From Chat to Checkout: MetaRouter and the Rise of Agentic Commerce

What happens when shopping moves from websites to conversations? In this episode of The First Mile Podcast, host Dom Burch sits down with Patrick Harrington, Head of AI and Machine Learning at MetaRouter, to unpack one of the biggest shifts in digital commerce yet — the rise of agentic commerce, where AI agents like ChatGPT drive discovery, comparison, and checkout without ever visiting a website.

From Walmart’s OpenAI partnership to the mechanics of the Agentic Commerce Protocol (ACP), Dom and Patrick explore how “chat meets checkout” is rewriting the customer journey, and what it means for data visibility, attribution, and personalization. Patrick reveals how MetaRouter acts as the first-mile control plane for this new commerce frontier, restoring lawful visibility and intelligence between AI agents and merchant systems.

We cover:

• How agentic commerce changes discovery, attribution, and personalization
• Why traditional analytics go blind when purchases happen inside AI chat
• How MetaRouter bridges the gap between AI-driven conversations and merchant systems
• The role of intent embeddings and policy tokens in powering lawful, AI-ready data exchange
• What “surface fluidity” means for omni-channel commerce and Retail Media Networks

Transcript

SPEAKER_00: 0:09

Welcome back to the first mum podcast with me, Don Birch. This is the podcast where we explore the very beginning of the digital data journey where every click, every conversation or interaction first takes shape. Now listen, when I'm away on my vacations, I don't normally check the stock price on the golf trip. But there I was in Portugal with my mates when I saw Walmart had jumped 3% overnight, and then I actually messaged the MetaRouter group chat and said, hey, what's going on? Why are they suddenly at 107? Turns out it's all down to Walmart's partnership with OpenAI, letting customers shop directly through ChatGPT. No website, no app, just a conversation that ends in a purchase. And at that moment, right there might mark the beginning of agentic commerce. That's right. And I said it out loud, Agentic Commerce. When commerce starts talking back, or as somebody else put it on LinkedIn the other day, when chat meets checkout, which I quite like. So to unpack all of this, I'm delighted to welcome back, and it's only been a few days, Patrick Harrington, who's head of AI and machine learning at MetaRouter. Patrick, welcome back to the first mile.

SPEAKER_01: 1:18

Dan, thanks thanks for having me, and great to see you again.

SPEAKER_00: 1:21

Like literally, I go away for a few days and boom, another major shift in our world. So we're just off the back of discussing GO versus SEO. Now this. So for people who've been hiding under a rock, just explain a little bit about what's going on and why this is such a seismic moment in the world that we're in.

SPEAKER_01: 1:38

This is a really interesting, call it historical fork in the road moment. And the direction of the fork has been chosen by the largest retailer in the world, Walmart. So the Wall Street Journal reported on the integration with Walmart and ChatGPT for orchestration of within ChatGPT commerce, meaning I can search for things, I can have conversations about maybe some project I have, and ultimately check out transact all within the app. So they Walmart now joins two other entities that have enabled this capability within ChatGPT, one being Etsy and two being Shopify. Both of them, obviously a little bit more in the long tail of retail commerce in terms of unique goods and services that both supply in different facets. But Walmart now is bringing, call it an 800-pound gorilla to the room perspective on where they see the puck sliding for uh agenda commerce. And it appears that they have a strong belief that within ChatGPT can help foster discovery and conversion benefits from such. And it's not a fringe one. It is akin to what Google search was in terms of discovery and the conversion funnel that follows it. And it appears that uh unlike the Google tactic of the past 20-25 years of referral traffic, sending traffic to retailers, digital services that they own, app, web, etc., this approach is all within. And so what's very unique about this relative to other discovery mediums is in this example, ChatGPT has full observability of top of funnel conversation, goals, back and forth, and conversion, bottom of funnel conversion visibility, which Google has never had, and they've been always trying to triangulate implicitly what people are buying because it helps buoy their advertising algorithms.

