B (3:17)

So cool, so cool. I gotta check this out.

A (3:20)

It is cool. Like San Francisco. I mean, it's not my town, but like, it definitely feels like I'm living a couple years in the future, taking Waymos and stuff like that.

B (3:27)

That's great.

C (3:28)

Awesome.

A (3:29)

All right, so our guest today is Rajiv gol from the CEO of PubMatic, who I guess both you and I have known for decades. At this point, we wanted him on not to talk about the boring old SSP exchange RTB stuff, but to talk about the new sexy stuff, because Pubmatic has been on the, I'd say the cutting edge on using this agent AI for transactions. They were, they probably can say they did the first real transaction using Agentic AI. They put out a press release in December and they've been working a lot on this stuff. So it's going to be a really interesting conversation about where all that stuff is going. And then we are just about, I don't know, five, six weeks away from architecture live.

C (4:18)

So.

A (4:18)

So what's going on with the Startup Showcase? So we announced it last week. How's it going?

B (4:24)

It's great. We've gotten a ton of submissions. And tomorrow, or today Rather today, Friday, January 30th, is the deadline to submit for the Startup Showcase. I have been asked a couple of times, why should I pitch? I'm not raising money right now, so I'd like to clarify something. This is not a pitch fest to raise money. There's. There's going to be investors in the audience, there's investors on stage, but this is not for that. This is to get the word out about your startup and do a demo in front of decision makers, buyers at agencies, buyers at brands, platforms that can be partners. This is to do a showcase for the industry and for decision makers. And this is for everybody. So I encourage everyone who's on the sidelines submit for the Startup Showcase. Put your hat in a Ring. It's going to be awesome and you will have zero regrets. Just filling out a Google form. Thank you.

A (5:18)

Well, all right, we got that PSA out of the way. Next psa. So we want to reward our loyal listeners to come to the market extra live event. And so we have a promo code. So if you register with the code ARIPOD30, that's a R, I, P O D30, you'll get 30% off, which is a pretty deep discount. So I'd encourage you to do that. Aripod30. And the website is marketecturelive.com Last thing, we have a lot of announcements. So we're a week away from the Super Bowl. I don't think anyone's super excited to see the Patriots there, but, you know, you gotta just suck it up and watch. But we're there for the ads and there's two different ways you can use marketexture to follow the ads. So first is my favorite project, MAD db. So we're using AI to classify all the news articles about the Super Bowl. So that's from Ad Age and Ad Adweek and Ad Exchanger and Podcast and everything like that. So if you go to maddb, aifeatured, or just go to the homepage, there's a link right there. And for the super bowl, you'll see every article we've been able to find that is about the super bowl in all of the trade pubs in one place. And it's updated every hour or so. So that's a fun way to get your news. And secondly, adland tv, which is, we believe, the world's largest repository of super bowl ads. It's been collecting ads since 1970. We are also collecting all the ads this year. So we are writing a newsletter about the best ads. We are collecting them, YouTube links, original videos, et cetera. So it's a great place to see all of the ads in this year's Super Bowl. So I recommend you check out both those things. All right, so that was a lot of promo. Let's go to the meat. So we got Rajiv Gol from Pubmatic coming on. We will hear about ad CP and how it's developing, as well as a lot of news this week, including meta earnings, AI, some layoffs, some other stuff. So stay tuned. Welcome. Rajeev, how are you?

C (7:21)

I'm doing great. How are you doing, Ari?

A (7:23)

I'm great. I want to take this opportunity to go on my long walk of shame and to apologize formally for misspelling your name in my book. My book Yield was Very well, fact checked. But I misspelled not one, not two, but three people's names. Joe Zawaski, which is, you know, that's a, that's a, that's a pass because that's a really hard day.

C (7:43)

Joe misspells his own name. So to be fair.

A (7:46)

But yours was totally egregious. I apologize. And the third will be a trivia question we'll leave for detailed readers.

C (7:54)

No problem at all.

A (7:56)

I was thrilled to be in the.

C (7:57)

Book, Ari, so I got. No, the bar is low with me.

A (8:03)

So thanks for joining us. Everyone knows you, everyone knows pmatic, but I really wanted to bring you on because you've been really kind of pushing the envelope on some of this agentic and CP stuff and give our audience a chance to find out what's really happening. Is this just fluff? Is it just press releases? Or is this really the future? So what's your point of view generally?

C (8:27)

Yeah, I mean, I can say unequivocally, we view it as really being the future. And there's a couple of reasons for that, I think when we think about programmatic advertising, you know, we've, we've been at this now, some of us coming on two decades, right? RTB for, I don't know, 18 years, 17 years, something like that. When you think about the media planning and buying process, you know, from end to end, RTB is, you know, it's like each atomic ad impression that is being optimized. Right. There's a protocol of how you describe the impression, how people bid on it, but there's all sorts of things that are happening upstream and downstream.

A (9:01)

Right.

C (9:01)

Like discovery and planning are having. Happening upstream. You have that moment of activation with RTB and then you have measurement and optimization. So I think as an industry we've been really focused on that point of activation. But there's actually a lot more to the entire process of buying and selling media that we haven't yet fully automated. And there's a ton of upside as a result. And I think if we get it right and we'll get into this, I think we can significantly expand the total addressable market for digital advertising and help a lot more brands generate more ROI and actually help consumers continue to get more relevant ads.

A (9:42)

Yeah, I mean, from the buy side perspective, you know, media planning and pricing and negotiations and, and all that sort of stuff, that all happens before you log into a DSP and buy anything programmatically and AI has a chance to kind of make that more of a holistic process perhaps. Right. Is that is that kind of what you're saying, right?

