# S3 E15 T70 - Regina Ye - Topsort

S3 E15 T70 - Regina Ye - Topsort 

[00:00:00]  

[00:00:00] Tom: You are listening to the Middlemen podcast  

[00:00:03]  

[00:00:07] Intro & Guest Overview 

[00:00:07] Tom Limongello: Welcome to The Middleman Podcast. Hello, Scott. 

[00:00:12] Tom Limongello: It's good to see you. It's been a while. We've been traveling a lot. 

[00:00:16] Tom Limongello: Yeah. Hopefully we'll be at something together again. It's been since Marketecture. but today we're gonna talk to Regina from Topsort. Interesting company, because If you had to say who they were, you would probably say they're a publisher ad server, but they are contending that they are something different. 

[00:00:35] Tom Limongello: And I think it makes it, you know, very easy to sort of think about where the market is going by the way that they characterize themselves. They're, they're saying that, the decisioning that they offer within their system and their, their belief in auctions , is bigger than just, publisher ad server. 

[00:00:53] Tom Limongello: What are your thoughts there? 

[00:00:56] Scott Messer: Certainly something I agree with and it's been fun to watch Topsort grow over the, the [00:01:00] last few years and sort of emerge as a, as a player in the ad server space. Although Regina would not like us, describing them as an ad server. But I think for our audience, we should think of them as an ad server because that's their predominant function. However, to, to that point, 

[00:01:15] Scott Messer: Auctions are hard, and the logic for them is complex. You can obviously construct them just from putting a bunch of line items together and letting them run in price priority. But once you start bringing like outcomes into the mix, it gets really, really complicated, and you might want a specialized piece of software that focuses just on running and optimizing auctions. So from that standpoint, Topsort is going after this ad server market sort of through the auction. They think everything should be an auction, and they're probably right in a lot of cases. And it's really interesting to hear Regina's perspective on why and how they accomplish this. 

[00:01:57] Tom Limongello: Yeah. And it's also interesting hearing, um, the founding [00:02:00] story of the company, that they have a lot of experience in, Australia and Latin America, and, they get to work with retailers that have different specialties and different resources and capabilities. My experience has been predominantly in the US, and the retailer is always this resource-constrained company that really doesn't have a lot of capabilities until they work with these outside vendors. 

[00:02:25] Tom Limongello: And I think she's seen the opposite of that, where, you know, they can really go fast and, and, you know, develop a lot of really sophisticated capabilities. Um, so, so that part of it is interesting. And I think also what's interesting is sort of this debate as to whether, uh, loyalty programs and first-party data are the most important thing, or is it the moment You know, is the advertising moment that somebody is, looking for a cheeseburger, is that, uh, the more important context than, the fact that you're in market for a refrigerator? 

[00:02:59] Scott Messer: [00:03:00] think the, the word context there is the one you were delicately dancing around, but that is the contextual moment of it, and they're strong believers that it's context over audience. And as somebody who's done a lot of, training and learning work for personalization algorithms on the publisher side, I can tell you that audience data isn't always the best indicator of what's next, that contextual usually is. Where the user is in their shopping journey and where they've come from and where they're going, that really is an enormous driver. So it's fun to see her talk about, not using data, although we do get to a point where there is an identifier in there, and you have to do some stuff like that. 

[00:03:42] Tom: I think where I would push back a little is Regina isn't really talking about contextual advertising. Context means page content or referring URLs, and what she's really talking about is search intent. That can be sometimes a much stronger signal than audience segments that were built six months to [00:04:00] two years ago. 

[00:04:01] Tom: To me, in my mind, that doesn't make loyalty or purchase history useless. Those signals are really very important for understanding brand affinity and repeat behavior. But it's an interesting tension, especially when you're thinking about the sponsored search, uh, on-site context. Which signal is gonna matter most in those moments? 

[00:04:20] Tom: Um, so it's, it's a good one to think through. And so with that, let's, let's get into the episode. 

[00:04:26] Tom Limongello: . 

[00:04:26]  

[00:04:26] Introducing Regina & Topsort 

[00:04:26] Tom Limongello: We are here with Regina Ye. She is the CEO of Topsort. Regina, thank you for coming. 

[00:04:33] Regina Ye: Thanks for having me. 

[00:04:35] Tom Limongello: So Topsort, uh, for the listeners who don't know the company, where are you based out of? How did it all start? 

