# S3 E9 T64 MetaRouter - Nikhil Raj

S3 E9 T64 MetaRouter - Nikhil Raj 

[00:00:00]  

[00:00:01]  

[00:00:05] Tom Limongello: Hello Scott. 

[00:00:08] Scott Messer: Well, Tom. 

[00:00:10] Tom Limongello: So today we are talking to Nikhil Raj of Meta Router. This is a long episode because Nikhil is a veteran. He's really a legend in this space. He was originally working at a company called Kosmix that was acquired by Walmart, that became Walmart Labs. He's gonna go all into that detail in history, um, which is fascinating. But he is now at a company called Meta Router. and it's a notoriously difficult company to understand because what they're doing is in the data infrastructure world. but Meta Router has some really big news that is going to get people very interested in learning what they do. So we learned at NRF from Mark Williamson that, uh, they are using Meta Router that means that Costco, uh, is using them as a data foundation. 

[00:00:57] Tom Limongello: And in the episode we just had last week. He [00:01:00] talked about how Meta router and some other, vendors that they work with were the data foundation that enabled them to go into Google product listing ads. So doing more bold offsite moves. Um, and then we heard at Shop 

[00:01:13] Scott Messer: It's a fantastic question and we're gonna get let Nikhil answer most of those questions. 'cause he'll certainly do a better job than we will. But some of the highlights that come outta this conversation for me were particularly around how retailers can now sort of build their IDs in their own infrastructure and don't have to rely on external sources as much. 

[00:01:35] Scott Messer: And I think the, the key part is that the ID graph is being stored in the customer's cloud. And so you don't have to send IDs out to be worked on, uh, as much in terms of attaching other identity and, and building them into the other graphs, which is a typical process that the industry has had to endure for the last 15 plus [00:02:00] years. 

[00:02:01] Scott Messer: So in my mind that's kind of like a, a dry cleaner that has a plant on premise. And if you wanna get your shirt done in the same day, you better hope that they can work on it there because if they just have to send it out to be worked on, there's no way you're getting it back in the same day. And that's where retailers are picking up speed and efficiency. 

[00:02:20] Scott Messer: And I think that's what's allowing them to go into Google, into these other platforms because you're working on identity so much faster, in such a safer way. That you can own the ID federation and the data enrichment portion, and have real time action across your website and across third party platforms, which is a big unlock for retailers. 

[00:02:44] Tom Limongello: I think this, you know, if you watched the John Flugstad episode at Moloco, um, John Flugstad just moved over to Meta Router. Um, what he was very focused on was. Why isn't retail media enabling, uh, the [00:03:00] data signal of just behavioral data when people are playing around on the site, on the retailer site? Why isn't that being captured and, and incorporated into how retail media is, uh, distributed and. I think that this is a great episode where Nikhil explains the foundations of, of why you can now do that. So that's a huge unlock because generally what was being done before in retail media was just audience targeting. and I think that, you know, we're naming this episode, um, something along the order of, this is the end of renting audiences or renting identity. Um, and the way that I think about that is, you know. The Walmart way that people talk about is about trying to understand how do you run a retail media network. In the case of Walmart where they didn't have a loyalty program. So most retail media has always  

[00:03:52] Scott Messer: Absolutely it. It was really great to listen to Nikhil talk about the way that he solved this first for Walmart, and then now they're bringing this [00:04:00] technology out into the public for other retailers to leverage as well. I think this is a really great advancement for the industry overall and can't wait to see where they take it. 

[00:04:09] Tom: All right. Welcome to the Middleman Podcast. We are really excited today to have Nikhil Raj, um, of Meta Router Nikhil. Hello. 

[00:04:18] Nikhil Raj: Hi. Thanks for having me, guys. 

[00:04:20] Tom: Yeah. You were at Moloco when I first reached out to you, and then I think we, uh, we, we were very lucky to be sitting in the same Criteo panel, at, where was at CES this 

[00:04:31] Tom: year.  

[00:04:31] Nikhil Raj: CES, yeah 

[00:04:31] Tom: It was a fortuitous, meeting. But I had many questions for you. Like how did you end up building out Walmart Labs? Why did you leave Moloco to go to Meta Router? So why don't you tell us a little bit of your background, especially the Walmart part to begin with. 'cause I think our audience would be interested in hearing how Walmart Labs started. 

[00:04:50] Tom: I.  

[00:04:50] Nikhil Raj: Well, in addition to fortuitous, I don't know how to say that word. When we met, it was also a bit embarrassing because. Tom as you know, you sent me a [00:05:00] message and I had completely not seen it. 

[00:05:01] Tom: You are, you are not a LinkedIn warrior at the level I am. 

[00:05:04] Nikhil Raj: I'm not, I'm not a LinkedIn warrior at all, so. 

[00:05:07] Tom: that is, uh, I think people like me who are, who use LinkedIn to the level that I do are because of other deficiencies, I don't code. So you. 

[00:05:16] Nikhil Raj: Well, I don't code either, so. There you go. So, anyway, so we, uh, so it's a good, good question. So look, let me start with the finish line. The reason I joined MetaRouter in general, independent of Moloco, was what MetaRouter is effectively is what my team built out at Walmart to support and start the Walmart advertising business back in the day. 