C (10:02)

So if you think about like the typical process, right, you know, somebody's launching a new car or a movie or whatever, an advertiser, so they have their agency, they got a planner. And that planner is going to maybe send out an RFP, you know, to 10, 20 CTV publishers or, you know, whatever is the case, they respond back, hey, these are my audiences. These are the placements, these are the price points. It goes into a spreadsheet. All of that happens. And that's gotten better, of course, with things like email or even APIs and standards around how we measure things and describe things. But it's still a very manual process. And so there's a lot of limitations that are introduced by virtue of the fact that it's manual. A human brain can only track process so many cells in a spreadsheet. So that means you can only have so many publishers, so many placements, so many audiences. You have to know who you're reaching out to. So direct sales, okay, do I know the seller? This is where the famous dinners happened to get to know people, or jeans parties, all that kind of stuff. And then when you fast forward a couple of steps, you get into trading. It can take a trader 30 minutes, 45 minutes, an hour to set up a complex campaign. And then maybe a trader can manage several dozen campaigns. You know, why, why can't we let them manage, you know, 2, 3, 4, 500 campaigns? And so I think AI could really change this, right, where we use AI agents to elicit input from publishers. What are your audience segments, you know, what are your price points, what are the placements, you know, what's all the performance data around those. The AI agent for the planner can make a bunch of decisions as to which campaign or, sorry, which, you know, which pieces they want on the, on the campaign. And then, you know, AI can optimize, AI can even set up the, the campaign, you know, in, in the trading platform. So I think there's a lot of scope for improving, you know, the processes that we have. And I think as we do that, what we'll see is that we can create a lot more efficiency in the advertising supply chain. And we'll get into this too. But I think this is much more than a technological shift. It's going to be a, a value chain, massive, you know, shift. And then I think we can make advertising a lot more effective and maybe and probably actually compete ultimately with walled gardens on performance. And I think that is really the big prize.

A (12:35)

Yeah, I mean, that's the crux of my issue around this whole thing, which is, you know, we could all agree that the advertising ecosystem could be automated, it could be more efficient, all that stuff. And you could make that argument within the context of the more traditional programmatic world where you have a DSP and an SSP and they're trading on a per impression basis. But a lot of what we're talking about is actually changing that and moving to a non programmatic impression basis, it seems. So maybe take me through the case study that you release. So arguably the first real campaign to be run using adcp. There were some demo demo ads run, but Butler Till ran with, with PubMatic. I love the fact the brand involved, it's called Club Tales, a flavored malt beverage brand. Eric, you drink a lot of Club Tales around the, around the pool?

B (13:29)

No, but it sounds fantastic.

C (13:31)

I mean, that's why they need to advertise.

A (13:32)

Right?

C (13:33)

This is a review.

A (13:33)

Are you a big fan of Club Tales? Have you tried?

C (13:35)

Yeah, I'm, I'm getting to know it better. So.

A (13:39)

Okay, so anyway, so we had to, we're trying to sell some Club Tales and. And instead of buying programmatically or direct, they instead decided to use a AI agent to book campaigns with PubMatic. Walk us through what this means, what happened.

C (13:55)

Yeah, so. And in fact Programmatic does actually play a role in this. So what, what we, what we did with, with Butlerto and Club Tales is they used Claude as a front end and they described their, that first of all, they, they connected to our MCP server, right, Our model Context Protocol server. And we've launched a product called Agentic os, which is what we were using in December, but we didn't actually announce the product until CES in January. And what Agentic OS does is it is a MCP enablement of, I would say all the standard use cases for digital advertising, which is creating a campaign, discovering audiences, pricing all of the normal use cases that you would go through between a DSP and an ssp. And so what the trader at Butler Till did is they connected their cloud agent to our MCP server. The MCP server described to them all of the services that are available. They said, okay, what inventory and placements do you recommend? Here's the URL of the brand so that our agent could go read about and find out about what is that brand, who might be the relevant target audience. Obviously we want adults, we don't want kids, et cetera. And then we recommend, recommended a bunch of placements, you know, from the 2000 plus publishers that we have. We also had over 200 data partners in our Connect Data platform. So we recommended third party audience.

A (15:29)

Wait, we recommend who's we in that? Your agent.

C (15:32)

We as our agent. Exactly. We is the primatic agent in a gentic os. Yeah, so it's a useful clarification. Right, so we, the agent responded back and then the trader again using Claude could say okay, great, like I like.

A (15:51)

The human being reads, read what you sent them.

C (15:53)

Exactly, read what we read the response back and then said yeah, great, I want more like this, I want less like that. And then when they were happy with it they said great, I want to purchase. And in our case they used Activate, which is our direct buying interface directly in the ssp. So they did not need to use a third party platform. So they could buy directly within the PMATIC SSP using Activate and again they typed into Claude, great, buy this and let's set it up. And then they get periodic reporting updates and they could do optimization and they can do that either in our UI non agentically via human or they get updates back to their cloud interface. And I would say that part we're still figuring out right in the sense of this is nascent and so what kind of data do you want, how periodically do you want it? All of those types of things. But now you can see how Agentix starts to take shape and how we can collapse the amount of effort that's required. So a couple of data points, 87% shorter time to launch campaigns, 70% faster troubleshooting. You know, so the, the, the manual upside is huge. Now we're in the process of post campaign measuring what the results were to try to see, hey, is there like, you know, better outcomes and all that kind of stuff. So I don't have that, that data to share yet.

A (17:20)

And so what did you train your agent on to be able to give an intelligent response to this buy side query about what would work for this particular brand on your particular inventory?