SPEAKER_00: 4:12

And it's called, you know, I think you might have even told me this. It's chat GPT, not ask GPT. I can't even say that. Ask GPT for a reason, right? Because you do have that back and forth. So, you know, let's walk through an example then. So a user on Chat GPT says, hey, I'm looking for a cordless drill, and then it's gonna come back with a selection of options, I guess, based on I guess understanding that user already to some extent. Maybe I've given a bit of description, you know. I'm trying to do a DIY project. I've got a, you know, the wife's having a go at me. I want to fix the bathroom. I'm not that good. Um, you know, my price range is sort of like mid. Um, what should I buy, right? ChatGPT, and it's gonna come back with some ideas. And then from there, how's that working? Like, what's going on in the background? What's the merchant seeing at that point? What's the, you know, what's really, what's the technical bit that's sitting behind that? What seems like a fairly innocuous conversation with my, you know, my robot friend?

SPEAKER_01: 5:06

Very well said. And and prior to getting into this, I mean, ChatGPT does maintain an understanding of the end user. I mean, you should ask it, what do you know about me as a prompt? And you'll be pretty blown away if you use it a bit. But what is under the hood of the car in terms of how this handshake between back and forth conversations, refinement, filtering, whether it be within the app uh typing, whether it be your voice while you're driving a car, um, these are now the services that are capturing intent. But what is under the hood between Jack GPT and Walmart is very interesting. The retailer in this example remains merchant of record. They own the relationship with the end customer, the fulfillment of the order, the orchestration of the payment associated with that transaction. So, how do how do those two worlds bridge together? A few months ago, OpenAI and the Payments Goliath company Stripe created an open source programmatic framework called Agentic Commerce Protocol with the acronym ACP. It's basically an APA API type uh handshake with a bunch of different uh capabilities to orchestrate agentic commerce and to be able to pass payment information without passing payment information. So, what does this mean here? Chat GPT is able to understand, you know, the back and forth and then ultimately present the consumer when they're ready to quote unquote add to cart, uh, that is when the merchant relationship starts seeing the demand. They do not see above that. So, really add to cart is where the handshake begins.

SPEAKER_00: 6:56

That's critical, right? Because up until that point in old world, I love that I'm describing, you know, literally four weeks ago as old world, but anyway, up until four weeks ago, the merchants seeing all of that before it gets to that point of sale, they're seeing so much information, aren't they? Right, so they're not necessarily now getting any of that journey that's gone before. So I might have been on Google, I might have seen an ad on Facebook or Instagram, I might have been having a conversation with somebody in a forum, and then I'm like, yeah, I really do need to get that cordless drill. But the first that the merchant might see is when I'm in Chat GPT and I say, Hey, I want to buy one of these things, it comes up with two or three, and I go, Yeah, go on then. Or not. And I guess the go there is one signal for them, but if I don't, does the merchant see any of that?

SPEAKER_01: 7:42

Let's say you have this really exhaustive conversation around that drill and the project that's accompanying the motivation for why you need a drill.

SPEAKER_00: 7:50

And my wife who's been telling me to do this project for the last six months.

SPEAKER_01: 7:53

Exactly. Unless you add that to a cart, of which that becomes the programmatic handshake with this example being Walmart, where their systems recognize the item that has been added to cart and they're awaiting that final authorization, that conversion, so to speak. They also see edits to cart, just to get that out there. But if you do not add to cart, that behavior lives solely within the realm of ChatGPT and OpenAI. There is no browse analogy that a retailer's e-commerce offerings would see when traffic hits their site, whether it be when they're searching, whether they're looking at things, et cetera. They only see, if you were to strip that behavior off, their pipes of ingesting that behavioral data stream really starts living at that add-to-cart moment. So much deeper down that funnel. And so this is a unique opportunity to try to maintain some degree of parity with those other consumer services, with this new discovery medium, which many of us know for other aspects of our life of high research intensive type projects. It's pretty good.

SPEAKER_00: 9:02

And to some extent, this is what crystallizes when we talk about the first mile, right? This is kind of what we're saying. Because what you've just described means that that traditional marketing funnel that we've all kind of got used to, right, is breaking up a little bit. So discovery, short listing, checkout, fulfillment, and all the little bits and interactions. And some of that might be weeks in the making, it might be minutes in the making, but all of those signals in old world are being collected, and at mass and at scale, you can do stuff with that data that says, okay, well, let's retarget that customer, or let's offer them a loyalty, or let's nip in with a voucher, right? So directionally, then potentially there's gonna be this big chunk of consumer behavior that's gonna be invisible to the retailer. But that that can't be sustainable, right? That can't be something that big retailers are gonna be happy with in the long run.

