25:54] Ari Paparo: Play that Google did. So Google in history, before YouTube became what YouTube is, just had [00:26:00] no O&O, right? They had tons of ad demand and they had tons of search supply, but they had no display and no video. So the network play, meaning aggregating many, many publishers, giving them tech, et cetera, was the way to monetize. 

[00:26:12] Ari Paparo: In the retail media world, um, I think, uh, or in the, in this sim related world, catalog ads, retail media, whatever, um, everyone has a lot of, O&O, at least by everyone. Facebook obviously not lacking for opportunities to sell ads. Amazon, you know, of course, et cetera. So, um, so why would you take lower margin, more complicated, you know, annoying, uh, you know, potentially conflicted clients? 

[00:26:42] Tom: It makes sense. Yeah. Why would you have all these, like the concept of a joint business plan or your annual planning with your, your brands? Like why, why do all that if you don't need to? Yeah. 

[00:26:51] Ari Paparo: Yeah. Whereas if you're a startup, if you're an ad server company or a DSP type company, this is a huge opportunity and it's kind of surprising more haven't, you know, invested [00:27:00] here. Um, they, I think the PLA thing, um, kind of turned a lot of people off because, um, the history of, uh, search engine marketing. 

[00:27:11] Ari Paparo: Ended poorly, where like the two leaders, Marin and Chu, both basically Chu didn't go outta business. They turned into the sky, but they certainly abandoned the search side of things. Turned out to be a terrible business. So PLAs, if you squint right, looks a lot like. SEM, um, and you know, if, if I was an executive at a company, let's say the trade desk five years ago, and someone said We should really be bidding on PLAs, I would say, how? 

[00:27:36] Ari Paparo: Tell me why that's not gonna be a very low margin business where we are squeezed out by the retailers. Um, and I don't have a good answer to that. That's. 

[00:27:44] Tom: No, but that's good context for understanding. I've been racking my head, why, why isn't BloomReach and, uh, you know, uh, country Sky, what, like, why aren't those guys coming in and just offering what Criteo offers? And 

[00:27:57] scott messer: Why aren't they ad server, which is why we think that [00:28:00] they probably are, or they should, or they're gonna listen to this and say, gee, maybe that's a good idea. 

[00:28:05] Ari Paparo: when you say ad server, you mean? I, I think it's probably helpful to explain what you guys mean. You're saying a sell side solution, a software product you would sell to retailers to enable, you know, enable ads. That's, that's the opportunity that you're trying to push on. 

[00:28:20] Tom: Yeah. And, but it might fall apart in minutes because we're moving to a world where longer form context kills the, the keyword as the, the linchpin of, of advertising there. So 

[00:28:32] Ari Paparo: Sure. 

[00:28:32] Tom: I don't know if I would've suggested today, I would've suggested it two years ago, maybe five years ago, but. 

[00:28:36] Ari Paparo: Yeah. Yeah. Uh, it is an interesting, uh, question about like, if I am shopping on Open AI chat, GPT and the feed of products is coming from that retailer, um, there's no, not as much of an advertising opportunity to, uh, to provide a different product or to, uh, suggest 

[00:28:56] scott messer: I wanna click into this transparency thing real quick here. I'm gonna take this [00:29:00] long segue. Um, something we say on this show, and I repeat from a Ari Paparo quote, is that. Um, log files are a constitutional right, uh, and surprisingly these retail media networks aren't always handing out their log files. 

[00:29:15] scott messer: Um, and I think that, I don't know if all the brands understand what it means to have access to log files and what they could be doing with them and how they can hold their retail media networks accountable to them. So I'd first, like you do, is explain what do you mean when you say log files are a constitutional right. 

[00:29:30] Tom: Yeah, let, let's be fair. Explain it to me. 'cause I think Scott gets it. I don't, I. 

[00:29:35] Ari Paparo: Well, okay, so fundamentally, if you're bidding for, um, a piece of inventory from a publisher, and the only thing you get is the aggregated data, meaning I spent a hundred bucks. These are the dates, these are some additional variables. Um. It's, it's very, first of all, it is very hard to forensically tell if that's true. 