[00:04:42] Regina Ye: Yeah, so, um, Topsort actually started in 2021. We're technically headquartered in Palo Alto, but when we first started it just. Didn't occur to us that COVID would end one day and we would actually have to figure out logistics to be in one place. So we started having a global team and [00:05:00] a global customer base from day one. 

[00:05:02] Regina Ye: Um, myself, I'm based in the Boston area on the East coast and my co-founders are both on, uh, the West coast uh, today we have team members in more than 10 countries. 

[00:05:12] Scott Messer: What's a quick overview of what is Topsort for our listeners. 

[00:05:16] Regina Ye: Yeah, so a Topsort is what we like to call AI monetization infrastructure for modern commerce. Um, that includes retailers, marketplaces, delivery apps, and increasingly we're getting into more of the brands and agency space as well. 

[00:05:31] Scott Messer: So that's a fancy, fancy terms for an ad server, but obviously goes much further than a traditional ad server might, as our listeners might think of. Is that correct? 

[00:05:42] Regina Ye: Um, yeah, I think ad servers one of the things that we do, um, but I feel like we might get into it, but like, 

[00:05:49] Tom Limongello: Yeah, we, 

[00:05:50] Regina Ye: feels 

[00:05:51] Tom Limongello: we can get into it. Now. I know that you bristle at the publisher ad server moniker, so let's get into it. Tell us what, what's wrong with the [00:06:00] industry? 

[00:06:00] Regina Ye: Yeah, I feel like, so  

[00:06:01] Ad Engine vs. Ad Server 

[00:06:01] Regina Ye: Ad Server it's a very, um, it immediately brings you back to the programmatic era and like thinking about publishers, um, and the websites. But really nowadays, when we think about retail, media, commerce, media, so much of it is really, um. Different from running a website, like a blog. So then, um, we really think of, um, Topsort more of an ad engine where it's not just thinking about the logic of serving the ad and seeing the ad being delivered as the outcome, but more that, um, we want to think about all the variables and like the transaction. the commerce nature of it. Um, and to do that you need to do sort of a lot more invisible work than just putting on the ad because you actually are kind of responsible for the outcome of the, the ad as well in in many ways. 

[00:06:50] Tom Limongello: we're obviously a, a podcast that talks a lot about retail and commerce media, um, but that wasn't like day one, what you were doing right. 

[00:06:59] Regina Ye: No. [00:07:00] Uh,  

[00:07:00] Founding Story & Early Product 

[00:07:10] Regina Ye: day one we started without knowing anything about what's retail media and like we were completely like oblivious to all of that. We had a much simpler starting point, which is, there are three engineers, me, my co-founders, Francisco and Michael Ostrovsky, and we were sort of losing our minds in 2020 towards the end of it. 

[00:07:20] Regina Ye: And we were just like having a lot of. These conversations and like Zoom, happy hours and chats. And, um, so Michael, who's, um, a professor at Stanford, GSB and really well-known economist, um, in the world and sort of, been working on ad auctions for the past few decades and very involved in early days. 

[00:07:40] Regina Ye: Pinterest set up in Google, uh, play store and LinkedIn ads, uh, Yahoo ads. And then we sort of got together and said, Hey, like, you know. Uh, it makes sense to have this built from day one in a way that's much more scalable instead of later on, bring. A doctor when the system starts showing symptoms of like, hey, [00:08:00] not working and being sick in a way. 

[00:08:02] Regina Ye: And, um, that's sort of what we started. We believe in auctions. Um, we felt like the know-how and the, the tooling for building a functional ad business at scale is very scarce and, um, to democratize. That was sort of one thing that got us very excited about. we started with much simpler product, just listings, just sponsored products, uh, only auto bidding, none of the business logic. 

[00:08:26] Regina Ye: Um, and uh, yeah, that was five years ago. 

[00:08:31] Scott Messer: a really 

[00:08:31] Tom Limongello: okay. 

[00:08:32] Scott Messer: starting point and obviously, you know, solving the, the small problem is interesting. I'm curious. How you thought about like going from auctions, right, and sponsored listings and sort of being an engine there, and then how does that expand into the larger function of what becomes the ad delivery system and, and expanding into the all encompassing ad server? though you're not 

[00:08:55] Regina Ye: Yeah. 

[00:08:56] Scott Messer: just have an ad server. 

