S08, E03: Amazon and Agentic Commerce (with Adam Epstein)

Let's [music] get real. Half the time you're not sure if your campaigns work. You just kind of hope they do. Attribution numbers look good, but what if those conversions would have happened anyway? Incrementality [music] helps you find out for real. No tracking people, no guessing games, just clear answers on which campaigns actually move the needle. If you want to stop guessing [music] and start knowing, go to Incrementality. Because good marketing isn't about getting clicks, [music] it's about getting results. Check them out at incrementality.com. That's i [music] n c r m n t a l dot com, or click the link in the show description.
The problem is that the distinction needs to be drawn [music] between the competence of the economists and the correctness of their analysis.
Hello and welcome to the MDM podcast. I'm your host Eric Seufert and I'm joined today by Adam Epstein who is

returning to the podcast I believe for the fourth time. Adam, welcome back.
Thank you very much. How many fourth time guests have you had Eric? Am I the first?
No, I've had a few guests that have come back more frequently, but I don't think with as short of a span of time since the previous appearance. So you're coming back on just a few months after we last spoke, but your expertise is is deeply relevant at this moment because we're going to talk all about Amazon and the Gent of Commerce.
And I hear you have some strong views on the Gent of Commerce, Eric.
[laughter and gasps]
I do. We'll get into those. Before we do, will you please reintroduce yourself to the audience?
Sure. Adam Epstein, co-founder and CEO of GG. GG is the AI media manager for the Amazon DSP. We work with some of the largest advertisers in the world to change the way that they work in this media channel. Our customers tend to be large agencies managing some very sophisticated brands and we cut the operational work of managing the Amazon DSP by 80% and

immediately improve advertise performance. We're all We're also the recent winner of a global Amazon ads partner award. It's our back-to-back global award. It's a prestigious award with Amazon, so we're feeling really good about our progress and continuing to grow.
That's right. Congratulations. I remember when you told me that you were up for the award, I remember thinking it would be bizarre if you didn't win that.
Well, we did.
You did. And you look quite dapper in your acceptance photo.
Ah, thank you very much, Eric. I think it's the first time I've worn a suit professionally not at a wedding in a very long time. So, I try tried to make sure that I was good to go for the award ceremony.
I I'm permanently dressed like a just total slob and I I went to this antitrust conference last year or no, this year and it was in Europe and I remember thinking, "Ah, maybe I should I should get a suit." And I just kept putting it off and putting it off and putting it off. And then finally it was too late and so I just wore like slacks and a sweater. And I looked so out of place. Everyone

It was It was all lawyers and so and then me and so I looked so out of place. And so I'm I'm giving a presentation at an event in November. This time I really do need to dress This is a good reminder for me to to get some clothes up that that don't look like they were bought at Goodwill. But okay, my own sartorial choices aside, we're going to talk all about Amazon today. You're the You're the Amazon advertising expert.
Something like that.
Something like Well, I think so. It's my podcast. But I've been asking this I've got a question that I've been asking the last few guests as this topic of agentic commerce comes up because it's kind of like right at the frontier of of AI times, you know, commerce times consumer adoption. And here it is. Am I wrong about agentic commerce being a mirage?
So, I think the definition of agentic commerce and the prism with which you point this thesis continues to have a narrower and narrower foothold given the breadth and depth of

agentic capabilities that are materializing over the last few months, especially. You yourself wrote an incredible blog post recently about how agentic commerce for OTAs is actually not necessarily a mirage, but in fact it's a desirable customer experience and a positive economic experience for the entire travel industry. So, this this lens of what agentic commerce could be as a mirage is is having much more of a narrower scope as more and more agents are completing commerce transactions for a variety of different use cases, but within that narrow scope with which you choose to define it, I think that it continues to be a mirage. And so, the scope with which you define it, forgive me for paraphrasing and if I'm butchering your thesis, feel free to jump in, but it's third-party general horizontal agents being able to make commercial transactions on your behalf ubiquitously across the internet. And every single action that's been taken by people that control those

commerce transactions, i.e. retailers, as well as the consumer adoption of AI agents indicates that your thesis is spot-on as it appears to be right now.
Well, that's a great way to start the podcast. You're right, I did hedge. I I hedged. I hedged in a sense, but you know, the the original thesis was kind of anchored to e-com. I mean, look, it's a it's a salacious headline, right? Agentic commerce is a mirage. And if you know, if you go back and read the original piece which I wrote more than a year ago, I was really making a point about Amazon. And and maybe more broadly e-com, but really it was about Amazon. I do want to give people the opportunity to tell me that I'm wrong because if I'm wrong, I want to know. But okay, so it sounds like we're aligned.
Let's continue to chat about why it's a mirage. And so, I think there's a mirage because of economic incentives. This is something that you've continued to write and opine about very well is that the economic incentives of large retailers like an Amazon do not beholden them to

