Agentic Commerce in Focus with ReFiBuy

Announcer:
Welcome to TD Cowen Insights, a space that brings leading thinkers together to share insights and ideas shaping the world around us. Join us as we converse with the top minds who are influencing our global sectors.
John Blackledge:
Hi, we are live from TD Cowen's 54th annual TMT conference. I'm John Blackledge, internet analyst here at TD Cowen. Pleased to have him on this podcast, Scot Wingo, CEO of ReFiBuy. Thanks for joining. This is old hat for you.
Scot Wingo:
I produce sometimes three podcasts a week, so I love the podcast format. It requires zero prep, you just go in and chat. It's my format.
John Blackledge:
That's great.
So we had you earlier today on a panel on agentic commerce. I though you did a great job. Super informative. I think the first question I have for you is just defining agentic commerce and also the debate around the definition. Maybe we'll start there.
Scot Wingo:
Everyone kind of has their own definition of agentic commerce and mine has a pretty big tent. So I think of anytime a GenAI or an LLM is helping someone research, find, or buy a product is agentic commerce.
And it can happen on a variety of different services. The most common use case that comes up in your and my world is these answer engines. Which would include ChatGPT, Gemini, Perplexity, Meta has one now, so on and so forth.
However, I think Rufus is agentic commerce. I count that. So it's a retailer hosted agent that's helping people find products in a different type of format versus the old format of the onsite search. Or even if you went to an OpenClaw and had just like a headless agent go do things for you, I count that as well.
I ended up getting in these debates. I was at Shoptalk and it was kind of like a famous thing. And it was spurred on through an online debate that I kind of jumped in the middle of. And there's a set of people and they tend to come from the retail media network world and then also the payments world. And they like this definition of agentic commerce that's very strict. And it basically says you, John, wake up one day and you're like, "I need a new pair of shoes." You tell an agent you need a new pair of shoes and then that's it. So in their definition, it's fully autonomous is kind of like how I think about it.
I know you follow AV autonomous vehicles and it's almost as if you kind of said, "Well, I'm only going to consider autonomous vehicles like where Waymo is now." But it took Waymo 20 years, 25, 30 years to get here. So that's their definition, which is fine. Everyone can have their own definition. But what they then do is they use that definition to say, "And you can't do that today, therefore it's not a thing and you shouldn't worry about it." Giving people advice not to pay attention to this I think is not good advice. So that's why I want to jump in there and say, you can have your own definition, but here's how the way I see it. And there's a lot of activity going on and you can't just ignore this.
John Blackledge:
To your thing on autonomous, Waymo's at level four. There are other players at level two, level three. And so what would be the compare, I guess, in agentic if you're looking at it similar to the spectrum for autonomous?
Scot Wingo:
Yeah, to help in this debate, I actually borrowed from the autonomous vehicle framework and I came out with the five levels of autonomy for agents and agentic commerce. And so we start at zero, which is you're doing nothing. So Apple is kind of like the one player in our world that's there. They have all these assets and they're not really bringing them to bear on this at all. They've really struggled with LLMs and GenAI and maybe they'll get Siri off the ground this year, knock on wood.
And then it goes all the way to level five, which is there's certain OpenClaw use cases. OpenClaw is like a little window into the future. I don't recommend people ... You have to be kind of a hobbyist. But it's kind of like this hobby way of seeing one year into the future. So I keep a close eye on it and I play with it in a safe environment for this reason. Because I think you're already starting to see ... I was just out at Google IO and Google Marketing Live. They came out with a consumer agent called Spark. And it's the first inklings of them adding their own OpenClaw like thing, but they're going to put Google-esque privacy and protections around it.
That is like your first inkling of fully autonomous. Where you can tell an OpenClaw ... I did a demo with a friend where he had the Ray-Bans and you can hack the Ray-Bans to tie it to an OpenClaw. And we could look at these microphones and say, "What is that microphone? I want to find one online and order it." And it will go do that for you. It can do it fully autonomously if you want, or it can check in as you go. So I think we're going to get there. You'll start to see these things in the next year, again, on these edge cases. And then over time the technology will be there. And then it'll take three to five years for consumers to get there.
But we have consumers at what I call level four. So many consumers, the younger consumers and the providers are at level four in many cases. So Gemini, for example, now has a universal cart you can decide to check out right there. Meta has a cart that's tied right into ads. So we're going to start to see these experiences this holiday where people get to what I would call level three, which is pay. Level four is advanced buy and level five is a full autonomy.