[00:08:58] Auction Logic & Business Rules 

[00:08:58] Regina Ye: It, it's a very sensitive topic. [00:09:00] So yeah, we started, um, thinking about ultimately like what's the best way to, so there are two parts to our thesis always. One is to democratize the ad know-how in the systems. And the second part was to think about how do all these retailers and, uh, vertical commerce come together and become like a sort of similar or comparable ecosystem and network, like programmatic. And, um, that was sort of like a personal thing for me because my previous business, I was actually. A vendor where I was working with retailers and um, also like DTC models and how to figure out ads was like taking the majority of my time. Uh, instead of, you know, like logistics, like inventory and things like that. 

[00:09:41] Regina Ye: And retail was sort of this black box. Um, and then like basically we, I think when you are an engineer, like when we first started, we were clueless about how to do sales. Like, you know, you have this beautiful idea of like. Um, there are three of us and like on paper we seem very smart, right? Like, you know, one of [00:10:00] them like got perfect score in the math Olympiad, um, not me. 

[00:10:03] Regina Ye: And then he's like spending his whole life with really smart academic people. But then, um, when you start talking to retailers, it's a very humbling experience, right? They would tell you, Hey, you know what? I have this very exclusive contract with Red Bull and we have to do things this way. Um. We have Black Fridays and Black Fridays sort of, you know, don't follow any auctions logic. 

[00:10:25] Regina Ye: Like someone just wants to pay a lot of money to be on the headline for like an hour. And, um, and then we're just like, well, the, the reality is that you either do that or you have a product that people can't really use and solve all the different problems that they would have to solve. So then, um, we basically just had to listen to our customers. 

[00:10:44] Regina Ye: So in a way we sort of built a reverse funnel, like from the very basic, very core like. Products and then backwards to mid funnel and top of funnel products. 

[00:10:55] Tom Limongello: But it also sounded like you were building out, um, potentially. [00:11:00] You know, business logic or business rules, uh, as a secondary capability to make sure that you can handle the Black Friday money. 

[00:11:07] Regina Ye: Yes. Yeah. And then actually a. Big part of it's actually, how do you make it coexist, right? Like, um, there's just such a wide spectrum of like, uh, maturity for retailers these days. And um, some of them really have majority of the business running on one piece. And even when a lot of them talk to us and say, Hey, we want to move towards a more automated. More scientific way of scaling our media business, but I still need to help the whole company with this change management and this transition. And that's where we're like, okay, we have to figure out how to bring it together. So, um, things like, not just the business logic, but like how do we do attribution on an exclusive campaign? How do you calculate the ROAS on those things? Like that has become, uh, something that I think was, I would say it's like interesting to bring it together and like we've spent a lot of time on. 

[00:11:58] Scott Messer: That's the, a fascinating part to [00:12:00] me of like where the, the ad server and what I would call the API driven commerce engines sort of come together. The, the publisher ad server never had auction logic inside of it and selection, logic and attribution. That was all handled. Uh, by the DSP and any of the direct campaigns, were just order fulfillment and making sure you deliver a certain volume to a certain place. So what I think is really interesting is how when you put the auctions inside and you bring attribution inside the models to talk about driving performance for something like sponsored listings where there's no creative, it's all about relevancy and users and, and journeys. You've now sort of taken the, the brain of the DSP. And moved it inside the ad server and it becomes a function for the, the retailer in this case to use it to their leverage. So they sort of smashed two systems together. [00:13:00] and then I think is 

[00:13:01] Regina Ye: Yeah. 

[00:13:02] Scott Messer: of what separates these from like a traditional Google double click gam. 

[00:13:06] Regina Ye: Yeah, for sure. And I think you have different motives also, like different, um, motivations as like companies here, ad servers. In a way it's very, um, it's very tempting, right? To say, Hey, we will give you all the control, but then if you mess up anything in terms of the economics. It's sort of on you. Um, which happens a lot. 

[00:13:24] Regina Ye: Like at scale, people are like, okay, wait, like we don't even know, like the people who set up all these rules and like, um, what feels like legacy traps, like don't even work here anymore. And then like someone who's being handed over that P&L needs to go figure out what's going on. but it's great because it's like, it's sort of saying, you know what, we're only handling the input, the output, not so much, but like really the magic. 

[00:13:46] Regina Ye: And I think a lot of the things we talk about, that's the beauty of like commerce, media, retail media is really that it's closed loop attribution. You actually have that feedback loop and without that input, um, it's sort of saying [00:14:00] like, okay, you can do everything, but there's an incentive to not actually give you an intelligent piece. 

[00:14:05] Regina Ye: And then there's the other one, which is saying, you know what? We really want the retailers to be as smart as possible, uh, with their data and like optimizing for, for, for that thing. Yeah. 