partner broadly and openly with personal AI agents and chatbots. And then I think the other piece that we're still forgetting is the customer experience is not objectively better for e-commerce within personal AI agents and chatbots. And I think we're just discovering that more and more as people begin to use these products or don't adopt these products on an ongoing basis. So there was a recent podcast, the Cheeky Pint podcast with one of the Colacean brothers and Zach and Alexander Wang. And what Zach was obviously talking his book in which he objectively said or tried to objectively say this is a flat-out better experience and retailers that adopt this are not going to move forward in this next wave of e-com. And I I question that proposition. Like what is an objectively better e-commerce experience from the consumer perspective that I want an agent to be taking an

action on. And I think the one very very narrow awesome use case was it what Alexander Wang said is that he's looking for super obscure vintage t-shirts around the world that he can pay hundreds of dollars for. And you know what? Having an AI agent find those for you is exactly the type of thing that you should be using an AI agent for. But I guarantee you that 99.9% of Americans are not using or want to use AI agents to purchase multi-hundred-dollar vintage t-shirts. If you think about the use cases like shopping and browsing for shopping is actually part of the experience. Whenever we talk about consumer adoption of AI agents, I always bring up my wife. And when I come home and my wife has some free time on her hands, you know what she's doing? She has her computer open and she's browsing e-com sites for clothes that she likely will not purchase. Because she actually wants to have that experience of browsing and thinking about it and seeing these brands. And so

e-com and this triumvirate of Shopify, Meta, and Stripe that's been created to fuel e-comm experiences is is generally speaking a lot in fashion and apparel. And in fashion and apparel, people want that browsing experience. Again, sorry for paraphrasing podcast that I've listened to in the past week, but I also was listening to the Knowledge Project podcast with Shane Parrish and the guy that's a real estate agent, this guy Ryan Serhant. And he very accurately said, "People don't want to be sold to, but they love to browse. People love to browse houses. People always are looking for new houses." But people do not want to be sold a house. And so, I think the customer experience that we're taking away when we think about agentic commerce taking all over these transactions is actually not objectively better. People genuinely want to enjoy that browsing experience. If you're spearfishing though for a very specific long-tail niche item, agentic commerce

can very easily be a desirable customer experience. So, Eric, you talk a lot about the commercial incentives, but I think people are at least within the echo chamber that we live in, people are underestimating the desirability of an agentic commerce experience relative to the current ways of interacting and purchasing products online.
You're right. I mean, I I try to ground my analysis more in the platform incentives and the platform economics, and I don't really pay as much attention to the consumer incentives. That said, I mean, I had Michael Munger on the podcast last week and he did a great job of of exploring that. That Shiki Pines podcast was I think very revelatory for my perspective because you had the one use case championed by Alexander Wang, which is this just very, very niche specific, very, very like high manual effort use case, which is kind of hunting down these these t-shirts. But you've also got a There's another

qualifier that's important here. On behalf of a consumer with essentially unlimited means.
Correct.
That That matters.
And limited time.
And limited time. That matters. That matters.
So, you're talking about a TAM of the billionaires in San Francisco. I mean, you're you're talking about a very small pool of consumers. You're not talking about the consumer experience broadly. But, in like Zuck's use case was I think utterly ill-suited to agentic commerce. Did you get me the stuff I need to bake bread every Saturday? Because you can do that. You don't need an agent for that. That's Instacart. Instacart, Amazon, they do that for you now. If you know which ingredients you need, and you know the timeline then there's no need to introduce an agent. I feel like this conversation is being anchored to these agents as they exist right now, which is they have