[00:05:39] Nikhil Raj: So that's, that's when I, when I met the team and I met the company and I found out what they're doing, I immediately knew what to do with it. Got excited and joined them completely independent on Moloco. But let's talk about what happened in Walmart's, a direct connection. 

[00:05:50] Tom: Yeah, when was this? When, when did Walmart labs start and you know, when did you get 

[00:05:54] Nikhil Raj: so Walmart Labs was started in 2011, April 18th. 

[00:05:57] Nikhil Raj: The acquisition was announced when my startup [00:06:00] Kosmix was bought by Walmart, right? We, we were, uh, a Silicon Valley, um, ad funded publisher side business founded by two guys who had sold their original company to Amazon in the nineties. So the, the entire Amazon backend for marketplace and, uh, item infrastructure came out of that, that acquisition. 

[00:06:21] Nikhil Raj: And Venky and Anand had uh spent a lot of time with Bezos directly reporting to him for many years. They came outta Amazon, started a fund, and one of the first, first few investments and founding stories was Kosmix from that fund. Right? They were also multiple other companies that had fantastic outcomes. They're some of the best known. 

[00:06:40] Nikhil Raj: They're one of the best known, I would say, best investors in the Valley entrepreneurs, right? So, so Walmart bought us not only because. Of the background of e-commerce and Amazon, but also because Kosmix was built on top of the search engine and a context engine that was way ahead of its time. I think [00:07:00] like the concept of embeddings and machine learning that is now so common and context is so common. 

[00:07:05] Nikhil Raj: We were trying to do it with manually, with statistics and taxonomies, uh, as opposed to machine learning embeddings, right. It was so early for its time. 

[00:07:13] Tom: Wow. 

[00:07:14] Nikhil Raj: So Walmart bought Kosmix. To restart kind of the, the hair on fire moment. I think, I mean, this is me interpreting this. I don't, I was not on the Walmart board, not, I was on the executive team of Walmart, so I don't quite know why they bought us. 

[00:07:29] Nikhil Raj: But my take from the outside in was, hey, they needed to compete with Amazon and e-commerce was a pretty big important, uh, thing for them, but it was not going well. When we, when we were acquired, Walmart was number eight in US E-commerce traffic. When I left six years later, we were number two after amazon.com, and we built a lot of the foundational tech and capabilities to enable e-commerce growth. 

[00:07:54] Nikhil Raj: Right? And one of the things that, um, we did at Kosmix was monetize Kosmix.com, right? Health, which I [00:08:00] ran bunch of websites, vertical websites through monetization, through ads. That's kind of how Kosmix properties were monetized. So when we came to Walmart, it was pretty natural for us to say, Hey, this is one of the largest publishers on the internet. 

[00:08:13] Nikhil Raj: We build an advertising business at Walmart. Turns out there was one, one person who was doing display ads through a, an ads server called 24/7 Real Media. I don't know if you guys have heard of it. And it was a $2 million a year in, in revenue. Something in that range, right? So we got in there and we got in touch with some of the big advertisers through the Walmart relationships, and we're like, okay, what do you guys want? So they're like, well, we got so much data on sales. We'd love to know if any of our advertising is converting in sales. We're like, sure, we'll we'll give it. We'll give it to you. And we turned around, went back in and found out that there is no way to do that. And the reason there's no way to do that is Walmart does not have a loyalty program. 

[00:08:56] Nikhil Raj: Right? If you [00:09:00] think about the world's largest company, largest consumer company, built without a loyalty program, and everybody talks about loyalty programs and. Walmart is one that went the other way, right? 

[00:09:09] Tom: Yeah. Everyday low price has its 

[00:09:11] Nikhil Raj: a very low price is a loyalty. So you have no idea who saw the ad and you have no idea who bought in the store. So what kind of advertising measurement, what kind of targeting are you gonna offer p and g when you have no idea who the consumer is? So a technology was born under the constraints of a lack of a loyalty program. We needed a piece of technology that. Allowed. Privacy, safe, anonymous, creation of audiences. 

[00:09:39] Nikhil Raj: Targeting of audiences and measurement of media completely anonymous without a loyalty program. 

[00:09:46] Tom: So. When you say it was anonymous, this was about the idea that you don't have somebody logged in or you didn't, because Walmart eventually has tons of people logged in, but this is sort of before that 

[00:09:57] Nikhil Raj: interesting you say tons of people log in, [00:10:00] uh, Walmart Plus, which is the loyalty program. I don't know, again, outside. 20, 30 million members at that time, it was zero and 0% logged in checkout, 100% guest checkout when we were acquired by Walmart. So even the logins were not stored. It was, it was EDLP implemented online. 

[00:10:19] Nikhil Raj: We got there. So, so literally we were, we had a bunch of email addresses for a transaction purposes. Hey, your order came through, your order shipped, your order was delivered, right? Other than that, we didn't use the emails for anything. So, and then in store was, even today in store, a large chunk of the transactions are anonymous because there is no loyalty program even today, right? 

[00:10:39] Nikhil Raj: Walmart Plus is largely online. So this problem or constraint, if it's not a problem, created an opportunity for us to invent something that we had to, otherwise, we couldn't give a connection of a, Hey, this person saw the impression this person bought in store. So we built the system to create that connection. 