[00:14:15] Tom Limongello: So, um. We, you know, we were excited about having you on because you're not afraid of spicy takes, and I think one of them is that, you know, you mentioned to us that you don't need audience data. So in a retail media world, what does that mean? And how, how, how is it possible? I. 

[00:14:31] Regina Ye: Yeah. Um,  

[00:14:32] Intent Over Audience Data 

[00:14:47] Regina Ye: it's not just a take. We actually live by it. Uh. More than two years. Um, like we've been around for five years. We scaled, I think, uh, one of the customers they were in, they were making about 80 million a year, run rate, ad business without actually having any, um, audience data. And we had to, in early days, like our engineers, like, and my co-founder, like CTO, would go on calls and sort of fight the [00:15:00] customers in a way. 

[00:15:00] Regina Ye: It's like, why do you need audience data? Like, what if I can prove to you that it works better without it? so I think like we sort of, I guess it comes down to like, first of all, what do you consider audience data, right? Like in the traditional sense of like, it's a snapshot of who you are at any given time. 

[00:15:16] Regina Ye: And then I use that piece of information and um, try to do it maybe like as a targeting parameter. Two months from now, you are no longer with that intention. Um, so that sort of. Static audience, uh, labeling is something that we don't think even really makes that much sense. It helps, like it kind of confuses the system and, uh, a lot of the piping isn't there for real time signals, et cetera. 

[00:15:41] Regina Ye: So we, um, always believed that, um, you can actually run really effective and profitable ad businesses. By sticking to the moment and to the intent. So if I know you are right now on a food app looking for a burger, um, I [00:16:00] really, my best shot is to deliver you or give you the best burger recommendation or anything adjacent to a burger. without having to guess like where you were like last seven days on a social media app, like whatever random article you browse like three days ago really has nothing to do with what burger you're going to eat in a way. Right? So, um, I think that's sort of like one thing that we decided to prioritize and I think at the time we're seeing the shift sort of manifest in a different way, where at the time when we first started it was a lot of things around cookies and third party cookies, if that's gonna go away. Um, now it's more like first party and just like, kind of became a little bit irrelevant in terms of that, that piece. But still like, um, we wanted to build a system that will be future proof, where if, um, apple changes the rule and uh, Facebook messes up the cookie or whatnot, like the business core of it won't take a hit. And we actually saw that number, um, like all the numbers really match to that [00:17:00] thesis and the customers we work with. 

[00:17:02] Tom Limongello: So just give me the burger. 

[00:17:04] Scott Messer: A. A, a technical, Tom wants to talk about cheeseburgers. I'm gonna go into a technical point from that standpoint. Like you're still using user data and identifying users. And I presume in that case, like, uh, the top sort, sort has a first party identifier that it's dropping on users in those sessions to remember it. 

[00:17:25] Scott Messer: For machine learning, you're not completely devoid of knowing who the user is at that moment. 

[00:17:30] Regina Ye: So the way we do it is, um. We, there's like a base layer and that's sort of like the core setup that everybody starts with. And then if someone is like a retailer is like, I really wanna give you more data, which happens, um, they're, they're like fancier, like, you know, things we can do. We connect to their CDPs, we work with a clean room or like we figure out how we can pass behavior signals back to them. 

[00:17:55] Regina Ye: But the very core, like starting point, we just need. promoted [00:18:00] impressions, clicks, and purchases. And then we don't have, um, like a very explicit identifier, but we do have a thing called opaque user id where we need to be able to know that this is the same person who made the impression click and purchase. And it could be super random gibberish, but like, we just need something to tie all three together. Um, and that's the extent that we have and that will be like to the marketplace to just tell us what that is. And they, they tie it together in a way. 

[00:18:28] Scott Messer: Gotcha. Yeah, that's about what I expected. Thank you. 

[00:18:32] Tom Limongello: Um, well, I think, you know, an example that you brought up when we were talking previously was about, you know, if I bought a fridge three months ago, so in my head. Something that's a high consideration purchase. Yeah. I think like, I'm not gonna buy that over and over again. So that, that does sort of distract from the, the session. 

[00:18:50] Tom Limongello: But is is, is your thinking that really it's the behavioral data that has more value? It's, it's what people are doing on the site, or you know, what they're doing in the moment. That's, is that sort of what your, [00:19:00] your thesis is? 