C (17:33)

Yeah, so what we trained it on is a couple of things. So one is we have extensive APIs in our platform already. Right. So building agents actually, you know, we can, we write an agent to write code to build the agents, right. By reading the, the APIs and then we trained on historical campaign data. So one of the, I think key advantages that we have at PubMatic is, you know, we're processing well over a trillion ad impressions on a daily basis on our platform. You know, there's tens of thousands of advertisers bidding on a monthly basis. So we have huge amounts of data to be able to be able to train. We have of course those hundreds of data partners, we have more and more commerce partners. So there's a whole architecture, and I've talked about this in our earnings calls, which we call the architecture of advertising intelligence, which is infrastructure, its application and its transaction, which really describes how we have embedded AI across our entire stack. It's not like a copilot kind of add on, like what you get from Microsoft these days in Excel or Word. So we've really thought through. We have a partnership with Nvidia, so we're using specialized Nvidia hardware which allows us to have real time data pipelines where then AI models can get data in real time through programmatic transactions. So there's a whole framework for how we've approached AI across our entire stack to be able to get to this point.

A (19:02)

So one of the themes that I've been writing about that seems sort of validated by this use case or this case study is that the sell side has a stronger role here than previously. Let's talk about the buy side first. So in the DSP paradigm, the DSP has a lot of intelligence, it's a brain, it knows data, it knows pacing, it knows all this other stuff. And AI kind of has an opportunity to sort of skip the DSP where you could just tell the AI what you actually want, that I want this kind of person, I want this kind of frequency and all that sort of stuff and hand it over to the sell side. Is that a fair prognostication, I guess, about the future of dsps?

C (19:45)

Yeah, I think that's true. And I think further what we're seeing is that there's a high degree of UI lock in or incumbency that a handful of DSPS have today, right? If you go talk to Big Holdco, they say, well I got to train my team on this handful of DSPs, right? I don't have time to learn something different. And so we get that and that makes sense. And now from what I just described, which is this trader using Claude and anybody can type in planing English or any language for that matter. You know, you entirely do away with the UI that they're used to. And the next step, right, which is a focus for Q1 is like don't even type in Claude, just use an agent, right, that you've, you know, pre programmed with some, some basic logic where now the agent can take on this campaign planning effort and buying effort. So I think what we're Seeing is that AI, my prediction, AI will upend and collapse the industry's value chain and bring advertisers and publishers much closer together. I think we will move away from, you know, the, I think the role of the DSP and the SSP is evolving where, you know, instead of needing multiple paths in that value chain, you might need only one path in that value chain. And I really view our opportunity as being the leading AI powered ad tech company that's delivering digital advertising performance. And I think that value chain compression and shift can significantly lower the cost of advertising and improve the effectiveness of it. Right. When you look at walled gardens, one of the reasons why they're so effective at performance is they have a single layer of tech that connects the buyer's ad campaign with the media that they own. Right. In the case of YouTube or Meta or TikTok. And I see AI is enabling that kind of trading paradigm in the open Internet.

A (21:48)

Yeah. So let's talk about the sell side then. So you're PubMatic is effectively an intermediary, an enabler of publisher use cases. And then I think going to the news that we're going to talk about later that pre bid is going to implement adcp as part of the prebid stack, which effectively allows almost any publisher to have their own MCP server that lets people buy ads. So will publishers who historically have had a problem of scale and fragmentation, will they be able to take advantage of AI selling or will it require someone with scale and intermediary like yourself to do that work for them?

C (22:32)

I think the short answer is yes, as in both. Right. So I think we'll see publishers launch seller agents as they should, but those agents, you know, they'll connect into buyers that know about them. Right. So if you're a big brand, sorry, if you're a big publisher, big brand on the publishing side, then of course you know, a buyer is going to know and they'll configure to, you know, buy it against ESPN for instance, or you know, Disney or NBC or whatever the case is. But I think the real opportunity is if you're one level below those folks and you don't have that sales team out there now you can actually get a lot of leverage and figure out how to connect your sales agent into an agency marketplace where now you might be able to get 1, 2, 3% of a big advertiser's buy that you weren't able to get to before because you don't have a sales team in Chicago and Detroit and New York.

A (23:27)

Yeah, historically There have been quite a bit away efforts to enable small advertisers to buy direct on publishers and they've never been very, they've never really scaled that well.

C (23:38)

Right, yeah. And I'm even, I'm talking here about maybe the flip problem which is small publishers to be able to work with, you know, large advertisers.

A (23:45)

Yeah, yeah.

C (23:45)

Then of course you're right, that you know, we're seeing now I think this huge influx of small advertisers, medium advertisers that you know, haven't been in the ecosystem before or have only been in the walled gardens. And now there's TV Scientific and Vibe and all Mountain. These companies are great companies going after that space. But a big part of that is this, I think AI enablement that is going to open up the playing field significantly and I think make it a lot simpler for them and through this collapsing of the value chain also create much greater advertising effectiveness. So I do think you're going to see. Back to your question, Ari. Publishers with their own sales agents. But I think as always, you know, a platform like ours which gives a buyer instant access to 2000 plus premium publishers, whether or not each of those publishers has built and launched their own sales agent, I think there's going to be a lot of value in that. Right. And that's what we're seeing today. You know, when we launched Agentic os, our launch partners were folks like WPP and MIQ and Butler till a license. Right. And so these guys are saying, hey, there's AI enabled scale already at PubMatic and that makes it super easy for me to go and transact.

A (25:02)

Right. Like you can imagine a small medium sized publisher like marketexture. We have some ads available for sale. We have GAM as our ad server because everyone does. And now we could have an MCP server so that if someone wanted to buy a banner ad on one of our slots, it could happen, you know, somewhat automatedly in a way that would be cost prohibitive for us to hire a salesperson and to do an IO and to deal with all of that.