[00:10:58] Nikhil Raj: In an anonymous, [00:11:00] privacy safe manner with third parties exchanging data between ba, Walmart, and the third party in real time, right? So at the end of that process, realtime process, Walmart owned an anonymous identifier for every consumer on walmart.com or for every consumer inside Walmart store. That anonymous identifier technology allowed us to do a bunch of things including launch the advertising business. 

[00:11:27] Nikhil Raj: That technology is essentially MetaRouterer, 

[00:11:31] Tom: So this is an ID that's not a cookie and it's server side. 

[00:11:37] Nikhil Raj: It's a, it's an ID that is built on whatever consumer information is available. It could be anonymous, such as an IP address, a household email, right? It could be an email that is less anonymous. It could be other things that can be used as match keys but cannot be stored. Remember, we could not store email addresses for marketing, [00:12:00] so we used email addresses to create a match key, or we used ips and other things to create a match key, and we stored the anonymous match, but we didn't store the the email, 

[00:12:10] Tom: Okay. 

[00:12:11] Nikhil Raj: Similarly, in store, there's lots of signals including payment information and other things that happen in store, and we use that information to create an anonymous ID in store and using the same third party to do. Anonymous online and anonymous in-store allowed us to match the same ID back. That's how we built the entire system. 

[00:12:33] Nikhil Raj: Even today, I think it's called Walmart Connect Today. So Walmart Connect even today, has to use that system. Otherwise, there is no way to match a large chunk of in-store sales to online traffic. This is the way to do it, and MetaRouterer is essentially the same system. Significantly better because we are now deployed in some of the largest retailers with largest with [00:13:00] large loyalty programs, and we add both the loyalty and the anonymous together to provide near a hundred percent coverage to their consumer data, digital and in store. And now I can talk about it openly because Mark at Costco, 

[00:13:14] Tom: Yeah. 

[00:13:15] Nikhil Raj: My partner in time at Walmart right back, I, we built the business together. So he was the head of Sam's Club bags. I was building the technology for the full company. I mean, uh, there's a guy called Brian Monahan who runs, um, Albertsons. So Brian, and so Brian joined two years after we launched Walmart, Walmart Exchange, we called it. He joined much, much later. But Brian was my internal customer for WMX, for Walmart, and Mark at Costco was my internal customer for WMX, for Sam's Club, and this one. Underlying infrastructure. Ran the whole ADS program. 

[00:13:45] Nikhil Raj: I mean, I even worked with Asda and other, other Walmart subsidiaries at that time. 

[00:13:50] Tom: So I'm happy we waited to finally get you on the podcast. 'cause we've had Brian, we've had Mark, um, and I was there at, at NRF when Mark showed the world what [00:14:00] his tech stack looked like. And that was, that was fun. So I guess all, I guess all of those, uh, if you're from one of those companies that Mark put on the slide, feel free to join us on the Middlemen. 

[00:14:12] Nikhil Raj: There you go. Yeah. A lot of the stuff that, um, that we invented back at Walmart is still powering that tech stack there. I mean, they've all obviously built, built a lot more, but at the end of the day, um, Monahan, uh, Williamson, these are all multiple others. Asda, there was guy called Steve Smith who now runs LL Bean's 

[00:14:30] Nikhil Raj: So his team I was working with at, um, at Walmart. So there's bunch of stuff that, that got built as part of my group at Walmart is powering a lot of the tech still today. 

[00:14:39] Tom: So that's amazing background. So thank you for going through with that because I feel like that gives us a really good foundation to start talking about what has changed. It is not 2011 or 2018, it's 2026. And if we think about, when I started to talk to, to Dave on your team. He is talking about [00:15:00] the first mile and shifting left, and it was all like engineering speak, but it sounded to me like something that was very fundamentally different. 

[00:15:09] Tom: And we've had guests on from Ahold Delhaize and other places, and they're talking about we need to move away from the old way of, of building audiences where, you know, the audience segment is created at the end of the, of the process of the data pipeline process. Um, and so can you talk a little bit about. 

[00:15:28] Tom: The approach of meta router and how you guys are different and what, what is all this first mile stuff 

[00:15:33] Nikhil Raj: yeah, so it's, it's a really good question. So again, it in many ways what we built at Walmart was way ahead of its time, right? So what's changed from then to now was, has not changed for Walmart, is the concept of privacy and security of the data and compliance, right? So privacy is about not, is about honoring consumer consent and making sure that what you do from an advertising marketing perspective is underneath that consent. 

[00:15:58] Nikhil Raj: After that consent, [00:16:00] right? And then compliance and security is about owning all of the data. So while the entire world went in one direction. To send their data out into third parties for a match key. Call it RAMP ID or this other or whatever at Walmart. We built it the other way. We pulled all the data in. So all of this identity infrastructure ran inside a walmart and had we owned the identity grant, right, so when I talk about the first mile, what the first mile is at the day in those days, and even now today is. Collecting, owning consumer data inside your first party cloud and meta routers. First Mile positioning is we are the ones doing that for our customers. We are sitting on these browsers, so we are on costco.com and all the other customers that we have, we are the first ones to see the traffic come in, and that's why it's called the first Mile and the connection back to Walmart is [00:17:00] all of that information. Heads back into the Costco cloud, just like we sent all the data back into the Walmart cloud for doing this. 