[00:19:01] Regina Ye: Yeah, we think the, the networks should be built around products, um, product, like there's a lot of data in the catalog itself, right? Metadata like the. Product information embeddings, you can build around products. But then also, like we do believe like behaviors around the products will become like that's really the, the gold mine. 

[00:19:20] Regina Ye: And that's what we've seen that really makes the biggest lift, um, to the campaigns instead of trying to. Not be creepy, but like basically try to piece together all the sort of background information in a way. And like, um, it, yeah, like, and I, and I think sometimes like there's that, like in our experience that we see tension early on where, um. 'cause you have a whole generation of marketers that are really, like, grew up with this like audience targeting is like your biggest lever for performance. But then once they see the performance and see that it actually works, like I think people really start to trust the system more. And then, um, we try to make it more open in a way where. We're not [00:20:00] just making the ad system take the um, CDP data or audience data from the retailers or marketplaces, we try to feed it back the signals as well. Um, and that's been really good. Yeah. 

[00:20:13] Tom Limongello: Yeah, I mean, we're seeing a lot of that with retailers where they're pushing more and they're bringing retail data and, and they're bringing their, demand over to GAM, DV360. And it's interesting for me to watch how the industry's sort of starting to play nice with each other and so interesting to hear. 

[00:20:31] Tom Limongello: Um, I mean, are there certain data infrastructure things that needed to happen before that could work? 

[00:20:37] Regina Ye: Yeah, I think it is becoming more open and transparent and, and people realizing that you have to, if you wanna hit scale. and there's a way to preserve, and I think that's kind of a big battle that we've been trying to balance is like, how do you preserve the customization all these retailers feel? 

[00:20:54] Regina Ye: It's very special. It's very different. Like they 

[00:20:56] Tom Limongello: Right. 

[00:20:57] Regina Ye: way, and at the same time you have [00:21:00] like actually a good auction dynamic and like, you know, you have all these things going on. That's kind of where we think things are headed. 

[00:21:07] Tom Limongello: Okay, so I wanted to take us in a direction where we understand some of your earliest clients and, and the fact that you have an international background. Um, are there, just at the outset, are there types of customers that are doing things that the US could learn from, that you, that you work with? 

[00:21:25] International Customers & Global Learnings 

[00:21:25] Regina Ye: Yeah. Um, we like to work with complex customers and who have like complex problems and, um, in many ways you see a lot of our customers, especially the early ones in new market, like really have, they have the same qualities in many ways. Like, and um, it's also interesting because like sometimes they become really, um, like they will call up each other. 

[00:21:51] Regina Ye: On unlike different topics and say, Hey, you know, what are you guys doing? And because they're not competing at all in terms of geography, which is really great, like it feels like we're kind of helping, [00:22:00] um, bridging some of this, um, like community in a way. But like, for example, um, early days we have a customer that ended up becoming part of Woolworth and then we had to like, um, go through different steps to work with Woolworth. 

[00:22:14] Regina Ye: And then, um, 

[00:22:15] Tom Limongello: Well we're Woolworth is, is that an Australian retailer or it's not The American one that went out, went out of business a long time ago. 

[00:22:21] Regina Ye: No, it's, it's the Australia one and they're doing pretty well. So, um, and they're one of the, the leading ones in Australia say, so then, um, like problems around 1P and 3P, right? Like how do you glue those together? also I think a lot of the international ones started with in-store and then. They actually, for them digital isn't really a given. Like people actually are living in much more urban settings and shop in stores way more. Some of them have way better loyalty programs and adoption there as well. So that's been interesting to see. We also got a lot of like early customers in, uh, Latin America or Europe where, um, because [00:23:00] I would say like part of the reason is because the countries are not as like. Some of the countries, right? Like in latam, like it's very common for the, the retail holding groups to be like owning, um, businesses in nine countries. And um, they have to then have a system that can reconcile different currencies, 

[00:23:18] Tom Limongello: Oh yeah. 

[00:23:19] Regina Ye: with different time zones, and, um, it forces you to be good, um, technology wise. Where then I think some of those problems like you see now in the American businesses, and it is funny because I think. Um, in a way the US could get away with, um, not having to think about some of those problems. But then I see the opposite happening right now where you will talk to a retail group in the US that has like, let's say three subsidiaries under them, three marketplaces. 

[00:23:47] Regina Ye: And they're saying, Hey, I have them all running in different, like ad servers and different systems. We use GAM for this, CRI for 

[00:23:56] Tom Limongello: Yep. 