implemented no form of monetization outside of the subscription. And when you actually want to start scaling that agentic commerce and monetizing it, there's going to be a take rate. If they go the take rate route, which I think they're going to abandon, just like OpenAI abandoned it. That I mean, Zuck's talking about a take rate, but that's only because it's more palatable than saying we're going to stuff this thing full of ads. Because I he knows there's already anxiety about AI, and if you then start combining that with ads, it could be a non-starter. So, he's got to say, "Well, we'll just add a little bit of a premium on." Who's going to pay for that? Because if the retailer pays for it, then you're probably A lot of these retailers just they won't be able to accept that. And also, I think all these agents are going to be commodity. So, you just you factor the agent without a take rate. But, at some point, you're probably talking about the price going up for the consumer. And if the price goes up for the consumer how is a repeat purchase for something that is known in advance on a timeline that is known in advance, how does that justify the added cost relative to going

to Instacart or going to Amazon and saying, "Set this up on a repeat purchase." No one is going to be able willing to pay for that. And so then you either have a race to the bottom with no take rate on agents, which I think is exactly what's going to happen. And then you have ads. Or you just have the complete abandonment of agent e-commerce altogether. If these companies insist on taking a take rate, no ads, and people say, well, guess what? I'm already a Prime member. I guess I'll just stick with Amazon.
Well, there there isn't even going to be a race to the bottom on take rate because the one experiment that initially occurred was what ChatGPT had, and they immediately deprecated that because they realized it was unfeasible.
Yeah.
The notion that there will be a take rate to merchants on AI chatbots [clears throat] or personal AI agents just it ceases to exist. Like, there's not even going to be a race to the bottom. No one's even trying anymore. They realize that's not the primary mechanism to monetize this behavior.
Yeah, this is a non-starter. Okay. The last time you were on, we both

agreed that Amazon would likely explore some kind of an advertising integration with ChatGPT. The company's just announced exactly such a partnership.
Well, I won't say exactly such a partnership because we don't really know the contours of it, but they announced something that I feel very confident is a partnership as we described it. They just they announced that last month. What do we know about it so far? You wrote a great blog post this week called ChatGPT ads and the Amazon DSP is bigger than an ads deal. I'm going to have a link to that in the show notes. That was the sort of catalyst for getting you on the podcast. But what do we know about this ads partnership so far, and why is it so important?
So, we actually know absolutely nothing about this ad partnership so far. The only thing that we know is that it's announced and that it exists. But I think the timing of the partnership is the most interesting indicator of it. The timing of this partnership was within 10 days of ChatGPT announcing that it was on a billion annualized run rate for

advertising revenue. So, if we if we roll back the past 18 months or so with Amazon's behavior with chatbots, there's a clear pattern. The pattern is they initially block chatbots and AI agents. They do so for a simple reason. They might say because of the customer experience, they might say because there's disparities in the value that's created, but really the primary reason why they block this is protecting downside risk of their advertising business. So, Amazon's advertising business for rough numbers, $80 billion business growing 20% year-on-year, of which let's call it 60-ish percent of that, let's say 55 billion are on-site sponsored ads, which are paper-click ads that occur on search and product detail page, native to the purchasing experience of Amazon. And [snorts] Amazon monetizes intent in a wildly profitable way through on-site sponsored ads, in which people are searching for products and that they know they want or that they know that

they might want, and they have to go through a myriad of clicks and a myriad of views on search and product pages, in which people are bidding on keywords and product SKUs prior to making that purchase. And if you look at rough numbers of the total profits of Amazon's retail business outside of AWS, estimates suggest that 80 plus percent of all profits come from Amazon's ad business. So much of the enterprise value that Amazon's created over the past 3 to 4 years is directly tied to the profitability of its retail plus ads business. And if you forego a significant amount of that ads business because people are no longer clicking or viewing products on search or product detail page, then there's a tremendous amount of enterprise value lost for Amazon. So, Amazon needs to protect that, like period, full stop. And the primary way and the fastest way that they protect that is by blocking AI agents. Now,

despite the fact that they've been blocking AI agents, Andy Jassy at literally every single earnings call over the past 18 months has explicitly said that he, Amazon, will be looking into partnerships with personal AI agents, like third-party personal AI agents. However, there needs to be a quote value exchange associated with that. And so, when you think about what a value exchange looks like, the value exchange is actually really simple. The value exchange is Amazon needs to know that if they play nice and if they partner with third-party AI agents, that there will be a commensurate offset from the loss of clicks and on amazon.com to incremental ad revenue that they would garner from the AI agents themselves. And the reality is that prior to ChatGPT in this right now in this moment of

time, no personal AI agent operated at a level of advertising scale that could even come close to the advertising revenue that Amazon could potentially lose. And so, of course, now is the time that Amazon is announcing a partnership with ChatGPT because advertising revenue share and advertising revenue capabilities is exactly the value exchange that Amazon is looking for in partnering with someone like ChatGPT. So, that's where we stand today, but again, we don't know anything else other than the fact that there's a partnership that was announced and there's more to come in 2027.
Great, that's a helpful overview. I would I would say we don't know anything about that particular partnership, but we do have examples of partnerships that Amazon has pursued, right? So, the one that I've been drawing attention to is the partnership that led me to coin the phrase walled compounds.