[00:29:57] Ari Paparo: So there's a, there's a true or [00:30:00] false or fraudulent nature of the inventory. What was the exact page? When did it show, et cetera. You really can't tell when you just get the aggregate data. Um. And secondly, you can't really tell if the price was right and, and the argument, some people make the argument, well, you were willing to pay so you paid, and who cares about where the price is? 

[00:30:19] Ari Paparo: But that's not really a fair point of view in the complex display and video advertising world where there are multiple middlemen and you may not be a hundred percent clear which middlemen you went through and what they paid. And there's a lot of opportunity for optimization. There's a lot of opportunity for, uh, perverse incentives from the people on the chain, including the vendor that you're working with. 

[00:30:41] Ari Paparo: You're working with the DSP, they might be manipulating your bids to go through a path that gives 'em kickbacks, all these sort of things and the. Easiest and best way to solve all these problems is to get the raw log files and have an, an ongoing analysis that tells you, uh, when things look a little funny. 

[00:30:57] Ari Paparo: Um, and if you don't have the log files, you could [00:31:00] almost be sure something funny is happening. Um, now in retail media, um, it may be less important, um, because the, um, opportunities for fraud, the opportunities for, uh, take. Manipulation are lessened because there's less of a supply path. Obviously I would still want it. 

[00:31:17] Ari Paparo: Um, but especially if I'm only one hopping. If I'm like basically buying ads on a given retailer and I already sell products on that retailer, I don't really need that much logging. I need to know what I paid and how many people bought something. 

[00:31:31] scott messer: about when they do off network audience extension for you? 

[00:31:34] Ari Paparo: Yeah, off network. I would demand it, uh, because I don't know where the, I would want, at the very minimum, you need to know the full URLs of where your ad served. 

[00:31:42] Ari Paparo: Uh, that's like a very important, uh, thing to have. Or else you're, you're guaranteed to be showing your ads in bad places, 

[00:31:49] scott messer: This brings up an interesting question, Tom, if you had to guess like how many brands do you think look at anything beyond aggregated reporting [00:32:00] with their retail media networks onsite and offsite 

[00:32:03] scott messer: percentage 

[00:32:03] Tom: mean, I think it's only, only, I think it's only the top brands that are spending the multiple, multiple millions on these 

[00:32:10] scott messer: already have like big, large established media orgs that are used to crunching this. 

[00:32:15] Tom: I mean, I think I, I think, you know, ones that have brought in agencies obviously are, you know, they have center of excellence around, blacklist, whitelists, , , that, that whole game of, managing, uh, how, what level of quality can we get by  

[00:32:28] scott messer: Yeah. And then how many, second question, how many brands do you think. Ask the retail media network what the off network stack looks like and like what DSP are they using and how are they going to market to buy those things? Do you think they ever look at it or they just say like, okay, let's run the money. 

[00:32:44] scott messer: run money. 

[00:32:45] Tom: A actually, I think it was, it was very easy to know back in the day when I was working on it because, uh, there was always 1D SP. Um, now it's gotten more complicated. So somebody like a Costco may have like 5, 6, 7 DSPs [00:33:00] going on, so they might not know. 

[00:33:02] Ari Paparo: Yeah, I, I guess I'd ask why they care. Um, you know, on the stack, I care about the sites. Uh, um, I, I don't really care about the margin if I'm, uh, we're using this as a performance metric, um, but I, I care about the sites and the, and the quality a lot. Yeah. And 

[00:33:19] Tom: Yeah. And, and to be fair, you know, the retailer is bringing their own data margin in here, and so the quality of the site usually suffers 

[00:33:29] Tom: unless there's, you know, some more transparency. So I think, I think that's fair. 

[00:33:33] Ari Paparo: It's, it's almost axiomatic that if, if someone's not looking at the site list, the site list is gonna be bad. 