[00:23:57] Regina Ye: Um, thinking about layering this thing on [00:24:00] top, um, bringing this other thing over, and also maybe clean room. And they realize that still they don't have a very cohesive system underneath that. And, um, I think in a way many companies in the past few years copy or try to copy Amazon ads in terms of its form, but not really getting the, the, the substance right. 

[00:24:21] Tom Limongello: Yeah, I mean, you know, my first days when I was working in Quotient, I started to realize that, you know, the websites were built by IBM and they have Adobe. Be. And so even on top of all the ad tech infrastructure, there was all this other legacy sort of outsourced capabilities that made it really hard to make any changes. 

[00:24:39] Tom Limongello: Um, and it made a turnkey retail media provider a very valuable partner. And I think as things have gotten, uh, more sophisticated in the US and probably in internationally as well, companies that are not doing every last part of the value chain are starting to to be interesting. And that's obviously [00:25:00] Topsort is, uh, obviously really engineering heavy, but you're not doing everything right, like you're not all of the, all of the channels and all of the, you know, offsite and instore and all of that, right. 

[00:25:10] Regina Ye: No, we really see ourselves kind of winning in the, um, the intelligence piece. Less so the, the delivery piece, but also I think the way to. way we see ad servers or the way we evaluate these platforms, it's kind of outdated in a way where you're thinking about, okay, this is the best onsite ad server and this is the best offsite or like best for video. 

[00:25:30] Regina Ye: And really at the end of the day, like the problem's kind of twofold, like you have or three, like you have the ad delivery piece, great. If you have a really good ad engine, it can point any of these directions. It can power video, it can power like products, like these different formats. And then what you, what you really need is sort of like a cohesive operational layer and also like the optimization intelligence piece. 

[00:25:52] Regina Ye: And if you get that right, like it really, the surfaces that it's activating, it doesn't matter as much. Um, that's, that's how I [00:26:00] feel. 

[00:26:00] Tom Limongello: So what are you most excited about? You know, now that you have a strong customer base and you're, you're moving into. All these new areas where you have a lead, what do you see is happening in the next two to three years that you're excited about? What's gonna change in the commerce, retail media space? 

[00:26:17] Regina Ye: Yeah. Um, our bet and what we're seeing right now in the market is that. I think it's actually the pendulum's gonna swing the other way. Where, uh, retailers are realizing it's actually a lot more, like, a lot riskier to put your eggs in so many different baskets. And there's a bit of a consolidation of, Hey, let's bring all these different tools into one platform. 

[00:26:39] Regina Ye: Now we've tested everything on the market, uh, right. We're seeing that still. We have all these. Different challenges, let's bring it into one. And so we think there's gonna be a consolidation and simplifying of like all these different toolings and layers. So very exciting. And then also with ai, like we're very bullish on ai. 

[00:26:57] Regina Ye: Um, I do believe that AI [00:27:00] will change how we shop and definitely how we operate and manage like ad businesses. And, um, for us it's not just about building the supply side infrastructure, but really kind of doing what we do best, which is. We are really good at auctions. Um, and we're really good at predicting the next product related behavior, right? 

[00:27:21] Regina Ye: Whether it's click or purchase. So we wanna focus on that. And with that being the core, you can really make a lot of different products or applications that benefit not just the retailers, but also the brands and the agencies and kind of building the pipes and connect connectors to have it all talk to each other. Um, and that's where we see like sort of our value. Um, we're building some of that, but with the, the rate tech is shifting, it's really hard to say what's gonna happen in three years, but hopefully we get there. Yeah. 

[00:27:49] Scott Messer: Well, I'm,  

[00:27:49] Agentic Shopping & the Future of Commerce 

[00:27:49] Scott Messer: I'm glad you brought up predictions and agentic shopping. Um, I'd love to get your take on, like, what do you think? The landscape of a, of ag agentic browsing and shopping looks like, and how does it shift the industry? This is sort of independent of a top sort, know, how it impacts top sort. 

[00:28:06] Scott Messer: We'll get to that and I wanna get into sort of your take on ag agentic workflow and planning, but how, what do you think's gonna happen to the, the general e-commerce economy? 

[00:28:17] Regina Ye: Yeah. Um, I think the stores will become. A lot more relevant in this new world. Um, which is sort of funny because we're talking about like agents and ai, but actually if you think about how agents shop, it's very similar to how you shop in store. Like there are aisles and you know, it's really about like clusters of products instead of. 