And to sort of put forth this walled compound theory, which is that if you have a nexus of identity, then you can partner with other platforms in seamless ways that allow you to broadcast that identity onto their platforms without incurring any sort of like privacy Such that you monetize that identity while also improving the economics of the partner platform. And the example I used in that case was this Amazon meta partnership, where they were essentially just doing demand injection of their own advertising client base. And so what that was was that you would link the accounts, Amazon would kind of select the ad to insert into Amazon essentially get a bid request. They would select the ad to insert into the auction, and then meta would essentially pick, you know, on rank the ads using just the kind of standard approach. And then the resulting ad, if it

Amazon's ad was selected, would open up like a a a sort of custom modal that would allow the consumer to to check out in something that looked like the Amazon purchase flow. And so retailers are buying these ads, but it's flowing through Amazon was was constructing them. Amazon was deciding which ad to submit. And so it's that's very win-win. But Amazon has pursued other partnerships that that weren't quite as cumbersome in terms of the integration with other companies. So they Pinterest, for instance, they insert ads. They're not like buying an RTB, they're submitting it and they're getting sort of like a bid request and submitting it. And so we know that those exist. And so I think we might assume that this looks like those. And if you if you do connect the accounts and Amazon still maintains that the sort of like incremental benefit of knowledge of the purchase as anchored to the underlying user, and what OpenAI gets is just more demand, more ads injected, more more

demand for the ad space to drive the prices up. Right? So it is very win-win in that way. And so if that is the case, I mean, that on its face just seems to be vastly superior to doing some sort of like affiliate style take rate on the agent kind of picking the item that it wants to display to you on your behalf.
Yeah, we can we can throw out that business model. That business model is just it's dead.
We'll just throw it away. We don't need to discuss it any further. Yeah.
It's thrown away. We'll throw it away the same way that we don't even need to juxtapose the optimal business model relative to that because ChatGPT's already thrown it away. I think the interesting thing is let's paint a picture of let's fast forward 6 months from now. Let's put a blueprint for what an ideal partnership looks like from a ad rev share perspective and a customer experience and what do you think both parties would agree to? So here's what I believe to be true. Obviously hundreds of millions of people are prompting ChatGPT every single day about

e-commerce transactions and it's in ChatGPT's best interest to ingest Amazon signals like ratings, reviews, inventory, price, anything else about the brand that Amazon uniquely has into the native ChatGPT experience. That's obviously a good thing that would enrich the customer experience on ChatGPT. Additionally, when people are making those sorts of prompts, there should be a proliferation of ads for customers that are interested in purchasing on Amazon and one click they will can purchase an ad via the Amazon DSP that will not allow them to purchase on ChatGPT. I don't believe that Amazon would agree to that, but I believe that Amazon would agree to a one-click unique direct link that sends you back to Amazon. And so what's achieved in all of that? Who wins? So obviously ChatGPT wins because now the e-commerce experience in ChatGPT is phenomenally better than what it is this right now

because they capture in incredibly rich e-commerce signals. They're now capturing 50% of US digital e-commerce share with Amazon and they're getting incremental ad revenue from the demand that Amazon's bringing in. You know, what ChatGPT's done over the past year since it's launched its ad business is trying to bring as as many demand sources as possible because it it knows it has a ton of supply. And so how do you get demand? Well, you get demand when you have a bunch of people that are ready to purchase, that are ready to click, and you have closed-loop measurement on the fact that click led to a sale on Amazon. And so, that's going to be the key is that Amazon and ChatGPT are going to have to share the click that occurs in ChatGPT in a closed-loop manner right back to a purchase on Amazon. And if they're able to do all of that, then everyone's winning. Amazon is likely winning because they're able to charge on a cost-per-click or a CPM basis on the ads in ChatGPT. ChatGPT is

going to generate revenue from that. Additionally, if it's done through the Amazon DSP, it's not just at the impression level, but the Amazon DSP charges a platform fee against every dollar that's spent in it. Additionally, the Amazon DSP charges a CPM for using Amazon's first-party audience data. So, on ChatGPT, you wouldn't be buying a click on a search or a click on a ASIN or product SKU, Amazon's version of ASINs, but rather you'd be purchasing the audience using Amazon's marked-up CPM on the audience. So, invariably, these advertisers that are going to be looking for incremental sales are going to begin to advertise on ChatGPT if all of this proves to be true. It's not unreasonable to think that 10% of a endemic advertiser, someone that sells on Amazon's budget, should be going to offsite ads that occur in ChatGPT redirecting people to Amazon. So, that's the framework of an ideal partnership that I believe to be true.