SPEAKER_01: 9:53

This is gonna be probably a multi-pronged lobbying effort of using carrots and sticks involving budgets, advertising budgets from retailers, from brands to try to price some visibility to augmenting this agenda commerce protocol to potentially invoke more visibility into interaction browse within that experience. But as it stands today, that add-to-cart, edit-to-cart conversion down is the extent of the visibility. And so the first mile interpretation of this is when those API calls associated with ACP, Agenda Commerce Protocol, associated with add-to-cart of the drill at interest for uh DOM in this example, where there is a payment token that Stripe effectively maps your credit card coordinates to a token, a privacy-safe token, and then it gets passed along to Walmart in this example, that then they own the payment rails, Stripe or otherwise, where that content gets orchestrated through their payment uh fulfillment systems, their fraud systems, etc. And ultimately the merchant is uh controller of data and merchant of record. So in order to extend those existing call it data monetization efforts that the retailers have in place that they use with browsing or add-to-cart abandoned cart retail media efforts based upon behavior of shopping at the end of the day, having some extension, a hedge, so to speak, to the funnel as it stands today in this new discovery, this new consumer channeled medium surface, whatever vernacular you want to use, is critical to then hopefully draw some creep back into upstream of that funnel that they currently have visibility today. And so the first mile moniker really extends from the existing web app physical storage sales coming off a point of sale to ChatGPT in this example, where that ad to cart down through conversion comes through. In this example, MetaRouter, who provides first mile capabilities, sees that before any advertising destination, before any routing to data assets, and is now in a position to take action on that data while it's mid flight. The same way that we talked about before when it's nested in web activity app or even omni-channel physical store uh sales coming off in real time.

SPEAKER_00: 12:30

I'm sat at home, right, listening to this, and I'm a marketer going, hang on a minute, this reminds me of like what's just happened in the world of TVs, right? We spent 20 years going HD, ultra HD, ultra 4K HD with the crystal TV that you know, right? And then people started streaming live sport from the internet onto their big screen in the house, and suddenly like the picture quality goes backwards, but people are like, yeah, yeah, yeah, don't worry about it, because the convenience of getting it now without having to, and it's almost like whoa, whoa, whoa, as a marketer, I'm going, but I'm gonna get like less attribution, I'm not gonna have visibility of what actually drove the conversion on a chunk of my customers, right? I'm gonna lose some personalization, I might lose the ability to re-engage with people who are hovering, maybe they were gonna buy, didn't buy, like all those things, and maybe my retail media networks aren't gonna be as efficient anymore. Hang on a minute. I was driving them to be like 5x more efficient this this next year, not like less efficient. So, I mean, these are big things, aren't they, that are occurring. People might not yet have kind of the penny might not have dropped yet. This potentially the direction of travel means there's gonna be quite a deterioration in what people have just assumed as the status quo, and if they're like any marketeer, is hoping was gonna get better and better and better.

SPEAKER_01: 13:47

This reminds me a lot of when mobile web, mobile app came out, like when we were at probably Walmart together, and there were differences in conversion characteristics associated with the traffic that hit that quote unquote new consumer surface at the time. We're in the same same game again, a different surface, right? And in order to ultimately control the marketer's persona type, you know, who has expectations, who is maybe uh performance-driven compensation is a function of the business's marketing targets, right? They're in a transient right now, right? And there is a need in order to control, you gotta first measure this new, call it, uh consumer surface. And having visibility into the behavior that is coming off of this new surface to understand, to route, to monetize from a marketing persona, like they do in other existing, call it classical channels, desktop web, mobile app, mobile web, uh physical store sales, there's a need to maintain that parity and extend to this novel surface, which we know is a potential big one as it when you think of discovery top of funnel search, right? Um, this is the first kind of challenge to Google, Google search business, the quote unquote perfect business model that we've seen in the past two decades. And so having an extension of the capabilities of what behavior do I have visibility into now that is normalized to my behavior that exists in my existing integrations, my existing revenue and margin and consumer behavior that I, as a marketer or chief data officer, extend into monetization efforts, downstream training of various machine learning models, et cetera, maintaining that parity to this novel N plus one new surface is critical to then begin efforts to potentially work their way up in gaining additional visibility further upstream in the funnel beyond Ad Descartes in this new modality.