C (25:25)

Yeah, that's exactly right.

B (25:27)

How would discovery work in this use case? Like short term, I think long term. I think I understand the vision that you're painting Rajiv, which is when the agents are buying and selling from agents, they're going to know about everything, right. The universe and you won't need a, you know, a door knocking salesperson. But like today with, with you know, the, the roadmaps, how, how would market texture if we continue to use this as the.

C (25:54)

As the example.

B (25:55)

Get discovered by some brand that might be a great fit and has just never heard of it before.

C (26:00)

Yeah. So I think this is where, you know, again, the. If you have your own seller agent. Right. You'll need to figure out how to get discovered. And I think we'll probably. It'll happen in a couple of ways. One is you have a sales team that's out there pushing it. I think we'll probably see marketplaces emerge where agents go to discover inventory. Right. So there'll probably be a place, Eric, where like you can go register market inventory and say, oh, great, architecture has a sales agent. Here's how you find it. Here's a little bit of metadata about us. Like come in and buy from us.

A (26:36)

Market might build that.

C (26:38)

Yeah, or you could build that. Exactly. Somebody should build that. And then the third would be through.

A (26:44)

An app for this call, right?

C (26:46)

Yeah, yeah, perfect. Why wait till after? You can vibe code as we speak. And then third is through us. Right. So we will be offering that service. We are already offering that service for all of the publishers that work with us because we have these deep integrations, often code on the page, certainly server to server. We know all the placements, we know all the pricing, we have all of the audience data, we have third party audience data, we have the infrastructure. So we're in a position to be able to say to Club Tales, who maybe would be a small advertiser that a big publisher sales team wouldn't spend time on, hey, great, you could buy from these premium publishers and that's incremental revenue.

A (27:28)

Yeah, it's a really interesting kind of dynamic how everything's changing at the same time. So you mentioned earlier that you think some of this technology could help the open web compete with the walled gardens. What's the state of the open web? There's so much negativity right now around traffic declines and fragmentation and just general malaise. What do you think?

C (27:50)

Yeah, so I'm actually quite optimistic about it because of the AI opportunity that we've been discussing. The current state is not amazing, that's for sure, given the traffic challenges that publishers are facing. Right. And you hear from publishers that are seeing anywhere from 10 to 50% kind of decrease in search traffic now. I think when you go look at. When we look at our business, we built it to be resilient in a couple of ways. Mobile, app and ctv. These are huge chunks of our business. And so those are largely immune to AI search shifts. We have built our business around Premium publishers. And so when we look and see there's data out there, the typical premium publisher, like a ESPN or NBC or whatnot, they have maybe like 15% of their traffic is coming through search. Right. People bookmark them or they go directly. Right. Just because they're such well known brands. And so when we take, okay, hey, let's say AI wiped out all that search, referral traffic, 15% on a subset of our business, you know, we get to like, for us it's a, maybe a mid single digit impact if the worst case were to happen. So I think we're going to see, you know, certain types of publishers. I think the struggle is going to be real. I think the counter is every publisher is working towards how do they get more data to use for measurement and outcomes. So logged in registration for instance. Right. One of the reasons why I think CTB is doing so well, mobile app commerce, these are all environments where people are largely logged in. And that persistence of identity means that targeting, frequency, capping, all of those things can be much better, measurement can be much better. And so that's where the performance is happening. And so what gives me confidence though is that the big differentiator on the open Internet is great quality content. Right. You can compare that content anytime to you know, what's inside the walled gardens. And I think open Internet has much better content. Open web is becoming the logged in web. And then as we talked about earlier, I think AI is going to collapse the value chain and create efficiency and effectiveness where when the cost of advertising comes down, cost of you know, conducting ads comes down and the effectiveness goes up, you know, through using AI to, to, to drive planning and measurement and optimization that I think will be a step function improvement in roi and that will bring more dollars into the open Internet. Right.

A (30:24)

And there's a funny thing going on and there's been some stats on this, that vibe coding is creating a lot more apps and websites. For the first time in a while, the number of apps submitted to the Apple App Store, I don't have a reference to this had been going down for many years, new apps and it's suddenly made a U turn. It's going up quite a bit because I mean you, you, they don't have a lot of scale or whatever it is, but it's fun to see like new websites, new apps being developed which you know, who knows little take off and what consumer behavior will be like.

C (30:57)

Yeah, absolutely. And I think probably the cost basis of those apps is way lower than traditionally made app. Right. And so it will be economically viable with fewer advertising dollars than would have otherwise been the case. And so I think that'll. That'll create, you know, a new level of vibrancy and innovation.

A (31:19)

So last week, we had Paul Connect in marketing CMO on the show, and he said that he was at ces, and I don't remember what event he was at, but they did show of hands. How many of you have heard of abcp, and it is words nobody, like.

B (31:39)

Clarify that I was in the room. It was an agency. It was an agency event. And, yeah, it was full of brands. The agency asked, you know, or the founder asked, before we get into this stuff, how many people have heard adcp? And what I said was, there were two hands that were raised with me and Joe Zawadsky.

A (31:56)

Yeah, sorry. I mangled the whole thing. I live in my own world. Should we care about that? Is it too early? Or should. Maybe brands shouldn't care about this at all.