And Eric, try and poke holes in that. Like, what doesn't work in that partnership?
No, I'm I'm absolutely 100% aligned with you. There's total overlap here. And and obviously, the consumer wins, too.
For sure.
Cuz they get better recommendations, driven by their own behavioral history.
Of course. And the categories that Amazon wins in are the categories that people are going to want this information. Like, if I'm Alexander Wang and I'm shopping for $100 vintage t-shirts in Korea, I'm not looking at Amazon ratings, reviews, or price and fulfillment timelines on that. But if I'm searching for a kids bubble bath that I want very quickly for my daughter's bubble bath, this is something I last purchased on Amazon. I I want ratings, reviews, price. Am I going to get it on Saturday and can I one-click get send it to me on Amazon? And the difference Do you think that again, sorry to to rant about the customer experience piece cuz it's not like an economically rational part of the argument, but Amazon has done such

an incredible job of minimizing the friction of purchasing. Like the delta between having an agent purchase that bubble bath on my behalf versus one-click into Amazon, one-click checkout is like so minimal that it's like why would I not just want to click the button on Amazon? They've done such an a phenomenal job of minimizing all of the friction on purchasing to maximize conversion rates that the experience of just the agent purchasing on my behalf is not demonstrably superior in any way. And so that's what I believe to be true.
Yeah, I mean I agree. I And I think it's also worth juxtaposing ChatGPT here with like Netflix, right? So Netflix has a demand deal with Amazon. They've got a demand deal with The Trade Desk, right? But it's a very different They're building a very different advertising kind of advertising business, right? So, you know, Netflix said that they projected their ads business. I mean this is months-ish old, so almost a year old. End of last year, they said that they

projected their ads business to be like 9 billion by 2030. OpenAI is already less than a year in on a billion-dollar annualized sort of trajectory. But OpenAI is building very much an ads platform in the image of Meta's. I mean there's a lot of ex-Meta people. It's an ad platform to serve the long tail of e-com advertisers and direct response advertisers. It's not for brands, right? And so all this stuff matters. The quality of the data matters. The targeting matters. All this stuff is really important. It's not just contextual targeting. We need to know a lot about this person in order to tailor an ad to them. And if that's the ad platform that you're building, then Amazon has the best data to deliver that singular ad. And if you don't have that data, you're going to be performing at a targeting deficit relative to that platform.
Yeah, definitely. And and not only is the richness of Amazon's audience data very impactful for the future efficacy of these ads, but what you described on the Meta-Amazon partnership, so that that partnership never really took off. It was announced, but you know, I've I've

lived in Amazon ad tech world for like 10 years. I kind of know when something takes off and then when something doesn't. That one didn't really take off.
Right.
But the reason why I think that this will take off is because the ad units that Amazon offers can very easily be extendable to ChatGPT as a native ChatGPT experience. So, let's just go through a few. So, every single person that's every single endemic advertiser that's advertising on the Amazon DSP uses what's called Rec Ads, which are responsive e-commerce creative ads, which are retail aware ads that just have the the snippet of the product detail page with the ratings, reviews, price, and a little headline. They've also recently announced that you can buy Rec Ads in a carousel. So, if you just picture the ChatGPT experience, it's the the bottom of that prompt is going to be littered with Rec Ads for products that are unique to that experience that you're shopping for.