[00:17:08] Nikhil Raj: The part about the audience is also really interesting point. The old way, like you said, take the data, send it to your third party cloud. So you've already sent your consumer data outside to a CDP or some other platform. Then you have your machine learning engineers bring it back into your Databricks. 

[00:17:26] Nikhil Raj: So now you've paid somebody an outbound and an inbound fee. Then you're paying Databricks Snowflake, a data warehouse to put all the data in, and then a machine learning team to run all the models. And then you build an audience and a day later you have an audience ready to match. And guess what all the audiences are PII based, known, known audiences only, right? So you've taken your set of a hundred percent of traffic with 20% logged in, and you've built an audience of 20% of the traffic, and then you match 50%, you get on 10%. Match of your entire user base is now an audience. Three days later. [00:18:00] The new way. Yeah. Go ahead Scott. 

[00:18:04] Scott Messer: I, I just wanna be f I'm gonna, I'm gonna ask you about sort of some some CDP things, but I just wanna be fair to everyone who was doing it that way that they didn't have your company. And those things were hard and they were difficult 

[00:18:17] Nikhil Raj: And, And, expensive. 

[00:18:18] Scott Messer: that sort of patchwork and expensive. to do those patchwork of solutions. 

[00:18:24] Scott Messer: But I think you're right that there's been multiple breakthroughs though that allow you to do this sort of ID federation inside your own. And it's, it's a larger concept of like owning your data graph.  

[00:18:37] Nikhil Raj: Exactly. So let's talk about that a little bit. It's five things that are, that CDPs used to do, right? And it costs a lot of money, complex. It takes time, behavior, data collection, identity collection, consent collection, audience building and propensities and distribution. Handful of things.[00:19:00]  

[00:19:00] Nikhil Raj: Put these audiences out on Facebook. Each of them was constrained by, in many ways, behavior was constrained because we didn't come to first party, so any third party behavior tags were blocked. ITP and Safari blocks it. That's what has changed Consent. Facebook's pixel on the page grabs all the data whether or not the user's consented consent was broken, right? 

[00:19:23] Nikhil Raj: Propensity will come to that. In the end, identity was only limited to known users. So if someone was anonymous, you lost them and a distribution was using only the known users again. So I have an email, I'll match it to RAMP idea. I'll go direct to Facebook, match it, hash email. I get an audience over there. Propensity the fifth one. This one was two days later and you said expensive things like that. MetaRouter  

[00:19:51] Scott Messer: Yeah.  

[00:19:51] Nikhil Raj: collects behavior data in a first party, so we are not collecting data into MetaRouter's cloud. We are collecting data into Costco's cloud on Costco's first party, so a hundred percent no [00:20:00] blocks. Full behavior consent. 

[00:20:02] Nikhil Raj: We read all the consent before we do anything else, and if the consent says 'yes', we make anonymous identity connections to all the third parties. Walled gardens are not Trade Desk TikTok Pinterest Snapchat, meta Google, right? And it's not Identity Federation, Scott. It's identity ownership instead of, instead of our customer data going out and matching to a third party, the third party ID comes in and sits in the it's it's owned and sits in our customer's cloud. 

[00:20:35] Nikhil Raj: So there is no more toll to pay. Once you have the id, you have unlimited use of it for a fixed PLA fee. So, huge commercial advantage for our customers. Number four, distribution because we have the IDs, we send all the data on those IDs because we don't need to do matching. We already match it's prem matched. And let's talk about the fifth on the propensity. So we have four of the five things happening real time on the browser without any storage. [00:21:00] And what you mentioned with new technology, and you mentioned Tom, you're now able to do realtime predictions the way ChatGPT does realtime predictions, the way any of these AI tools we've built inside the MetaRouter 

[00:21:13] Nikhil Raj: Infrastructure, real time inference capability. So as the user is clicking, browsing, searching every single event generates a prediction in real time. And that prediction results in an audience, people who bought Pampers, people who are likely to buy Pampers. And the audience identity and consent is already available right there. So we immediately update an audience on any third 

[00:21:38] Nikhil Raj: party media platform that you want. As the user is clicking and browsing, the audience is updating on meta, on TikTok, on Pinterest. So if you leave the retailer app and open Instagram to check your feed, the retargeting ad is there. Ready, privacy, safe consented. 

[00:21:54] Nikhil Raj: No third party grabbing the data fully, real time in your cloud.  

[00:21:59] Tom: [00:22:00] So I can see why Mark from Costco would want that. He's his, his shoppers are all out on the trends. You don't want Facebook to be the only place that can, that can do that type of retargeting. So. That makes a lot of sense. This is  

[00:22:11] Nikhil Raj: And you don't want Facebook to do it directly from your page. You want under your control what Facebook is able to do. 

[00:22:19] Tom: Um, one little note here. 

[00:22:22] Nikhil Raj: Mm-hmm. 

[00:22:22] Tom: How do you as a media buyer deal with the fact that these audiences are so predictive that it's kind of a black box now? 

[00:22:30] Tom: Like, 

[00:22:31] Tom: you know what, what, what am I 

[00:22:32] Tom: buying? Like, I don't know what I'm buying anymore, do I? 