C (32:08)

I think brands shouldn't care. Right. It doesn't matter. ADCP is the how. It's the plumbing. You know, no publisher ever asked me if I use a Windows PC or a Mac. Right. It's. It's kind of irrelevant. Right. I think what matters is that those brands, whether they know it or not, are starting to use AD cp. Right. We launched with wpp, W promote miq, Eliza, and Butler Till. And we have more and more campaigns launching every week that are, you know, Gentex campaigns that are launching. And those agencies represent real brands and real dollars. So I think, A, we're early, and B, maybe they'll never know. And maybe that's perfectly fine. Right. And my guess would be that's a bunch of CMOs in the room. I've rarely met a CMO of a Fortune 2000 brand that wants to talk about RTB. Right. Just like, you know, I don't want to talk about what kind of oil goes into my car. It's kind of irrelevant. But I like to have, you know, nice, fast car. It's great. So I don't know that will ever get to a point where, like, that is the right metric.

A (33:12)

So the last subject, I think the last time you and I saw each other in person was in the Virginia courthouse when you were testifying. Right. I was sitting in the pews and, like, clear.

C (33:21)

Neither you or I were on trial.

A (33:24)

No, no. Yeah. We're both innocent. But every. Every witness would go on the stand, and then they come down and they would fist bump me on the way out. And the other reporters were like, how do you know all these people? So I know since then you filed a lawsuit against Google, private lawsuit. So I guess we can't talk about that. But I will ask you, with the remedies supposedly coming out, you know, within the next four to eight weeks, any predictions?

C (33:52)

I won't make any predictions, but I mean, I will say this. Yeah, so we do expect, you know, remedies to be announced in the not too distant future. You know, there's been a big question among investors. Is it going to be structural remedies or behavioral remedies? Right. And so, you know, personally I'm expecting that there's at least behavioral remedies. I think, you know, you've made the case that structural seems like a stretch, and that may very well be the case. In any case, I think the behavioral remedies will be much quicker to be enacted, you know, just by the, the very nature of the, of the remedy. And our view is that when we do the math, each percentage point of market share gain that we could achieve as a result of these remedies being put in place, it's worth about $75 million in revenue and most of that dropping to our bottom line, we're about a 4% market share player. Today. We estimate Google to be about a 50% market share player. So it's a very significant potential catalyst. And so we're eagerly awaiting, and I think we've started to see a little bit of preview where Google has made some concessions. And so I think we're starting to get a little bit of an early window into what it might look like.

A (35:10)

Yeah, there's two forces here. There's one which is them making their auction fairer, which would allow you to compete with your existing demand more fairly for GAM publishers, which is almost certainly going to happen. They've already done a lot of it in Europe. And then the second thing is if they force AdWords to start bidding through your inventory more aggressively, which is just potentially a windfall of demand coming your way.

C (35:36)

Yeah, absolutely. And I think both of those would also make the playing field level for advertisers and publishers, where you would expect to see fee efficiency, you would expect to see higher ROI for buyers, higher CPMs for publishers. So I think these are very important remedies that are in the balance and hopefully are being announced soon.

A (36:00)

All right, this good place to take a break. We'll come back in a minute with a lot of news, a lot of earnings, meta, of course, AI news and some layoffs. So there's A lot to discuss. We'll be right back.

C (36:15)

This episode is brought to you by Credit Karma.

A (36:18)

When it comes to your money, Credit Karma keeps you ahead of the game. You can count on Credit Karma to keep up with your financial needs needs as they evolve, helping you monitor your progress and giving you personalized recommendations so you can make strides towards your goals. Make sure you're on the right track.

C (36:35)

No matter where you are on your financial journey.

A (36:37)

Intuit Credit Karma, Karma you can count on download today.

B (36:42)

All right, back everybody with the refresh and we have a bunch of things to go over today. Let's start with Meta earnings. So yesterday was the kickoff of earnings season and one of the companies that we closely tracked, Walled Garden, kind of a bellwether, Meadow was the first to announce and they had a bang up quarter. So they did. Last year for the calendar year of 25, 200 $201 billion in overall revenue with ad revenue just over 196 billion. So this is a monster business and a 22% year over year increase. So not only was Q4 good for Meta, 2025 was good. Two things, you know, kind of stood out on the, on the, on the, on the transcript. So the first was their CapEx. So everybody knows the tale of, you know, AI. All these, you know, mega scalers are spending a lot of money. Meta is planning on spending between 1:15 and $135 billion this year on AI capex versus 75 million in 25 versus Google at 93 billion. So he is foot on the gas. And then on the other thing, I think, you know, you have to hand it over to them. Meta's doing a fantastic job of talking about how AI is driving these business results in terms of better ad targeting, higher engagement across their apps. You know, CPMs going up. Like I think they've really nailed how to frame this such that investors are, because the stock is up almost 10% this morning, the day after, like supportive of all of these expenditures that they're making. So this maybe start with Rajeev. Two questions. Number one, his overall reaction to this. And then number two, interestingly again, you talked about your partnership with Nvidia and to the extent that you could talk about this as a public company, how you're thinking about these CapEx numbers, both that we're seeing at the high end and you as a public company CEO.

C (38:44)

I feel like the person at the kids table at Christmas dinner. We're measuring our capex budget in the tens of millions, not in the hundreds of billions.

B (38:55)

Million 135.

C (38:57)

So when I hear these numbers, I'm always reminded of the movie Brewster's Millions.

A (39:03)

Right?

C (39:03)

I don't know if you guys saw that, but it's like somebody like held a gun.

B (39:06)

How do you spend it in one day, right?

C (39:08)

Yeah. Like, okay, you gotta spend a hundred billion dollars this year on capex. Like I'd be like, man, I believe in, I give up, right? Like, I don't even want to try to spend money that quick.

A (39:18)

I'll give you some suggestions in Iceland. Yeah, to you got to build data.