I have a podcast in a Substack where I published this framework and update it. I just updated it coming out of Google where I moved them forward. And ChatGPT was at the buy and I actually moved them back a step because they've abandoned the checkout. So they kind of took a bit of step back on this framework. But I think we may get to it later, but there's a method to their madness there that we could talk about.
John Blackledge:
Yeah, that's great. Maybe we could talk about adoption. You just mentioned maybe there's a certain cohort of people in the level four arena. Maybe you talk about consumer adoption and then who's leading it now. And then also the verticals within retail that are kind of the leading cases.
Scot Wingo:
Yeah. So I'm talking about answer engines specifically right now. And then we can talk about the onsite retailer engines if you want.
Within the answer engine category, the leader is Google. So Google is confusing. They have three flavors of their AI. The one I'm talking about is AI mode. So if you go to core Google and you look at the riot, you can flip it into AI mode. I know you know this, but for our listeners. And then you have AI overviews. You can't really transact into AI overviews. So AI mode. And then Gemini is really for super hardcore people to go kind of touch that ... Like get close to the actual underlying engine. So you can go to gemini.google.com and do that. Most people are using AI mode and that seems to be the future of where this is going.
You're in AI mode, you now have a cart that persists. It works in Chrome, it works in Gmail. It's going to kind of follow you across your Google world, your Google ecosystem. And at any time you could just start throwing products into that cart and check out at your leisure. And it'll alert you, it's a really smart cart. That's why I consider it advanced buy. If you're in a loyalty program, it ties automatically into that. It watches prices and inventory for you. And you may have something in your cart, it may say, "Hey John, you've had this switch to in your cart for one of your kids. It's getting low in stock. You may want to pull the trigger right now." Or, "There's a price drop and this is unusual. This is actually a pretty good time to buy." Or, "This is kind of a high point for this product. You may want to wait 30 days. It typically will dip into this kind of zone."
So it's really a tool you would never ... You could do that online, but you'd have to combine six different tools and really spend a lot of time on this. And just having it all in one centralized area is what puts Google at that ... They're kind of ahead of the pack.
Then everyone else is clustered in the next level, which is they're helping you buy but not advanced buy. So here we have Meta, Copilot and Perplexity are in that bucket. I used to have ChatGPT there, but they went back a step because you can no longer buy on ChatGPT. You can only research and find products. So they're on level two, which is the find category.
So that's the lay of the land with Apple on one end of the bookend doing nothing, but they have so many assets. When they get in the game, they could go fast. We'll see how that goes. And then the other end is the most advanced is Gemini at level four and everyone kind of right around the middle of the pack.
John Blackledge:
Makes sense. What verticals like CPG or consumables are kind of leading right now in terms of agentic commerce?
Scot Wingo:
Yeah, there's a lot of data out there and I kind of put it into two buckets. There's stated preference, which is surveys. And you guys put out a bunch of surveys, which is great. I kind of put those into the bucket and you do a really good advertising survey that I appreciate. And then there's observed preference.
Observed state data is more scarce and you kind of have panel data companies, so there's similar web and sensor tower. Those are two ones I look at. They cover a lot of Rufus data, which we can get to. And then you have Salesforce and Adobe. I'm sure you've seen all the ... Around the holiday, they were reporting like 400% increase in traffic from a small number. But the observed preference data is showing that this is lighting up for sure in the discovery part.
And then the stated preference, what you're seeing is consumers are warming up to it, especially younger generations. Retailers kind of smush together Gen Z and millennials and they call them Zillenials. And Zillenial females are really adopting this at a pretty good clip. And what they do is they use it to manage their life. It's become their personal assistant. So they may have their beauty routine in there, their meal planning, their fitness routine, their schedule. If they're a student, their student stuff in there. And it basically helps them run their life.
And they like the voice mode. I'm not a big fan of the voice mode, but they'll just chat with it and they give it personalities. It's wild to kind of watch the different behaviors out there. And it becomes like their personal assistant that has a personality. And that cohort is adopting beauty, fashion, and we're starting to see a lot of grocery as they do their meal planning and meal prep through there. And then we're also seeing pet come along for the ride. So those are the big categories kind of leading the charge right now.
John Blackledge:
Yeah, it'd be interesting as we get to the holidays if other categories kind of move up. But those all make sense. A big debate in the investor community is the implications for the retail media advertising marketplaces. You mentioned Google launched the universal cart. It's designed as a single cart across multiple merchants and services. The question is if the agent never visits the company's website, what are the implications for advertising on these large e-commerce marketplaces?