[00:22:35] Nikhil Raj: so so at this point, you're buying outcomes, Tom. That's kind of what you're buying. You are buying the fact that, um, someone is likely to do something that therefore generates an outcome for your business. So the, the old, like the, the one end of the spectrum is I'm gonna buy an impression. I'm gonna buy reach. And then, you know what? I don't want just reach, I wanna know someone interactive. So I'm gonna buy a click. I'm gonna [00:23:00] buy a CPC. And you know what? I'm not interested in that either. I wanna buy a conversion, so I'm only gonna pay you on conversion. So that further step is, I'm not only gonna only buy on conversion, I'm gonna buy on an outcome of conversion. So target roas. Right. And what is P max? What is Advantage Plus? It's really,  

[00:23:23] Scott Messer: Oh, a lot of people wanna know what P  

[00:23:25] Nikhil Raj: yeah. Well it's really this black, black box of trying to figure out an automated conversion optimization system, right? It's not,  

[00:23:32] Scott Messer: Sorry.  

[00:23:33] Nikhil Raj: no, no. People don't like P max. Right. If P max had a little bit of, uh, I, I don't wanna digress too far here, but if P Max had a little bit of transparency I think it would take it much more uptake. If it had a little, a few more dials, it would get much more uptake. 

[00:23:51] Nikhil Raj: Don't trust the black box. Right. Anyways, so, but, but the broader 

[00:23:55] Nikhil Raj: point is technology and AI has made 

[00:23:58] Nikhil Raj: capability of p max [00:24:00] like capability 

[00:24:00] Nikhil Raj: possible today where you can just buy 

[00:24:03] Scott Messer: But yet they're not charging on roas, right? They're still charging on CPM for impressions.  

[00:24:08] Nikhil Raj: no. It's actually P max is outcome. There's no CPM right? You're saying. 

[00:24:16] Scott Messer: P max is, but what about everyone who's on Meta router?  

[00:24:21] Nikhil Raj: So Meta autos. So meta autos play in the ecosystem is not on the media side to be clear, right? We are enabling audiences in real time to show up in Facebook or Google where their ad servers are able to deliver p max like outcomes or advantage plus if the case of meta, right? But to put an audience on that ad server. That is right now interested in Pampers as opposed to you figure that out two days later. Oh man, this ID is a Pampers potential pamper shopper. I'm gonna update this ID on Facebook. By that time, the person's [00:25:00] already finished the Pampers pack already. It's too late. Right? The kid has already gone poop four times. 

[00:25:05] Nikhil Raj: Right? So my my point is that you wanna be in the moment ready to activate audiences. On top of that, leverage P max Advantage plus type technology to to optimize further. So if you give Facebook a bad audience or an old audience, or Google an old audience, the p max Advantage plus doesn't, doesn't help bad unqualified audiences coming in too late. 

[00:25:32] Nikhil Raj: It's gonna give you bad outcomes anyway.  

[00:25:35] Scott Messer: Yeah, let me, let me take us down to technical, slightly technical, non non code technical because that's the op level I operate at. Uh, if someone was to install meta router And sign up, uh, what does, what do they have to integrate with and what do they get to rip out  

[00:25:53] Nikhil Raj: Yeah, so it's a great question. So when this, when someone installs meta router, they install two [00:26:00] parts, two things. They install something on the website or in the 

[00:26:04] Nikhil Raj: app on the, what we call the client side tracking. They own it's white label, first party code, and then they install a second part which receives the data from the client side and that sits 

[00:26:16] Nikhil Raj: inside their cloud inside the server, right? 

[00:26:19] Nikhil Raj: It's two parts, website and cloud. And those two talk to each 

[00:26:23] Nikhil Raj: other to share identity, behavior, consent, propensity, and distribution. I talked about the five 

[00:26:29] Nikhil Raj: things A CDP does, right? It's, it's those two things, 

[00:26:31] Nikhil Raj: those five things. So who does identity resolution? That's one somebody you take out. 'cause now you own the identity and you have it the identity graph in house. 

[00:26:43] Nikhil Raj: So you don't need anybody else who does all of your onboarding, measurement, targeting, audience building, whoever those players are. Now you start to wonder if that is necessary anymore. [00:27:00] We now have the first customer who's like, well, I'm paying so much for my web analytics solution. With MetaRouter, I can cut it down by half because these guys, so we are detecting bots, we're detecting unnecessary data that doesn't need to go in,  

[00:27:17] Nikhil Raj: and the web analytic solutions cost us half now because half the data that was going out is no longer needs to go. 

[00:27:24] Nikhil Raj: It's not that the whole thing has come off, it's just lot less expensive.  

[00:27:31] Scott Messer: And then what about an integration with consent management? I noticed on the, in your materials, you said you  

[00:27:35] Nikhil Raj: we don't, we don't, we don't manage or create consent. We integrate. So we, that's one thing that you need. There's lots of other things you need with MetaRouter, but there are some things you don't. I talked about the things you don't, but consent is a thing you do need. So OneTrust, TrustArc  

[00:27:49] Scott Messer: Right. You need your  

[00:27:50] Nikhil Raj: Your internal system, whatever it is, we integrate with that. Exactly. So is that helpful, Scott? Was that non-technical answer to the non-technical question?  

[00:27:59] Scott Messer: Yeah, that, [00:28:00] that's a, that's sort of the perfect level I think that you get of like, what is this thing? I think, you know, our listeners were, and I was thinking like, this is great. It's fantastic. It's amazing, but what am I really doing here and where does it plug in? So that  

[00:28:12] Nikhil Raj: Well, again, going back to the Walmart days, there is no third party tag. No, there was some discussions, uh, Tom earlier about third party tags, so all of the third party tags come off the page. That's one thing you don't need anymore because now you have the identity graph. You can talk to those third parties on the server side. 