[00:28:37] Regina Ye: Kind of search grid in many ways. And then, um, we think the lines and the boundaries become a lot more blurry, which is why I said earlier that ad delivery, it's really not about, oh, you are just boxed into an ad delivery like for ad server for onsite or, or things like that. So lot of those boundaries that we made up, I think they are not going to matter, uh, very quickly. And, [00:29:00] um. like I was sitting in, um, I think at an, an event last week where people were talking about, oh, gen Z shop this way. Or like, you know, like, we would never do agents like, uh, I would never use agent to buy this. And, and I think it's happening so quick because I remember I looked up my calendar. June last year, um, I was in, uh, Seattle for, for a meeting. And then that was when we first talked about agents, I think, as an industry. And then, um, some of the developers were saying, you know, in the future it's going to be agent to agent negotiation. And nobody, like, they, they looked at them as if they're crazy. 

[00:29:34] Regina Ye: And then like a year later, I think it sounds less than a year later, it sounds a lot less crazy. 

[00:29:38] Tom Limongello: Oh yeah. 

[00:29:39] Regina Ye: so I think that like, we're very bullish. We think like. For example, search, like, which is the core of, um, retail media volume today, that's gonna change. It's going to greatly reduce. I don't think it's necessarily gonna fully go away, but it will change a lot. A lot of these search companies that previously like optimized for relevance in like the search, um, grid layout [00:30:00] or the, the way that human beings consume information. They have to adapt to how hu like how agents, uh, consume information. And um, when you have all these different agents, they will fight. There's no conflict resolution mechanism right now. So then, 

[00:30:14] Tom Limongello: I make, I make them fight. I, I'm like, ChatGPT said this. What do you think Claude? 

[00:30:18] Regina Ye: Exactly, so then like I think retailers like really will have to kind of adapt to that and to do that. Catalogs. Um, and just like the basic, um, intelligence piece matters even more. Um, that's really what we think. Yeah. 

[00:30:34] Tom Limongello: That's very interesting. I know Scott wanted to follow up, but I, for me, the most interesting is I first, you know, looked at your company. It was very much about that grid and about that. So it's interesting for me to like, you know. Your future, you know, is defining sort of that new shopping behavior. But you know, it, it sounds like you're not scared that the grid might, might collapse. 

[00:30:55] Tom Limongello: Um, like what, what, what is the new thing that happens? Like what, [00:31:00] what's the brain going to do for you? 

[00:31:02] Regina Ye: Yeah. when ChatGPT first had its viral moments a few years ago, we had our scare moment back then. Uh, we basically started preparing for it since then. So, um, really we have been training our own like large marketplace model, which runs very low latency and trained without sort of dependency on third party cookies or like audience data. 

[00:31:25] Regina Ye: In essence, it's much just about products and behavior and, um. What we think needs to happen is that you need to basically compress the entire recommendation and promotion engine into very small surfaces. And whoever can do it with the most relevancy actually gets the trust of the retailer, of the the customer. 

[00:31:47] Regina Ye: And that's the one thing that we need to focus on very well. And um, it's also actually a lot more similar to an auctions problem than. Sort of like an ad delivery or like, um, a serving [00:32:00] problems. So in a way, like I think if, if we live in a world, a future world where all the line items and all these sort of like key values and these things go away, um, it's a little bit anarchy. 

[00:32:11] Regina Ye: And that's very exciting for an engineer, I think in many ways where, um, you're just thinking about, okay, like, you know, have to just compete on whoever gets the recommendation, right? And does it the fastest. 

[00:32:26] Tom Limongello: Very cool. Um, so yeah, we're coming close to time, but I do have one more question, which I'm starting to ask all the guests. Um, and that is, you know, you're, you're living in this retail world. And I'll give you a few examples of retailers who've done bold things to try to create their markets. Um, the first one, Sam Walton, when he was running five and Dime before it was Walmart, um, took something called Zori Sandals, sold 'em for 19 cents, put 'em on a table, and they, you know, people thought they were crazy, but they went like hotcakes. 

[00:32:56] Tom Limongello: Um, dollar General had an overrun of pink corduroys. [00:33:00] They didn't know what to do with, so they priced them to, to move, and it turned into a, basically a big ad for Dollar General all over the town where they were, they had their first store. Um, and then finally five below ships. Their, uh, their, their basketballs deflated. 

[00:33:15] Tom Limongello: Um, because they wanted to get more, uh, in there and have really low prices. And so those types of bold moves, if you translate that into the ad tech world and things like that, it may not, maybe doesn't sound as cool, but for us, the middlemen we're excited about it. So if you could give us anything that you've seen that you've done or even just, you know, one of your retailers has done that you thinks is, is really bold, I'd love to hear it. 