Scot Wingo:
In our world, we call it retail media network and over 60 or 70% of it is Amazon. With their $70 billion Amazon ads is what Amazon calls it. But the whole industry's seen Amazon success and replicated it. So now Target has an active program. Costco, Best Buy. Obviously Walmart has had a big one. So that has become a really big part of the retail economy. Uber, DoorDash, Instacart, they all have it.
And it turns out ads are a great thing to put in front of consumers and that's how ... Kind of goes back to network TV. There's a model there that if you're going to give the consumer something free, an ad is a great way to do it. And that's why ChatGPT is really leaning into this as well.
I can just give you the bull and bear arguments. I don't really know which way this is going to go. I think we need a year to see. I rely on you smart Wall Street analysts to tie it into what's going to happen in the stocks.
The bear case is, just like you said, that number one, if there's Gemini. If Gemini is answering and you're buying up there, you never go to the retailer site at all. So that's Checkmate, right? Now ChatGPT is actually ... Then the argument would be, well, ChatGPT's not doing checkout. We're okay. But what they're going to do is they're either going to send you right into the cart pre-populated so skipping a lot of the journey where you show ads. So they're going to pass all the billboards and just go right into the back of the store to the transaction.
So again, even if you are excited that ChatGPT's not going to be doing onsite buying up there, it still is not good for retail media. Because best case, they'll land on just a product detail page. But those typically, they may have an ad very far down on Amazon, for example. The juicy ads are in the discovery funnel.
John Blackledge:
Right. The sponsored product ads.
Scot Wingo:
[inaudible 00:11:40] at the search results page, that's the money. That's where 90% of this money is made. That's the bear case.
Now the bulk case is, well, people don't want an aggregated experience. They're still going to go to insert favorite retailer. They're still going to go to Amazon. They're still going to go to Walmart. They're still going to go to Target. They're still going to go to Lululemon and Anthropologie and all these places. And when they go there, they're going on a brand experience. And yes, we may need to move the ads from the search and we may need to get rid of search and/or augment it with our own agent. So everyone will have their own Rufus is how I explain it because people are so familiar with Rufus. And we're starting to see it, right?
So it started out, Amazon was the lead with Rufus. Now called Alexa for shopping, but I still prefer Rufus. And Walmart now has Sparky. Target has one that's wildly called TSA, they need to work on the branding of that. And Target Shopping Assistant, I don't think they though about it. They should have done Bullseye, but we'll talk about that. And then Lowe's has one called Mylow.
And what you're starting to see is they're starting to differentiate. So the Mylow, what's really good about it is it kind of knows what projects you're working on. And the Home Depot and Lowe's of the world, they have really good information on how to do a deck in your backyard. Or whatever project you're working ... How to fix a hole in your drywall, whatever it is for that DIYer. And it's kind of tied into that. Where Alexa for shopping doesn't really know much about that. It's such a horizontal. It has to know a little bit about everything. It can't get super vertical.
So their argument is we're going to get people to stay on our site and they're going to go into that. And that we're going to show them ads in there. And you're starting to see Amazon and Walmart are already putting ads into Sparky and Rufus. The format they use is called a sponsored prompt. Where you can kind of buy a prompt and say, "If someone's shopping for groceries, I'd like to introduce my organic milk to them." You can target these prompts on a variety of different things.
John Blackledge:
That makes sense.
Scot Wingo:
So we'll see. I think after this holiday, we'll have a pretty good inkling of where this is going to go. Because in the captive retailer agent, there's two or three formats up at the answer engine. So we have ChatGPT and Google/Gemini. I think between them, they have 30 ad formats right now. So there's definitely a ... We're going to have a lot of these ad formats and then we'll standardize on four or five. And I think in this next year, it's going to be really interesting to see which of these ad formats we kind of settle in on.
John Blackledge:
Right. That makes sense.
So you were at IO and before IO, it was announced that Amazon, among others, was joining Google's Universal Commerce Protocol. Just curious on your take. They held out for a little bit, then they've joined. So just curious your take on Amazon joining the protocol.
Scot Wingo:
So I think a lot of people assume that they joined this thing and they're going to make their inventory available. I actually think it's the other way around. I think they're going to try to soak up the UCP inventory and make it available on Amazon. That's my prediction. And we have inklings of that. So number one, you can't find Amazon inventory on ChatGPT or Google. They have purposely blocked all the bots. Oddly, you can on Copilot. I think it's just one that kind of leaks through some. I think they just don't care about Copilot because it's so small. And they don't worry about it. Or maybe there's a Seattle thing going on between those two guys. And there is no Venn diagram of I'm on ChatGPT and I can buy things from Amazon. They've basically blocked that at every possible way.