[00:28:31] Nikhil Raj: You don't need to put their tag on the page and ship all your data away to a third party, right? So it's not cost you're replacing, but you're adding performance because now your website is faster. You get extra revenue because of that, you have more page per day. Like if you have a hundred seconds in a day and you load a hundred pages, that's a hundred ads you show. 

[00:28:51] Nikhil Raj: But if you double the site speed, now you can show 200 pages in a day, right? So now you can show 200 ads. So ad revenue goes up. So it's literally, time is [00:29:00] money, right? That's what we enable. 

[00:29:02] Tom: you were super early and you figured this out at Walmart. Now you're still a little early, but you're figuring it out for people who are not Walmart, does the future look like? I am actually really excited about, you know, sort of what things that are happening in ai. Are you basically saying that small retailers now can have their own data spine and. 

[00:29:25] Tom: Cut out some costs or whatever. What happens, like what are these retailers gonna do with this extra time and money? Like what? What do you So this is, this is, uh, this is exactly where we are going. Um, it's a great question, Tom. So at the end of the day, nobody should need to invest. Millions of dollars in expensive AI infrastructure and expensive AI people. There's just so few of them in the world. What you do have, which is unique to you and your business, is your first party data that is absolutely gold that you need to protect. So we at Walmart, we [00:30:00] made a strategic decision to put all the data in our cloud. Nobody gets access to it. That's the way it should be for everybody. In order to do that, you should not have to spend multiple millions of dollars and then find engineers and machine learning people to go work on that data. So my vision and our vision at MetaRouter is to provide that level of real time AI capability to every consumer business on the planet. Where as the consumers are interacting with you in any mode, they could be on your website, they could be in your app, they could be talking to you on the chat AI interface. They could be sharing images with you and videos with you. The mode of communication with you and your consumer should allow you to learn from that consumer interaction in real time and act on that learning immediately without having to invest in expensive data infrastructure and storage costs, and machine learning engineers and [00:31:00] analytics platforms and everything else to give you the answer three days later. No, that's not gonna be the new world. The new world is, everything happens right now with a lower cost and you don't need to invest like we are in two pub, two big POCs right now on ai, where the chief analytics officer of this public company told us, Hey, I don't want to build a machine learning AI, machine learning operation. 

[00:31:26] Nikhil Raj: I don't want the data warehouse. I don't want the machine learning engineer, I just want the MetaRouter API. That'll make me the prediction in real time and go off and do what I need to do on the website. That's the future. That's where we are Pointing Pointing the company.  

[00:31:42] Scott Messer: we talk a lot about like core competencies of companies and like it's not strategic to your business to own a data warehouse, so you shouldn't own  

[00:31:51] Nikhil Raj: Or or hire 15 ml engineers at half a million a pop and on the low end.  

[00:31:55] Scott Messer: Yeah. Don't need that. Um, okay. [00:32:00] Let's, let's pull this crystal ball back out again. My question here is like, how does a agentic commerce and agentic browsing like change? We can, we can start at like the big retail media level and then we can bring it down to like MetaRouter  

[00:32:14] Nikhil Raj: Yes. Look, um, retail media, there's obviously a lot of conversation going on about whether. The opportunity to upsell cross-sell, someone with an ad on your own website is gone because the discovery is happening offsite. Right? And to a certain extent, I think that will happen. But my view is a little bit more, uh, a little bit different in that it is yet another compelling discovery channel for consumers like TikTok. 

[00:32:43] Nikhil Raj: Shop was a compelling discovery channel for consumers built entirely on the similar infrastructure of inference. That, ChatGPT works on, there's no difference. Now, that has not stopped consumers from going and buying things from Walmart and Amazon. 

[00:32:59] Tom: Right.[00:33:00]  

[00:33:00] Nikhil Raj: So to my, my point is that there will be, this is a new marketing channel where you have to reach the consumer and be discoverable just like it was. 

[00:33:06] Nikhil Raj: Insert social, mobile, whatever, open web. That stuff is gonna be there. The second, the second point is what kind of brands are going to get discovered on these surfaces? Right. Again, the, the disruption on shopping is going to, my, my view going to be in the D two C single brand companies. If you are a Procter and Gamble and you have 400, 300, 400 brands in your portfolio, and you're selling to all the large retailers, I mean, discovery is not, you are gonna want get discovered, but that's not where the purchase is for you. 

[00:33:42] Tom: Right. 

[00:33:43] Nikhil Raj: You are supporting your retail partners. You're supporting your e-commerce partners. 

[00:33:46] Tom: Yeah. People don't need to figure out what Colgate toothpaste is. Right? 

[00:33:51] Nikhil Raj: Or what? Charmin toilet paper is. 

[00:33:53] Tom: Right. 

[00:33:54] Nikhil Raj: so there is a question. So there is, so there's the category of the product, like utility, a [00:34:00] hundred percent penetration. Like everybody hopefully has toothpaste & toilet paper, at least in the US right? So it's a hundred percent penetration category and it's just simply a brand building exercise at that point, right? 