[00:33:38] Tom: Yeah. Because you can get away with 

[00:33:40] Ari Paparo: yeah. Um, it, it kind of begs the question about whether. There's, why are we doing this? Like why is audience extension through retail media even a thing, 

[00:33:51] Tom: I mean, for me, if it were, if it went all CTV, I'd be happier 

[00:33:55] Ari Paparo: If they owned it. Like, I, I love, I, I tweeted this like a week ago when I was [00:34:00] watching, uh, Thursday night football on Amazon and just like every single ad had a call to action. And I just love that. Um, because it's Amazon's media, Amazon's DSP, um, Amazon's data. I love that. That's a great use of retail media, especially if it's priced on performance in some way, which it probably isn't, but it'd be, especially if it is. 

[00:34:19] Ari Paparo: Um, but like the compare that to like, you know, some, uh, your retailer, your, you work with 10 retailers and one of them has a trade desk account and is showing your banner ads to random people. And like you have to figure out if it was actually worth it or if it was duplicated against your Criteo and your other stuff. 

[00:34:39] Ari Paparo: It's like. 

[00:34:39] Tom: So do we need to revise Terry's, uh, idea about who the trade desk should buy? They should be buying a retailer. I mean, what, you know, 

[00:34:48] Ari Paparo: well, that you just, you just did a pretty big u-turn on me there. Um, so is it, so do you wanna explain to your audience what you're talking about or do they know? 

[00:34:56] Tom: well. Well, they might know if they watch the Moeloco [00:35:00] episode. We, we, we, uh, in homage to you guys, we stole your clip of Terry Kja and Architecture saying that, uh, that, that the trade desk should buy Moeloco. Um, and so that's, you know, getting into sort of performance, but it sounds like what we'd be more interesting is getting them on tv, which is, you know, my wife is very upset that she's seeing so many trade desk ads on Hulu these days 

[00:35:23] Tom: because they're targeting me.  

[00:35:24] Ari Paparo: I see a lot of trade desk ads. I see a lot of ads for all kinds of things. Well, I, 

[00:35:28] Tom: It's really getting me on the hard strings. I don't know, 

[00:35:30] Ari Paparo: I think there's two different things when we talk about the trade desk, which is first. Should they be in per more performance settings? Um, and secondly, uh, the channel. Um, so on the performance side, I, I think it's a fair criticism that the trade desk is not the best, uh, cutting edge performance company out there. 

[00:35:50] Ari Paparo: Um, I, I don't know how well they're. Perform on ordinary campaigns, but it's not what they're known for in the way that like Moloco or App Lovin or TV Scientific are known for that. [00:36:00] Um, so that would be clearly an area I would look if I was them. And the second question is by channel. You know, CTV performance is different from, you know, retail media performance. 

[00:36:10] Ari Paparo: Um, so those are all good options. I'm not, I'm not, I wouldn't say not to do Moloco. I think Moloco a great company. 

[00:36:16] scott messer: But I think to answer the obvious question you asked was, why are we doing it? And there's not enough O&O inventory to scale massive media budgets across these retailers. 

[00:36:29] Ari Paparo: Well, that's why retailers want to do it, but I'm asking why should, okay, so let's just do a scientific, a, uh, a thought experiment, right? Uh, which is, and, and the example I gave, I, I'm a brand, I sell across 10 retailers. I also buy media myself. Um, and so the retailer comes to me and says, in addition to the PLAs, which we know are great, uh, we're gonna take some of your budget and buy banner ads for you across the internet using our data. 

[00:36:55] Ari Paparo: Um. Should I do it or not? As a brand? [00:37:00] The, the, the positive is it's some, it's some signal I don't have otherwise. Right. That's the main positive. 

[00:37:07] Tom: Well, I mean the negative has been that if everything's done on ROAS or I roas, then those banners, uh, you know, unless they're on a mobile device. Are hard, it's hard to tie that in enough of a cost effective way to, to roas. And so that, that's been challenging and that was the pain point that we had. 