[00:33:36] Regina Ye: Yeah. Um.  

[00:33:37] Bold Retailer Stories 

[00:33:37] Regina Ye: I love meeting the founders and like the, the people working behind these like really resilient, um, retail brands, especially around the world where it's, one thing that retailers do very well is like the survival instincts, right? They really stay in business through like ups and downs and like technology shifts and, and whatnot and like. Two, I think, I mean, I have, I have two or three stories if, if you're interested, but like one I, I think would be Woolworths, like how we got that business. Um, we won that because we ran a LinkedIn ad. So, 

[00:34:11] Tom Limongello: Wow. 

[00:34:12] Regina Ye: yeah, it was 2021 and, um, we had no idea what we're doing. A bunch of people, like we, none of us had sales experience at all, and we had no idea how to talk to, um, salespeople. 

[00:34:23] Regina Ye: So we were like, okay, let's just run an ad like LinkedIn ad. I thought it was just running in the us. Um, no idea how it ended up targeting Also Australia. And then, uh, one day we got a call and then the guy's like, Hey, we want a demo. It's like, okay, cool. Like, sure, we will talk to anybody that would talk to us. 

[00:34:41] Regina Ye: And then, um, then they ended up like ghosting us and just like went completely dark for, uh, about six months. And then they would come back chat again. Six months dark again. And then one day they're like, oh, by the way, we just got acquired, um, by this company called Woolworth. We convinced all of them, you know, they [00:35:00] should meet you and like, let's do this. 

[00:35:02] Regina Ye: Um, do you wanna give us a demo? And we also got demo from so and so. We didn't like it, so we really recommended you guys. And, 

[00:35:09] Tom Limongello: great. It's like, you know, it's like the shopper. We don't know what the shopper is thinking. We think we have all these signals, but yet they're working very hard behind the scenes. 

[00:35:18] Regina Ye: yes. So then, um. Yeah, so then that's how we got, got to meet everybody and um, and I think that even ultimately led to the W23 investment that also brought in Tesco, ahold, uh, ShopRite and Empire, which was great for us on top of Wooly's. So that's how we started that. And then I think, um, gave us a foothold in Australia, which now we also work with Cole's, the other big grocer. Um, and then, I mean, two other stories, I actually, you guys can keep it or not, but like, I, I think it's really interesting. 

[00:35:45] Tom Limongello: them. We'll take, we'll take the stories. 

[00:35:47] Regina Ye: Yeah, so I love the, just the crazy founders of these retail groups. Like they're in latam, they're, there are two, so one is called Cencosud. They're also the parent company of Fresh Market in on the east coast, [00:36:00] um, in the US as well. And the founder, um, I think. famous in Chile, which is where they're headquartered for just showing up randomly in a store. And he would take the mic in the store and starts like talking to people, 

[00:36:13] Tom Limongello: All right. Love it. 

[00:36:14] Regina Ye: like motivational speaks and like, you know, also just like start singing happy birthday songs to people. it's hilarious. Um, and then, um, in Brazil, the largest retailer, Magalu uh, it's named after the woman named, uh, Luiza. 

[00:36:29] Tom Limongello: Okay. 

[00:36:30] Regina Ye: it's magazine, luiza is how they got started. Um, and then they got shortened to magalu and she's probably, I, I don't want to guess her age, but I think she's like, um, you know, like in her older years and, um, she would be the first, like she was at NRF and she would be the first front row to all the events. 

[00:36:49] Regina Ye: 8:00 AM in the morning. And still like runs weekly meetings with her team. So I think just that sort of drive and um, passion that the retailers have is always very, um, inspiring in many ways.[00:37:00]  

[00:37:00] Tom Limongello: I love those stories. Thank you very much for joining us on the Middlemen, Regina. Is there anything that you want to share with us in terms of how, new customers can find you, um, or things that you need to, to tell 'em about, 

[00:37:12] Regina Ye: yeah, I think, uh, it's going to all come down to performance and speed. Um, we wanna be the best in those two things, and we wanna stay the best in those two things. 

[00:37:22] Tom Limongello: Great. Well, thanks again for, for joining us on the middleman and we'll see you at the next event. 

[00:37:27] Regina Ye: Thanks. 

[00:37:28] Scott Messer: Thanks, Regina. Have a good day. 

[00:37:30] ​