This is literally an always-on strategy for almost every single endemic Amazon brand. There's a carved-out percentage of their Amazon DSP budget dedicated exclusively to Rec Ads, and they're super easy to produce because it's just the product detail page of Amazon. There's not really any like unique creative, and you can pump them out really, really quickly, and they're really, really effective as well. The next two are relatively new ad units which Amazon has created, which is one sponsored prompts. So, injecting yourself into prompt itself, obviously that could exist within ChatGPT with a one-click checkout to Amazon. And one of the unique data points that Jassy said on the last earnings call is that sponsored prompts convert a sale 48% more often, and users spend 21% more on average than those who don't click a sponsored prompt. So, when we're talking about offsetting advertising revenue, so like a loss of clicks versus a net new opportunity, one of the things that you'd look for when trying to offset advertising

revenue is understanding the effectiveness of these ads. And so, if the conversion rates are going up 48%, then the price competitiveness of the ad will soon follow. So, an advertiser who may only be willing to spend X number of dollars on a sponsored product ad, if a sponsored prompt converts at 48 more percent, then they're willing to spend 48% more for that ad unit. And so, if you're if you're picturing the very sophisticated Amazon economics team who are game theorying this out in a war room in Seattle, this is the type of stuff that they're looking at. So, sponsored prompt, super effective early ad unit at Amazon. And the other thing that Amazon recently created is what's called branded conversations. They just announced this at Unboxed a couple weeks ago, and branded conversations is in which you can actually pay to kick off a conversation with a branded chatbot, and that could very easily be extended to the chat GPT experience in which you pay

to kick off that conversation, and then you can curate the rest of that conversation and redirect them to something that comes back to Amazon. So, I'm thinking of a direct example that's not specific to someone that is traditionally thought of of an endemic brand, but as an example, like I recently got a new car lease, and I actually used a variety of personal AI agents to help me with the new car lease. And Amazon is invested very heavily into their autos product, which is effectively digital showrooms that occur at that provide a unique retail experience on Amazon. That's this really rich touch point along the path of purchase if you're browsing a new car. Well, branded conversations for auto OEMs that could live within chat GPT that redirect to an Amazon auto page or brand showroom is exactly the type of customer experience that Amazon is uniquely able to provide. And the super cool thing that Amazon also has with all of this is they have their advanced

measurement suite, Amazon marketing cloud. And so, if you can start to bring signals of this path of purchase to Amazon marketing cloud, you'll probably start to see that, oh, if I saw a sponsored prompt on chat GPT, those customers might increase my new to brand rate on this ad unit relative to the native ad units on Amazon because there's much less commercial intent on chat GPT relative to someone on Amazon who might be ostensibly further down the funnel. And so, I think that there is just a really, really complimentary set of data point signals and ad units that allow for these two properties, chat GPT and Amazon, to natively interact together to create a really powerful advertising and retail partnership.
And OpenAI launched their own sponsored agents product last month, right? So, that's the same thing. You You kick You kick the conversation out into a setting that is branded by that company. And

that solves for, you know, there's this the kind of whole one of the sort of I don't want to say criticisms, but but vulnerabilities or weaknesses of the ads and chatbot models that comes up a lot is that like, well, okay, but then that'll deteriorate the trust. The user needs to be able to trust that they're getting like objectively good advice from the chatbot, and if they feel like, oh, it's being paid for, then they won't trust anymore. Well, no, because if you kick it out into the sponsored environment, you know this is not from the chatbot, this is from this advertiser essentially.
You know those channels your colleagues keep bragging [music] about? The ones getting all the credit? Yeah, they might be doing squat. Attribution makes every channel look like a hero, [music] even when it's a zero. Incremental tells you who's actually doing the work. It's like a lie detector for your marketing budget. Start using Incremental today. Get your demo at incremental.com. That's i n c
[music]
r m n t a l dot com. Mention that you came through the Mobile

Dev Memo podcast for [music] a special 15% discount for the first 6 months.
Talk to me about the DSP because this I feel like is the least understood aspect of Amazon's advertising business. You talked about 65% is like the sponsored search. I think people anchor to that, but 65% still leaves 35%.
Sure, and that's a really, really rough guess, Eric. So, I'm not on Amazon, but all of their like if you think about the math of Amazon's advertising business, sponsored ads have effectively been relatively capped out on revenue for the past few years because the CPCs a flat line. The auction has gotten too competitive that people are not willing to bid anymore, and there's no new ad inventory that Amazon can create. Actually, they just created sponsored reviews, which I think is going to be a billion-dollar ad unit, but outside of sponsored reviews, there's very limited new ad inventory that Amazon can create because almost everything's already an ad on Amazon. So, the Amazon DSP instead uses Amazon's first-party shopping data and a variety of first-party