[00:37:26] Ari Paparo: Yeah, so you, you're basically, you're expecting the user to see a banner ad for like some mayonnaise and click on it and then go back to the retailer that where they abandoned mayonnaise from the shopping cart and then buy some 

[00:37:38] Tom: Yeah. Well the truth was, and this was, this was sort of the innovation, which was just the exposure was enough if you had a transaction log, um, and so you didn't have to click. Um, and, but you never got to a place, and this is why I got excited about attention companies and things. But how do you level the playing field where. 

[00:37:55] Tom: It's not just the cheapest inventory that performs the best. So I think this comes back to [00:38:00] the Google discussion, which was search ads were so good, so easy to measure, so easy to take credit. Um, the next level up in retail media for us was mobile ads. Um, but unless you level the playing field of measurement, um, so that it doesn't always, you know, award the, the sale to the, the la you know, the last click. I, you know, you're, you're right. It, it doesn't make sense to do audience extension because you can't measure it well enough. 

[00:38:26] Ari Paparo: And you're really, uh, if it's a call to action of go back to the retailer, you're really just subsidizing their retargeting. Yeah. 

[00:38:34] Tom: Not gonna argue with you there. 

[00:38:35] Ari Paparo: Yeah. 

[00:38:35] scott messer: is the like. 

[00:38:36] Tom: This is where, the two goals of the retailer and the, and the brand. This is what they talk about in their joint planning sessions, which is, you know, the retailer obviously wants category sales. 

[00:38:46] Tom: They don't really care about brand 

[00:38:47] Tom: sales. A brand wants brand sales, but if they both are pushing traffic, then they hopefully both win. Um, but, you know, you know, it's not, it's not as strong an argument, it's just a brand centric 

[00:38:58] Tom: argument.  

[00:38:59] Ari Paparo: Right. [00:39:00] Yeah. Interesting conversation. I'm, I do wonder, I mean, obviously there's some value there. If I was a brand, I would say like, get me people who you think would buy mayonnaise but don't currently buy mayonnaise. I don't want to spend any money on my own customers. 

[00:39:13] Tom: Yep. No, I think that's fair. And that's sort of the predictive audience concept. I mean, now that there are GLP ones, you know, you can't target those people from 2-year-old audiences. And actually, this, this might be a good time to ask you the, the AdCP question. Um, 

[00:39:28] Ari Paparo: For you went from GLP one to add CP in one sentence. That's like that, that's a very, uh, flexible brain you've got right there, 

[00:39:37] Tom: Well, I mean, the reasoning is that the GLP one customer, so to give you some context in retail media, pretty stale audiences up till now. Basically it was look for category buyers or brand buyers. Those audiences might be three, six months, even two years old. Um, but you can't do that, uh, if it's a GLP one product that, you [00:40:00] know, like somebody's buying a protein yogurt or something like that, um, they weren't doing that before. 

[00:40:05] Tom: So how do you predictively find those people and find the things that they might want to buy? Now, maybe you don't even know if they're g GLP one user, but how do you predict the trends of those new products and  

[00:40:15] Ari Paparo: Yeah. Are you hearing this from actual customers of yours that they're seeing, like drop off in effectiveness of like high fat foods and increases of like high, high protein foods and things like that? 

[00:40:27] Tom: I mean, that's, that's what the retailers were talking about. All, all grocery shop  

[00:40:30] Tom: with me, but  

[00:40:31] Ari Paparo: Yeah. Right. All right. Well, 

[00:40:34] Tom: And then the reason I, I jumped to AdCP was that, you know, you had an in, I mean, the only thing that I've read on the internet that gave me sort of a feel for what AdCP could do. Was the idea that deal IDs were a little stale too. 

[00:40:49] Tom: They were prepackaged. Um, but potentially using AdCP you can get to real time understanding of audiences. 