C (39:22)

Centers, you got to shovel 300 and 100 million out a day, right? Whatever the. That's a couple 300 million out a day. So I think actually what we're seeing is something very interesting, which is that the fastest realization of AI to AI investment to revenue is actually the advertising industry, right? That's to me that's the takeaway from Meta earnings, you know, Google, last quarter, maybe last two quarters, is that advertising is actually like the built in machine that's best positioned to take advantage of all of the benefits of AI. And I think when you compare it to other industries, right, I don't know, transportation, oil and gas, cpg, so on and so forth, the flywheel is basically already there in advertising. Whereas other industries have to figure it out and set it up. Like Tesla announced yesterday too, and they're moving to more robots and all these kinds of things, but this is still like multiple years away. Whereas you're seeing the revenue and the return in advertising right now. And in part I think it's because of all the things that we've been talking about, how Agentic can sit on top of RTB and Programmatic. It's not an either or, it's both together. And then the fact that we have these auctions that happen in real time and advertisers are looking at the ROI on a hourly, daily, weekly, monthly basis and then putting in more money where it works. So that is in part what makes me really bullish about where we are focused.

B (40:54)

Producer Mike, please clip the past 90 seconds or two minutes. That was gold. Ari, what you got?

A (41:02)

I'm following kind of a sub story here, which is whether Meta is getting serious again about fan Facebook audience network. There's a couple of things going on here. So Eric Seufer, always good to quote him, he had a newsletter this week about whether Meta is getting serious about in app advertising. Advertising again and makes the case that they sort of dropped off after ATT on in app. Right. Because in so much in App was like user based and it's fan based. And Applovin came out of nowhere and built this huge business that's a huge threat, honestly, to Meta and its advertising probably. Meta probably considers Applovin among their one or two most strategic competitors at this point. Connecting some dots. You know, we had Jim Payne from CloudX on the show and he, a big part of his pitch was like, meta, meta, meta. They're helping him get in with publishers, app publishers. And that seems like a very clear situation where they're using their leverage to try to make some chinks in the armor of Applovin's SDK presence. And then there was a Twitter thread today. I could put it in the notes about app publishers saying Meta's coming out of nowhere. And now they're becoming a pretty big source of monetization after Applovin and after Unity, like number three in Apple app. So I think there's a little strategic change here on metaspart from de emphasizing fan and in app to starting to make it a real priority.

C (42:31)

Yeah, I think, you know, after ATT rollout, right, I mean, Meta was kind of hit pretty hard and took them what, about 18 months, two years, something like that, to basically kind of recover the business, which I think they did through all of their AI investment, right, to be able to figure out, you know, create signal from where their. The signal went away. So I think to the extent they feel like, hey, our AI capabilities can work effectively on third party inventory. Right. As opposed to our own inventory, where of course they have more signal, then it would be logical for them to say, okay, great, like even beyond the financials, it's, I think, strategic in nature, which is, hey, let's get more of that advertiser's budget. And so then we can, you know, optimize the steer to meta owned properties versus third parties. So that could be the driver behind that resurgence.

B (43:26)

Makes a ton of sense. All right, let's keep talking about AI. So this morning, shortly before the POD dropped, there was an announcement that Pre Bid is taking over ADCP's code for creating sell side AI agents. And Bach dropped a newsletter on this as well. I think they're kind of one and the same. Ari, what's your breakdown of this whole announcement?

A (43:48)

I think this relationship is similar to the way PreBID implements RTB. So OpenRTB is a standard from the IAB and then PreBID implements it. And this is similar in that the ADCP protocol has its own organization that was founded recently that has Randall Rothenberg and Matt Eagle as leaders. Matt Eagle will be on our podcast, actually, I think, I don't know if next week or two weeks from now he'll be our guest. So we'll hear about that. So they're creating the protocol and then pre bid is implementing it. So you could bait. They said within I think 30 to 45 days it'll be available and publishers will be able to effectively have a plugin to pre bid that will give them the capability to spin up an ad CP server. Very exciting. It's great.

C (44:33)

Yeah. I mean, I think we're seeing a race to establish the standards, like a healthy kind of race. And I think having reference implementations is going to help open up access to this capability for a lot of folks. So I think that's going to be a net positive.

B (44:53)

Yeah. Box newsletter was interesting. I'll pull a quote out and it's very much aligned with what you said earlier, Rajeev. And I'll start the quote. Prebid sales agent is open source at the Boys in an afternoon and it could double the sales of your premium ad products without adding a single headcount. So effectively what you were saying, which is, you know, just expose your inventory to a wider universe of potential buyers, once this thing is at scale, who can then, you know, kind of discover all the great things about your products.

C (45:24)

Yeah, and this is exactly the point I was making about efficiency and effectiveness. Right. Where now the cost to transact will come down because you don't need an expensive salesperson. Right. The seller agent will interact with the buyer agent and then I would expect effectiveness to go up too, because whereas before a human might be able to, you know, put a buy on 20 publishers, now you can go to 250 or 2000 publishers if you want to, and the machine can figure out, you know, what's, what's optimal and what's, what's hitting the KPIs versus what's not.

A (45:56)

So, Raviv, are you in, Are you in Bach close now?

C (45:59)

Are you tight?

A (46:00)

Because I'm old enough to remember when you posted a blog post of a crying baby referring to one of his blog posts, Is that right?

C (46:08)

I don't even remember that. We're definitely talking a lot.

B (46:14)

So, you know you are founding ADCP member company, right?