SPEAKER_00: 15:55

So, what do you think is going to happen? And how does MetaRouter fit into that? I mean, you've kind of touched on already, right? But there's potentially an opportunity here for companies to really grasp, oh, okay, that's what you guys are doing then. That's the the bit at the start that I hadn't quite appreciated until now, but now I'm sort of slightly worried about not having visibility of some of the things I have got you grown used to. Just describe it to people in a way that, you know, they're sat there at home with a mug of tea in England, I'm thinking, got their nice Yorkshire tea. Just explain it in a way that people can wrap their head around because it it can get quite technical quite quickly, but just walk us through that kind of step by step.

SPEAKER_01: 16:36

The goal of the pro the marketing persona is to orchestrate various campaigns, whether they be brand level, brand and behavior level leading to different audience segmentation. We're all familiar with this, right? We've all been appreciated, I think, a little bit more of a personalized tone of voice with a lot of the marketing-based mediums. I, for one, do. So if you look at the probably what's going to happen, a monotonically increasing volume of conversion and revenue in this new modality, say within ChatGPT, and I'm sure this will extend to others, perplexity and Gemini, et cetera, being able to have those same type of campaigns that are at least orchestrated on abandoned cart-based tactics, post-transaction. People who bought this may like that, retail media-based efforts, where there is uh add-to-cart or conversion-based behavior of the brands corresponding to the items being bought. Marketers are going to need an answer to this novel and probably super exponentially growing critical new channel. And in order to do that, they need to have visibility around proper consented, normalized data in real time of where that data is going, akin to what they're doing today with their existing assets in desktop, mobile, and store. And so that allows them to go to their the board, their C level, et cetera, to say, hey, we have a hedge in place. We are activating audiences based upon this new modality. We are monetizing the behavior in this new modality. Let's let this run. Let's observe what the statistics are across different modalities and how we think of the future of budget allocation relative to driving the ROI. So that first mile posture really does allow you to extend those capabilities and manage it in real time. As those agentic commerce protocol calls, API calls, are being hit. They're being interpreted the same way in the first mile as any other behavior, so to speak, across the other mediums that we discussed.

SPEAKER_00: 18:55

And that real time, right? It's so easy to say that phrase, isn't it? But actually, real time being you know, milliseconds from me clicking or switching or flicking or looking or saying or whatever, right? Browsing or changing the chat or pausing after that response, hovering over the three items, clicking on one of the three items and coming back, and then picking up my phone and getting that instant ad on Instagram that says, Come on, let's get you over the line, right? That's that's what's happening at the moment, right? That's what happening. People are maybe not fully appreciative of certainly at a consumer level might not notice it as much as we know that's going on. And some of that rug is being pulled from beneath a chunk of people at the moment who are starting that journey or finishing that journey or halfway through that journey within a large language model, and as you say, it's not just gonna be Chat GPT today, it's gonna be you know everybody's large language model tomorrow, right? What should that look like? So let's thinking about, you know, maybe let's rewind back and we'll go up a level and go, okay, so we've got Stripe and ChatGPT have already figured out some of this and they've got an ACP. What are some of the other things that from your experience of 20, 30 years of Google kind of gradually almost like commercializing how the right way to do this is for customers? Because when we were talking about geo and SEO, right, it was similar in a way, wasn't it? It was like, how do you practically help the big companies are gonna grab a large chunk of the conversation, let's just call it that, the conversation, the traffic, the people who have intent, and actually find a way for them to integrate at a kind of industry or framework level that then allows everybody to play ball in a way that is relatively consistent, which ultimately means people get what they want, right? Because you know, if I get frustrated with my large language model not giving me back the three cordless grills uh grills, cordless drills I actually want, then I'm gonna jump back out to search, right? What needs to happen?