partnerships that Amazon's created to serve ads that occur off Amazon. So, these ads are traditionally display ads, CTV ads on Amazon-owned and operated properties like Prime Video, but you can now buy CTV ads everywhere in premium CTV outside of YouTube, and audio ads. And you can see now the very natural extension of that. They've actually renamed the DSP. It's a subsidiary of Amazon's ad agent, and the DSP is now known as DVA plus, display video audio and plus. Well, guess what's going to be plus? I think they're going to add a uh an A another A or a C for agents and chatbots. Because what effectively Amazon is doing with chat GPT is saying, "Here's the playbook. You want our signals, you want Amazon shopping signals in your personal AI agent, and you want to control commerce. The price to play is creating a scaled advertising

platform." So, you know who's going to be really good at that? Is Meta. Like Met Meta's going to have ads in news. There's not a question of that. Zuck might say take great potential, but ul- ultimately Meta's going to have ads in news. Meta is uniquely suited at building incredible advertising businesses. The ads in news are an inevitability, and they'll quickly get to a billion in ad revenue like faster than we know it. But, it's going to be really challenging for a personal AI agent like Instinct to to to do this. And so, I think that you'll start to see that these personal AI agents are going to start to veer towards more vertical specific use cases. Like Instinct 40-plus percent of their GMB is travel. Great. They're going to continue to go in on travel, but they're not going to be able to own e-commerce because if by default to own e-commerce, you need to partner with Amazon, and to partner with Amazon, Amazon is saying, "Hey, you need to build a billion-dollar plus ad business, or else we're just going to block you." And they don't really care. Like they don't they're not here to pick winners. They're here to protect revenue

and drive incremental revenue. In doing so, they are selectively choosing who they partner with because partnering protects their advertising revenue.
Catch everyone up to speed about those identity partnerships cuz we we we talked about so it was CAN last year, right? Not the most recent CAN, but the one before that. So, two CANs ago, when you were on the podcast, they Amazon announced the Roku, the Disney Plus These are identity partnerships. They're specifically ident- Talk to me about that.
Yeah. So, Amazon has this incredible sense of identity through the Prime subscription. And in addition to the Prime subscription, they also have other signals that enrich this unified sense of deduplicated identity across American consumers like all of the Amazon hardware products that they have as well. But, they are always in the interest of strengthening that signal. And I think the the primary partnership that did that was their Roku partnership, cuz Roku has the most ubiquitous OEM for connected TV and smart TV

devices. And so, everyone saw that Roku partnership and thought that this was an inventory deal, but actually it's an identity deal in which Roku is now leveraging their OEM hardware to enrich identities the identity spine of Amazon, such that Amazon has coverage on 90% of US households now for identity. So, the Amazon DSP, in considering the like unparalleled and unfair strength of it relative to any other DSP, now has deduplicated reach frequency identity across 90% of US households. So, if you're a brand that's objectively if you're just starting your DVA plus agent budget somewhere, why would you not start it on Amazon? Because you have the sense of identity that no other enterprise DSP has for buying similar sorts of inventory.
And and identity also enriched with probably the most valuable data set that's ever existed.

Right.
Right. I mean, okay. So, then you touched on this, but maybe we could just define it in like more concrete terms. What is What is the optimal relationship look like across agents, retailers, and retail platforms? Like, let's assume everyone behaves rationally. Let's assume everyone optimizes for, you know, maximizing their own outcomes. And let's assume that incentives and market power matter, which is an argument that I I keep sort of like pounding the table on, because I think it gets lost in a lot of the um agentic commerce optimism. Like, what what does this What does this settle into? What does it look like?
So, I think that all retailers should not be bucketed into the Amazon story, because I think the Amazon story is actually very different, because they uniquely have something like the Amazon DSP as a demand source for purchasing these ads. Whereas Whereas all other retailers don't. All other retailers will rely on other third-party DSPs like the Trade Desk or DB360 to leverage their audience data for buying inventory on other places. So I

think that Amazon is uniquely capable of monetizing the intent and offsetting the potential ad revenue loss on purchasing behavior on amazon.com. So So just to sum it up, the ideal partnership for someone like Amazon is a an advertising revenue share with someone like ChatGPT in which Amazon is making a percentage of all of the ad revenue that comes from ChatGPT, the platform fee on their DSP, the audience fee offered through their DSP, and in return, they're giving back ChatGPT all of their rich retail signals to make a purchase. Ratings, reviews, price, inventory, everything that you would normally see with an Amazon on ChatGPT now has that as context that it can share to a prospective customer with a one-click deep link that can send you back to Amazon to make a purchase natively back in Amazon. That is the ideal use case. Does that look the same for Walmart? Maybe, but again, no one