C (46:18)

Yeah, yeah, yeah, exactly. Yeah. So that's been clearly an area of mutual interest. And you know, Brian of course is a force of nature. And so I think a lot of what we've seen in terms of the energy and the focus and the movement behind this agentic protocol And ADCP more broadly is due to Brian. And so we definitely appreciate that and are staying engaged with him.

A (46:47)

He and I have also buried the hatchet. So we're all buddies now.

C (46:51)

Excellent. I like it. We're all wiser in our older age.

A (46:55)

Exactly.

B (46:56)

Some of us. Somebody's got to go on archive is to find said blog post and maybe resurface it. Okay, let's talk about. We have to talk about OpenAI again. So last week everybody was fired up about OpenAI just announcing an ad product and a million dollar price tag and no targeting, so on and so forth. Today we've got open a couple more details. So effectively they're targeting a $60 CPM for these first campaigns without advanced targeting, without really kind of advanced measurement tools. And people are continuing to be really mad about this and I don't get it. I don't get it. I don't believe why people are so mad.

C (47:40)

It's a little bit mystifying to me to like announce a cpm right? Like that to me is a little bit odd because we all know that advertising doesn't really work that way where you just say, hey, but we can.

B (47:55)

Also do math, right? Like, what's the CPM of search? What's the CPM of B2B search? It could be in the hundreds or thousands of dollars.

C (48:04)

So that's the mystifying part, is like, why wave a flag that says 60 bucks where people are going to fling arrows at it. In reality, you want to be charging $1,000, not because you're charging $1,000, but because the performance is there.

A (48:19)

Right?

C (48:19)

And the advertisers are like, yeah, great. Like the math makes sense that a thousand bucks, like I'm getting great roi. So that, that's the part that I think for me at least is, is mystifying. I'm not angry about it by any means, but I just, like, I don't, I don't understand why they took that approach. And also because they have so many smart ad people. So that's the other part that, you know, leads to the, the, the surprise.

A (48:43)

Yeah, I don't know what people expected. They're going to come out out of nowhere with a fully functioning, you know, AdWords clone. You know, that takes some time and I don't even know how they're going to do billing. Like, this is complicated, right? You know, making sure quality ads and that they run properly get reporting all, you know, bots are screened out. There's like a lot of work. Advertising is not so easy as A lot of people think so.

C (49:09)

You know, that's a really great point in that maybe what they're, maybe the reason they're taking this approach is it's the cold star problem, right? Which is like, how do I get some campaigns running through so I can start to build the optimization engine and all that kind of good stuff. And so maybe the way to do it is okay, we just pick a price point. Well, Netflix chose $60, seemed to be, seemed to work out well for them. So we'll go with that and let's get 1000 campaigns running and then we can start to deliver on the performance side of things.

A (49:43)

I think that's right. Like this, you know, as soon as you start running ads, there's going to be a million fires to put out and it's going to be all the unknown unknowns they can't predict in advance that they're going to have to deal with. So just like start running something, make a little money, make some headlines and build the relationships with the advertisers and then iterate.

B (50:01)

But I guess Sam Altman doesn't deserve any of our sympathy. So I understand why people will just react this way.

C (50:07)

Let me ask you guys a question on that one. So where do you think whose budgets is OpenAI going to eat from? Is it search? Is it social, is it open Internet? What budget do you think they're going to eat from?

A (50:21)

Well, I'm assuming that over time it's going to be largely contextually targeted based on what you're searching for or querying about. And this could mostly be performance, in which case it's definitely going to be search.

C (50:32)

Search.

B (50:33)

Yeah, okay, yeah, I agree. Search, maybe search plus social media. But how much of this is the. Also the kind of like venn between ads and commerce. Right. Because like ultimately it's going to be like showing you an ad and then you know, encouraging or like literally directing the transaction of said ad. So it's like, you know, like ultimate funnel collapsing.

C (50:57)

So yeah, that could be Amazon and Walmart it could be competing with.

B (51:01)

Yeah, exactly. Exactly to that point. There was a note that opened is taking 4% on storefront purchases. So basically when a user completes that transaction, it seemed like a low incremental amount. But again, shout out to Eric Suefer, he thought it was like really bad and like could, you know, kind of turn some merchants who have very thin margins ultimately that negative on transactions.

A (51:24)

I thought this was interesting in a contrast to Google's approach. So Google announced its UCP protocol a couple weeks ago. It's been feels like it's been an era since then. And my understanding is that in the Google world, you're able to put your storefront in their shopping experience and they're not charging you a fee for that, but with a giant caveat that in order to get to that product, Google has all kinds of advertising opportunities that they'll charge for that. You know, Google Shopping has a paid system. So I think it's a little bit apples and oranges here in the long term. I think Eric's right that taking a percentage on just showing the storefront is a tax that maybe is a bridge too far. But driving the demand, top or middle funnel is an obvious place where you are willing to pay. I think this is one of the most interesting things going on right now is how are the Shopify, Walmarts and other merchant tools trying to work their way into the AI world?

B (52:29)

Here's one that I'd actually love to get your take on for a nuanced reason. So Yahoo announced Scout, which is Yahoo's search, Yahoo's answer engine. And they were pretty explicit about how they're going to be building in ads in day one. So Rajiv, you were a leader in the display era. You're now a leader in the CTV and mobile app era. How much time are you thinking about like answer ads as the next format? And what do you think on the other side of this, like formats in answers look like?

C (53:07)

So we're actually spending a good amount of time trying to figure this out and we definitely do not have answers to any of those questions yet. But I think there's a growing opportunity for, we'll call it AI chat advertising or AI chat as a, a consumer kind of engagement model where then you can put ads in. And so some examples of that. Right. So Scout is, you know, let's say the, the most recent one. But if you go to many e commerce sites like Home Depot and you're like looking for a new fridge, let's say you need a new fridge, you know, there's an AI agent that can answer all sorts of questions for you, right? Like I need it to be this size. I want it to have bottom freezer, not top freezer. It this kind of energy efficiency and it'll Rufus Amazon.