SPEAKER_01: 20:57

Well, keep in mind the Google's whole claim to fame in the late 90s was page rank. The concept that different websites, different publishers, different news outlets, retailers had a context of a quality score. And that was used to help buoy the results when someone searches for things, uh, which previously was devoid of that quote unquote quality of the end uh result. There will be an extension of that type of concept into this new modality. If I look at you know, some cordless drill sold by whom, right? And now those individuals that comprise those who could fulfill that order, there's going to be some concepts of quality or affinity, brand affinity, general mathematical alignment to DOM, maybe separate that and maybe different than someone else, right? And so there will be this additional alignment exercise, which takes your past behavior, what it knows about you, and aligns it relative to the context of the conversation, the back and forth, and then the selection. There will be such. It's not clear yet what that paid advertising model paradigm will be to call it affect that alignment, but we all know it's going to come. And what's really unique is if you have that ability to muck with the alignment and you see bottom of a funnel conversion, you can close the loop. And that's really why this is such a big deal with an entrant like Walmart at the scale that it operates globally, with not seeding that observability, but being content with having another party have that observability. But they strongly believe that they can, they're there to fulfill orders at a low price and ultimately view this as a mechanism to drive and potentially even dislodge competitors. So having a dog in this fight is going to be critical in the in the war for this revenue in this new channel. And the role of a marketer or a chief data officer or a privacy officer is going to come through. What can we do with that data? How can I monetize it, often at higher margins, especially with retail media networks, and have a hedge in place to experiment and iterate, but call it capture that tailwind of growth and revenue and adoption, and then marry it with existing capabilities and investments without having to necessarily go and reinvent the wheat.

SPEAKER_00: 23:26

Now we're at 25 minutes already, right? Which is bonkers. So what haven't we said that would be useful, helpful?

SPEAKER_01: 23:33

One key distinction of this ACP uh protocol that is worth noting. We talked about GEO before. Um basically how to let's see a new SEO analog. Part of ACP are when a retailer like Walmart, like Etsy, like Shopify integrate with this, call it commerce enablement of ChatGPT. ACP provides a very structured API endpoint for retailers to publish their catalog into them. So there's no crawling necessary here. The work is being done on behalf of the retailer who is publishing, enumerating through all their catalogs containing the name, the brand, the price, the URL of any media, images, videos, et cetera, that all then used for the LOM, in this case ChatGPT, to index and represent and then drive alignment based upon the conversation to that geometric representation of these items in a catalog that marry the geometric representation of the conversation. And so it's a bit of an interesting change here, both in terms of everything we've spoken about, but retailers are effectively publishing their catalogs in, and they did this with Google Shopping and other things as well. And so we see a lot of analogs taking place here.

SPEAKER_00: 24:53

That's interesting, isn't it? Because it's almost like over the last three years, two and a half, whatever, ChatGPT has almost warmed up everybody to this idea that we're gonna come and get it, or you've already given it anyway. So the scale of Walmart says, hang on a minute, we know how to do this. We can plug in to your thing and make it easier for our customers who happen to be on your thing, find the things that they would be looking for from us anyway. So it's quite in a way, you kind of go, Walmart are like being a bit savvy here, aren't they? Either they get dragged there kicking and screaming in six months' time, or they go all in up front. It's quite a Walmart play to some extent, isn't it? It's kind of like, what are we doing? Like, are we doing it or are we doing it? We're doing it.

SPEAKER_01: 25:40

I will say, after a few sleepless nights after I saw that, because it was such a big splash in the direction of Call of Commerce going forward, and Walmart does take some legitimate principled risks. They have an incredible fulfillment network, relationships with brands and merchants, and the ability to negotiate on price and are happy to fulfill any incremental demand with their considerable investments and capabilities in moving products and goods around the world and selling them in a storefront agnostic medium. They want to be there, right? And they certainly have come out of the gates, and there's probably a first mover advantage with them doing such relative to their competitors.

SPEAKER_00: 26:23

Now, if I could ask you just to recap on the bit that matters, right, in the first mile, because I guess three things come to mind, right? One, you want to maintain that visibility, right? So in this agencera that we're now in, you want to see what you want to see, you want to see what you've already seen before. But then there's also this bit about making sure that you're enabling lawful privacy-first data routing, right? Because that's something that is increasingly the bar is being raised globally, whether that's in California, whether it's across Europe, right? That that bar is being raised, it's not going to come back down again. But also, you want to be able to power AI-driven personalization and retail media network measurement. It's almost like we don't want to lose what we've always had. And you know, I guess that's where you're I mean, I've seen the conversations, right? And you know, we can't share all those conversations on the podcast, right? But I'm lucky to be party to some of the thinking as it is emerging within the within the team. But that as that crystallizes out, right? That's why having as an organization that has a lot of data, really understanding how you own that data, how you control that data, how you route it, to whom and when, and also how you then make sure that when customers make a decision like, don't show me that, I don't want to be part of this, that you're still able to do all of those things. And as this world changed, I mean, like we were literally only talking about something three weeks ago, who knows what we'll be talking about in three weeks' time, right, Patrick? But but that's I mean, that's so important, isn't it, in terms of that first mile. While why control matters so much if you're a an organization or a brand or a retailer that has a lot of customer data.