else has a demand source like the Amazon No other retailer has demand source like the Amazon DSP. Walmart has actually had a different perspective in which they have a Walmart DSP that was powered by the Trade Desk, but given in the existential doubt of the Trade Desk very existence, Walmart has instead said, "Our audience base and first-party data should live in any DSP." So now you're starting to see Walmart partner with folks like Yahoo, like DB360, and with Magnite directly. And so the the challenge there is like what does a ad revenue share look like with Walmart? How would Walmart be able to execute that? Would they rely on their own DSP, which is powered by the Trade Desk? Would they have to rely on others? It just gets a lot more complicated when you don't own the whole ad tech stack. And Amazon is the only retailer that owns the whole ad tech stack.
And then what do you think about? So, I mean Shopify occupies like a different it plays a different role here. And then occupies a different part of the

landscape because it doesn't do fulfillment at all. Right? So, it has I think the luxury of just saying sure, we'll work with that. Like but you could say yes to everything. It doesn't matter if the if the transactions are flowing through your pipes, you you make money and it doesn't really matter what happens after you hand that off. Right? Like so there's a So, Google just launched the universal cart yesterday. And what's really interesting about that is okay, it uses UCP to aggregate the catalog information from the retailers, but also to manage the handoff. So, UCP is is obviously the agentic protocol or you know, the universal commerce protocol. But the reality is Google's trying to aggregate just the discovery in AI overviews AI mode so that they can show ads. They've got an ad product for this discovery mechanism in AI overviews. That's direct deals. And so, what they're willing to do is to say look, if you want, we'll hand it off and the checkout happens on your website and we'll just give you all the information. Or if you if you don't care, we'll check you out here. It's just that gives everyone the opportunity to optimize for what they care about. If I care about the customer and and also

all the checkout with the universal cart happens where the retailer remains the merchant of record, they have the customer relationship, which again is really important. And so, there are like just different sensitivities here and I think, you know, every time Shopify does an agreement or Stripe does an agreement with an agent, I think it's blown out of proportion with respect to how indicative it is of agentic commerce as a new consumer paradigm. Okay, I want to give you one more question before we I know we you have to jump. Talk to me about B frontier and B eclipse. These were just launched or announced yesterday. These are the hardware devices and for Amazon collecting more consumer data. How does that fit into the broader platform and advertising strategy?
It's interesting. So, that those products are effectively listening products that exist as a bracelet or exist as a plug in that you attach to your phone and it'll listen to all of the conversations and everything that you do in a private and secure way and feed into Alexa Plus to create a more personalized Alexa Plus experience. I think what it tells us is that Amazon is

not taking their investments in Alexa Plus lightly. Amazon will continue to invest in that. And the thing that every agent needs, whether it be a vertical AI agent like my company or whether it be a consumer personal AI agent, is context. And so the more context you can enrich the agent, the better the customer experience is going to be. And so I think that this is a small announcement that it probably won't take off. They're pretty invasive products that I don't know how many people really need every single aspect of their life transcribed. But for those that do, this is an option. But I think the most compelling piece is that we haven't heard the end of Amazon's ambitions with Alexa Plus. So Alexa Plus is was what used to be called Rufus. It's the AI agent inside the amazon.com experience. And so naturally you could say Alexa Plus is

going to listen to your conversations and just decide to purchase that toilet paper for you on Amazon. I think that Amazon is smart enough to know that those use cases are just so limited by now. But I think that Amazon is using this as a perhaps wedge to strengthen the future story of Alexa Plus. And it's in Amazon's best interest to do that. Amazon is a behemoth. They're not willing to seed all customer interactions to third-party AI agents. They need them to operate on their terms while Amazon simultaneously makes its own bets. Same way that Amazon builds its own chips. They have their own chips. They use Nvidia chips. Like Amazon makes a lot of bets and do not want to be beholden to anyone else for its very existence. It's critical bets that they need to make like agents and like technology infrastructure like chips. And this is a really early signal that Amazon is going to continue to invest in Alexa Plus.
Adam, as always, it has been a pleasure. Thank you for elucidating us and

illuminating this topic for us. How can people learn more about Giga?
giga.ai. I'm on X, I'm on Twitter, or I'm on X and I'm on LinkedIn. And we have some fun announcements coming in 2027. So, excited to hear from you all.
Thanks again and enjoy your weekend.
Thanks, Eric.