[00:40:57] Ari Paparo: Real time. Yeah. Um, I think [00:41:00] AdCP, um, is really a protocol for letting agents, you know, interrogate and discover audiences and potentially transact them. Um, and so in the world of retail media trying to bring it to that, um, you could imagine that a retailer might open up a, um. An MCP server that lets the buyer mix and match and discover, um, based on segments and say like, you know, find me, Hey, let's do these, the GLP example, find me people who were buying lots of fatty foods a year ago and aren't anymore. 

[00:41:35] Ari Paparo: Um, maybe. Uh, I'm just kind of speculating and, and you know, you could do that in a query environment or a, uh, ad ops environment, but the ability to do it, um, arm's length is the benefit of AdCP. Um, but you still have to have your data. AI doesn't solve the problem of having the data and making sense of the data. 

[00:41:55] Ari Paparo: It just enables the discovery Once they, the segments have been made, um. [00:42:00] Right. Um, and to some extent de taxon mizes it. Um, so instead of like a hierarchical approach where, you know, you have Oreo buyers underneath cookie buyers, underneath, you know, whatever, um, you might be able to ask the AI unstructured questions like the one I gave or other ones like that, um, which is exciting. 

[00:42:20] Ari Paparo: But once again, you need a lot of data. You need to have it organized before any of this works. 

[00:42:26] Tom: And that organization, that tax sonification is not, is not conducive to this 'cause you're basically saying, let me cherry pick, right. 

[00:42:32] Ari Paparo: Yeah.  

[00:42:32] Ari Paparo: Yeah. I, I think there's a more in retail media. I'm, I'm talking totally off the top of my head here. Um, I think there's a bigger opportunity in AI for changing the ads themselves and making the experience better than targeting. Um, you know, I, uh, I'm a, in. Investor, small investor in Marpipe, which is a company that does, um, like retail feed enhancements, using ai, you know, make 'em prettier or, you know, offer different options. 

[00:42:58] Ari Paparo: I think that's a very [00:43:00] promising direction for us all to go. And, and, you know, PLAs just suck. They just look bad. They're 

[00:43:05] scott messer: that a personalization opportunity or is, is that, is that contextual or is that personalized based to the, to the user? 

[00:43:12] Ari Paparo: Those are all options. I, I won't speak from our pipe specifically, but if you just go from a, from the feed that you have lying around your CMS to a feed that's gonna work and perform in ad environments, there's some changes you need to make in terms of the quality of the photos, the text messages, you know, seasonality, put a little Christmas tree behind things, you know, stuff like that, that AI can do at 

[00:43:33] scott messer: So, so in honor of the platform war or premise here, like, and we'll just use this marpipe as an example if you want, but like when do they come up and they say like, actually, we could be your entire ad server. 

[00:43:44] Ari Paparo: Once again, I'm not sure we need an ad server. So like you guys are really stuck on this ad server thing. You know, buy side doesn't need an ad server at all, right? They just need some metrics and some, some reporting and maybe log file analysis. Sell side needs a lot of different things, none of which [00:44:00] really are an ad server. 

[00:44:00] Ari Paparo: They really, they need some sort of like sell side really needs to understand data segment users enable usage of that data in various places. 

[00:44:10] Ari Paparo: And the That's an ad. 

[00:44:11] Ari Paparo: is 

[00:44:12] scott messer: That's Ad Server does for For 

[00:44:14] Ari Paparo: Maybe not if they don't control the media or if the media's limited. 

[00:44:18] scott messer: Yeah. 

[00:44:20] Tom: I think, I think maybe that's, uh, that's where we call it because this was a great discussion. I don't know how to push it any further without, uh, saying that I have a flexible brain or falling a foul of a, of, of Ari's thought processes. So, but I really appreciate, this was a great discussion. I think we're gonna have a lot of fun putting this out, out into our audience and, you know, we'll hopefully we'll get some feedback of. 

[00:44:41] Tom: You know, we need, we need to rename our series. We're, we're up for that 

[00:44:45] Ari Paparo: Well, thanks for having me. That's a great conversation. 

[00:44:47] Tom: all. 

[00:44:48] scott messer: Thanks, Ari.  

[00:44:49] ​