[00:34:12] Nikhil Raj: And then there is. Not a hundred percent penetration category. The other end being the beauty, luxury goods and everything else in the middle, right? And then there's direct to consumer brands and there's brand houses. I think the ecosystem is complex enough that what will happen to a certain segment of that complexity, of that ecosystem, which could be a 100% checkout on chat, fine, there will be a segment that, where that will happen. But to expect that to happen for all market segments, for all types of brands, for all types of commerce, I think it's naive to be honest, 

[00:34:48] Tom: One thing though that I'm interested in is. Is the retailer potentially in a better position to have better data to talk with the agent platform, uh, than a brand is [00:35:00] directly because they don't have enough data. Um, that sort of came up in a discussion that Scott and I were having and the, you know, the agent to agent concept. 

[00:35:08] Tom: If the brand doesn't have a lot of data to play with, potentially there's, there's not as good of a collaboration as it could be with a retailer. But I don't know, you know. How I mean, I mean there's the meta route angle and there's, let, let's stay with the sort of, you said earlier, let's talk about the retail media angle and then the meta route angle Yep. The, my point is that, you know. As, uh, a retailer, and I was part of one of them for a while. You wanna be playing in any part of where the consumer is and you wanna build the, into those interfaces. 

[00:35:41] Nikhil Raj: Like my search marketing teams, I had all of MarTech, all of ad tech and the media budget for Walmart when I was there, right? It was, uh, a fairly large, complex technology stack that we built in house to manage search, affiliate, email personalization, media buying for at least the e-commerce business. In one place. 

[00:35:59] Nikhil Raj: And [00:36:00] as new channels came up, we added capability to go buy there. Like when Facebook came up, we added there like Facebook was not even public when we were in, when we were in Walmart labs. Right. And then Twitter came up and then Snapchat 

[00:36:10] Nikhil Raj: came up. I remember 2013 CES when Snap came to us and pitched, Hey, Walmart Media, if you wanna buy teenagers. the only place to buy in the world is Snapchat, right? So we had to build an integration there. So we keep doing that and now commerce companies will build agent integrations. That's like one more thing to build. It may be a lot of hype, I don't 

[00:36:29] Nikhil Raj: know, maybe, but it's a true new channel, right? So in that channel, the interaction is through commerce protocol, a CP from chat, 

[00:36:36] Nikhil Raj: GPT and UCP from Google, and maybe something else from Anthropic, whatever, right? 

[00:36:40] Nikhil Raj: There'll be multiple protocols. You gotta integrate with those, deliver those embeddings and those retrieval, whatever you wanna, you have to play. And you will play, and you'll either build the technology yourself or there'll be third party vendors, tech startups that will 

[00:36:53] Nikhil Raj: give you the capability. And we are doing that for some of our customers, right? 

[00:36:56] Nikhil Raj: So this is the inevitable [00:37:00] evolution of how this, this industry marketing will follow the consumer. And when the consumer goes somewhere, marketing and media and advertising, we follow them there. That doesn't mean it's the only thing that's ever gonna happen. 

[00:37:12] Nikhil Raj: Like Google is not gonna let their search business get cannibalized overnight like that. 

[00:37:16] Nikhil Raj: Meta is not gonna let the feed advertising business just 

[00:37:18] Nikhil Raj: go away to chatGPT like that. They're gonna, 

[00:37:22] Nikhil Raj: they're gonna figure out ways to, to continue. And it's gonna, as long as you have consumer distribution, you're gonna be relevant 

[00:37:28] Nikhil Raj: and you're gonna attract marketing dollars. 

[00:37:31] Nikhil Raj: And then you'll, you'll, you'll have retail media propositions on that, and then meta out will enable that. 

[00:37:35] Nikhil Raj: And I think that's, 

[00:37:36] Nikhil Raj: we are seeing ourselves as the enabling piece no matter what the protocol of interaction is. Could be a pre-bid JS request, or it could be an agent e-commerce 

[00:37:45] Nikhil Raj: request. For us, it's another way to just connect our customer's data 

[00:37:50] Nikhil Raj: to ecosystems that they wanna connect it to. That's it, that's the way we see it.  

[00:37:55] Scott Messer: Do you, do you think like the LLMs and AgTech browsers [00:38:00] will start off by sending data back to the  

[00:38:03] Nikhil Raj: I think that is in this way and giving you  

[00:38:04] Nikhil Raj: that's exactly what's the deal with, with Gemini and Walmart. Conversion data is owned by the retailer, not by the, not by those platforms, and that's why they did those.  

[00:38:14] Scott Messer: And not. And when And when we say conversion data, are we also talking about like getting an audience id?  

[00:38:20] Nikhil Raj: own the whole  

[00:38:20] Scott Messer: I guess. I guess. you have an No, they own the whole consumer. They own the everything. Because they are the ones for the system of 

[00:38:26] Tom: The emergent of record. Yeah. 

[00:38:27] Nikhil Raj: Yeah. They're the merchant of record. They're the ones owning the customer relationship, the return, the 

[00:38:31] Nikhil Raj: shipping, the loyalty program integrations, all that happens on the Agentic Commerce integration.  