SPEAKER_01: 27:57

If you look at this new consumer surface, and this is not a fringe one, this is going to be increasingly more critical as a key source of revenue and likely high conversion rates that go with that revenue. The ability to take the data associated off that stream, that new modality, and capture and enforce consent of how that data may be used for advertising purposes. A lot of the capabilities that we at MetaRouters support our existing customer bases, we are able to suppress the routing of a potential event corresponding to a consumer if they have not yet uh consented to advertising. Right? So this plays well in the EU, the GDPR, and stateside increasing legislation. California is obviously uh out of the gate. But it's not that that data makes its way down to some last mile data destination of which an audience is created. It is not clear in the moment what that consent reflects. We are able to suppress or allow that data to flow through to various destinations, whether it be bestowing consent for advertising purposes but and analytics or some people with analytics but not advertising, et cetera. And that first mile, that real time and that data is in flight, we're able to simply deploy business logic that checks for that consent and simply doesn't send it alone, right? And so maintaining that parity, you know, if I'm a chief or privacy officer, compliance, and I'm looking at and supporting our the marketing efforts of a large company that is trying to maintain, call it marketing parity in this new important channel, this is going to be a risk consideration that comes up, and they're going to be looking for solutions that are able to do this in real time such that they don't have any litigious risks being uh you know, case in their armor, as they do want to make some critical business decisions and move quick in this new. Modality and be able to report that back to their leaders, the board, shareholders, etc.

SPEAKER_00: 30:06

And there'll be people listening who might, you might have just piqued their interest in that last few sentences and they're going, oh, who can I trust to have a conversation about this? And like without being too salesy about it, pick up the phone, right? I mean, there are guys here at MetaRa who can walk you through that conversation in confidence and help you figure out some of this stuff, right? Because it is going to be setting alarm bells off. If you're head of compliance or you in charge of data security, this world is moving quickly. And you might be just sat there going, okay, who can I just have a conversation with this about? Who can who knows what they're talking about in this space?

SPEAKER_01: 30:38

That is a great point. Our the team uh is on the road a lot, whether across Europe, folks were just on the East Coast this past week, and throughout California, throughout the entire world, right? We're having conversations with leaders right now, helping them think through this changing landscape. Uh, and especially coming up to where budgets are in flux right now, as they look at everyone's definition of a fiscal year. These are conversations right now that are being had. And uh literally just last week I've had a few. Helping folks think through this new world, which we know two quarters in is going to be profound, let alone going into the holidays, where there's much more of an uptick in commerce-related activities. So uh the team is very good based upon the backgrounds of individuals of helping leaders think through this. And uh and it's also on a on a personal note, I think what's really fun about this because we get to see the whole gambit of differences and similarities across a lot of conversations. And we we do recognize that this first mile capability overlaid with where deep expertise in AI and retail media and the advertising ecosystem and experience and omni-channel commerce and identity come together as a confluence of themes to really support and normalize data and capabilities of this new critical modality, this new consumer surface, with all of it, with the goal of explaining all 100% of revenue. That's the ultimate goal. All the data, all the behavior that comprises the business of revenue, we want to be there to support in the first mile.

SPEAKER_00: 32:16

Patrick, always a pleasure. I couldn't have summed it up better, so we'll leave it there. But for the time being, thank you so much for being on the first mile with me.

SPEAKER_01: 32:23

It's always a pleasure, Don.

SPEAKER_00: 32:24

So that's a wrap. Today we explored why the first mile control has never been more important. Maintaining visibility in the agentic era, enabling lawful privacy-first data routing, and of course, powering AI-driven personalization and retail media measurement. As Walmart and OpenAI bring shopping into conversation, MetaRouter is building the data foundation that keeps merchants connected and compliant when discovery, engagement, and purchase happen everywhere at once, right? You've been listening to the first mile with me, Don Birch. Thanks for listening.