However, what Amazon has done is they've started a program called Shop Direct. And what that does is it now adds ... We've always had two buckets of inventory loosely on Amazon. First party and third party. First party is wholesale, third party is marketplace. And there's flavors in there of like, is it FBA and all this? But that's the basic hierarchy. We now have a third bucket of products and it's this off Amazon inventory that you can either go visit and buy as a shopper. Or a subset of that you can actually buy through an agent called Buy For Me is the name of their agent. And it sends a browser-based agent to go buy for you down there. The constraint is it has to have Amazon payments underneath it. Which is kind of rare for a retailer or brand to have. So that's why it's a small subset.
So I think what they're going to do is they will be able to have people buy on Amazon, off Amazon items using UCP. I think that's going to be the use case. Because all these people are joining UCP and they're doing it. I view it as a defensive mechanism because Amazon, I'm sure you've seen the flywheel diagram. I call it the Death Star diagram. It's like literally the Jeff Bezos napkin diagram of how the marketplace would look. It's like more selection brings more sellers, lowest price brings more consumers, and the flywheel spins faster and faster. It's a three-legged stool and one of those legs is selection. And I think Amazon is very keenly aware that they got here by having the world's largest selection. That's in the whole A to Z promise.
The doomsday scenario, the bear case for Amazon is there's this set of selection that's not available on Amazon that you can only get through UCP. And in there is all the brands that never would sell. There's Sephora and Ulta, like all the big retailers, Home Depot and Lowe's are in there. And then all these brands that would never sell on Amazon, like Birkenstocks famously won't. Nike's been in and out. I can't even remember where they are right now. A lot of the shoe guys won't do it. Adidas and all these guys.
So if you coalesced all that, you'd have an interesting set of inventory for the first time that's different and not available on Amazon. And that's a threat to Amazon on that part of the flywheel. So I think their strategy is let's embrace that selection and make sure it doesn't escape our ecosystem, that the prime user can still get to it and stay in the Amazon family. So I think it's a defensive mechanism, but we'll see, maybe I'm wrong. Maybe they will actually kind of come and say they've invested in tons in Anthropic and ChatGPT. Maybe someday they say we will make some of our selection available on these things through UCP. We'll see.
John Blackledge:
It'd be very interesting to see how it plays out. I also wonder how we think about ... They've invested so much in the fulfillment infrastructure and they're setting record speed of delivery seemingly every quarter for the past several years. And they just rolled out this Amazon now, which is 30-minute delivery. All that is to say, does the significant tens of billions of dollars that they invest every year in their fulfillment and logistics infrastructure, does that help them at some point as agentic commerce evolves?
Scot Wingo:
Yeah, it's huge. It's the biggest moat they have. However, if they become the backend, I don't think people like you would be as excited about Amazon if they're basically a 3PL. So they don't want to be commoditized. They want to own that front-end experience with the consumer. There's no existential risk. There will always be an Amazon around because of that. I used to say never. I think it's unlikely you would have one of these companies build that. But then look at their CapEx. Have you ever done the math of how much has gone into the build out Amazon has?
John Blackledge:
The mix between AWS and the retail other?
Scot Wingo:
No, no, the fulfillment. How much did it cost them to build out fulfillment? You should do a report that compares that to what is Anthropic and ChatGPT spending on CapEx right now? And I think it's actually small.
John Blackledge:
Yeah. Well, the mix this year, for instance, the mix that they do overall, the company does 200 billion in CapEx is what they got it to. Say 35 billion goes to retail, other 165 billion is going to roughly AWS. So still a big number. I mean, relative to-
Scot Wingo:
I don't think it's crazy that one of these people tries to build something. It doesn't have to be as big as Amazon, but there's a world where ... Or maybe they buy FedEx or UPS. We're in a world where almost anything is possible because we've got now these companies are going to come out in $2 to $5 trillion market caps when they come.
John Blackledge:
Right.
Scot Wingo:
Anything is on the table in that world. So we'll kind of have to see what happens. It's unlikely that someone would compete with them on that level, but someone may get serious enough about it that they do it. I don't know. Who would've said they would spend a trillion dollars on GPUs? And here we are.
John Blackledge:
Yeah. I mean, thanks so much for doing this. Thanks for doing the panel and this podcast.
Scot Wingo:
Sounds good. Yeah, I look forward to seeing your reports that you published out of all this.
John Blackledge:
Thank you.
Announcer:
Thanks for joining us. Stay tuned for the next episode of TD Cowen Insights.