[00:38:37] Scott Messer: But they wouldn't see like all the other things They browsed 

[00:38:41] Nikhil Raj: They would see the  

[00:38:41] Nikhil Raj: conversion. They wouldn't see all the other things they browse, correct? No. That is just like you wouldn't, if you didn't get the click from Google. If I did 10 searches on Google and I didn't click on anything, you are not getting the 10 searches. You're only getting the one I clicked and you're getting the one I click. 

[00:38:57] Nikhil Raj: Click to your website, not to the other websites either. Same [00:39:00] idea.  

[00:39:02] Scott Messer: Yeah. I, I just, um, it's interesting to think that, uh, the agentic browsers like have to deal and play ball. With the retailers, if it was a regular publisher, they would be like, sorry, no data for you. Thank you very much. Um, but because the retailers have the products, they're bringing budgets, they're very important to the ecosystem on multiple fronts. 

[00:39:28] Scott Messer: They're, those LLMs and Gentech browsers are willing to say, okay, table stakes. We're giving you data back.  

[00:39:35] Nikhil Raj: I think for them to, retailers have learned in the last two waves, right on search and social that they got locked out and now they're not gonna get locked out. So they're already demanding that out of the gates and they'll get that data. But here's actually, there's another unrelated points card. That's a great thing you're talking about. 

[00:39:51] Nikhil Raj: Agent browsers, one of our largest US retail 

[00:39:53] Nikhil Raj: customers had a 30% increase in 

[00:39:56] Nikhil Raj: traffic. Overnight, which [00:40:00] was all agentic browsers that got right through the browser bot detectors. These agentic browsers are posing as humans and getting right through the, the bot detection systems these people employ. 

[00:40:14] Nikhil Raj: We've been asked, and we've started to show them results, is can your edge first mile AI. Learn from known agents, like Google Bot comes in and tells you, Hey, I'm Google Bot, I'm gonna get your site. And it's allowed. And some other bot comes in and doesn't say that I'm, it's a bot, but works like a bot, clicks browsers and does things like a bot. 

[00:40:36] Nikhil Raj: So our real time AI learns that that on matches and any new session starts that behaves like that. We immediately predict it's a bot within two events. We know it's a bot. So instead of blocking it, we are giving this customer the signal that 30% of traffic that came in, these are the 25, 30% of them that [00:41:00] are the bots. 

[00:41:01] Nikhil Raj: The other stuff is not, so you don't know which 30% your traffic went from 10 million to 13 million, 15 million. You don't know which 5 million is the bad stuff, and we are marking it for them in real time. Right. The sort of first mile thing that you asked about earlier, Tom. Allows us to not only do audiences, retail media, behavior, consent. 

[00:41:20] Nikhil Raj: Now we're getting into bot signaling from our position in the cloud in the front of the, in front of the consumer on behalf of our customers. Sorry, Scott. I just  

[00:41:31] Scott Messer: Because,  

[00:41:31] Nikhil Raj: Yeah. Bring it back to you. Yeah.  

[00:41:34] Scott Messer: theoretically, you would've built an ID on those bots.  

[00:41:38] Nikhil Raj: And then you would update it on Facebook, and then you would've shown an onsite ad server email impression to a bot. Literally tomorrow I have a call with that customer to turn off their ad server, onsite ad server when the bot is on the site. 

[00:41:55] Tom: Yeah, makes 

[00:41:56] Nikhil Raj: about the, think about the wasted roas there, right? 

[00:41:58] Nikhil Raj: You're taking [00:42:00] a much higher ad spend on the bottom of the ROAS equation and making it much smaller, and the ROAS goes up just by blocking the bots. 

[00:42:09] Tom: Yep. I didn't think about this. There, there was, you know, I think perplexity kind of like turned this whole thing into like, aren't we gonna be starting to show ads to bots? I didn't think about the problem of just the fact that the ads won't get turned off if you don't know who the bots are. 

[00:42:22] Nikhil Raj: So what's going on for, for MetaRouterer now? Because we are sitting on the front end, like I said, the two parts, right? We're sitting in front of the consumer. We are processing the data inside the customer's cloud. We are now passing that bot intelligence back to the server side infrastructure that we have. 

[00:42:37] Nikhil Raj: We also have the id, so we are turning off audiences in third party media platforms. We're turning off web analytics data, so your analytics bill is lower. We're turning off data from going into Snowflake, Databricks, and now we are turning off ads to bots on site. 

[00:42:56] Tom: That's great. I mean, I think, you know, this brings it all back [00:43:00] in terms of, you know, the initial pitch for retail media was. Higher efficiency. Don't waste a bunch of ad spend if it's not somebody who's gonna be a buyer. And so I think you're doing that in, in every channel and, and everywhere. So really wanted to thank you for being on the middleman. 

[00:43:15] Tom: This was a great discussion. Um, and thank you for, uh, connecting not only you know, the dots for what MetaRouterer does, but all the players in the industry. So thank you. 

[00:43:26] Nikhil Raj: Thanks Tom. I'm, I'm so glad we connected despite me not seeing your message on LinkedIn. But 

[00:43:32] Tom: It's, it's even more fun when it takes more time, so. 

[00:43:35] Nikhil Raj: yeah, yeah. Good things come for those who wait. Right. So there, here we are. So I hope you enjoyed the conversation. Thank you for having me. We'll be in touch. Thanks  

[00:43:42] Nikhil Raj: Jets. Bye now.  

[00:43:43]  

[00:43:47]