B (53:55)

Rufus has ads, right?

C (53:56)

Yeah, you can go to health websites now and you know, you can do chat and those, you know, travel is turning into that as well. So kind of vertical by vertical you can see, you know, these chat experiences coming up and they're a very different environment, you know, in which to run advertising. It's not going to be like just slap a banner ad in there. There's all sorts of things to look at in terms of, of, you know, what's the context that you could pull from. How far back do you go? You got to deal with user privacy, of course. How do you share that signal with an advertiser to figure out, like, which campaigns are going to be most relevant? And then your question, Eric, which is, you know, what's the right ad format in that environment? Right. Do you put a banner in? Do you put a video in? Is it a standard IB video, or do we need something else? You could have a background ad that appears so. So I think it's an entirely new kind of format or medium for advertising, and I think none of those questions are answered yet. And so we're definitely spending time on trying to figure that out.

B (55:00)

Yeah, same. All right, let's get into some not great news and then we'll come up with something to lift us at the end. So, two layoff announcements this past week. This stinks. So first was Pinterest announced that they were cutting 15% of staff to focus on AI and they're going to be actually closing some offices. And then the second one was Amazon is laying off another 16,000 corporate workers. This is after 14,000 were laid off in October. And again, sounds like for the second time, a big company loved doing this. Like, there was email leaks and it was pretty sloppy. So these are real numbers. I mean, Amazon, on an absolute basis, big numbers, but, you know, on a percentage basis of the company, relatively small, they've got 1.5 million employees worldwide, but this ain't good. And it seems to be, you know, part of what we're. What we've been expecting with relation to AI.

C (55:57)

Yeah, I mean, there's always this question of, like, how much of this is AI related versus not.

B (56:02)

Oh, yeah, over hiring.

C (56:03)

And.

B (56:03)

Yeah, the zero barrier, for sure.

C (56:04)

Yeah. And of course, like every, I don't know, cfo, CEO, they have an incentive when they're making cutbacks to say, oh, it's because we had this great unlock with AI. Right. Because, you know, you'll try to get some credit with investors by, you know, by, by making that claim. And so it's all. It's always hard to know. I mean, the Amazon one, I mean, my question is, like, how much over hiring did they do back in the day?

A (56:28)

Right.

C (56:28)

Like, it seems this is like, I don't know, is it like the fourth time that they are kind of whacking the corporate site.

A (56:34)

It seems like Amazon is part of their strategy. Just keep hiring, hiring, hiring, then call, then hire, hire, call. Right. So I'm not sure that it's like they wake up one day and are overstaffed. I think it's like part of their ongoing approach.

C (56:47)

Yeah, just. Yeah, it's like performance management maybe, right?

A (56:50)

Yeah.

C (56:51)

And, and it feels like a big number, but it's like they have so many employees and I don't know how many that one and a half is like, you know, working in factories or transportation, logistics, all that kind of stuff versus the corporate. But, but it, you know, there's such a big company that a 5% end of year reduction for performance could be like 10,000 people. Right. It's at that scale. I think in the case of Pinterest, what we've seen is that the walled gardens that are not measured in the billions of users, it's been tough. Right. And so nextdoor, Pinterest, X Roku. Right. I mean they're all different businesses, but when you're running a walled garden from an advertising perspective where you say, hey, the only way to buy me is like through my own tools, you're competing with Meta and YouTube and TikTok which are measured in the billion plus, maybe multiple billions. And so it's I think really hard to make that smaller walled garden kind of approach work. On the flip side, if you take any of those, you know, small to medium sized walled gardens and put them in the context of the open Internet, they're huge. Like a big publisher in the open Internet is like 30 million. I don't know what Pinterest has, I don't remember off the top of my head, but I think it's in a couple hundred million. So they're massive. So I think we're going to see more and more of these kind of mid sized walled gardens move to being more open. Roku did that and I think that's worked very well. Where they sunsetted Data Zoo acquisition. One view. Yeah, one view. Yeah, exactly right. And totally opened up. We see Roblox going with an open approach. We see that X has opened up. So I think we're going to see more, more movement in that direction.

B (58:43)

Makes perfect sense. Ari, want to hit one more and then call it or just call it.

A (58:47)

Listen to our friends at Data Links.

B (58:49)

Absolutely. Shout out to Joe Lux from Data links. Raising a $4.2 million seed round. Appearium was happy to invest in this round, as did you.

A (59:00)

Right. Personally, Joe was the Second commercial hire at Beeswax, our first salesperson, and he was a big part of our growth and he's a great guy and did some time at Amazon after Beeswax and then with a bunch of other Amazon product engineering folks, had this sort of insight that marketers needed a sort of a data prep before using their data for AI. So he's building an infra business that helps with that. Maybe I'm not describing it the best way, but I'm really excited for him and I invested.

B (59:39)

Yeah, you're right. So the insight that he had was, you know, he said anecdotally, maybe 80% of enterprise marketers, I think there's some data out there that supports it. You know, somewhere between 70, 80%. Their data is a mess and unable to be activated for AI use. Cases like personalization, all the stuff that we, you know, we kind of love to think about. So that's what they're doing. They're building a data readiness platform. So it makes a lot of sense and it could end up being something really, really important in this, in this age of Regentix. So happy about it. Congrats and shout out to Joe.

C (60:13)

Yes, congrats to. Congrats to Joe.

A (60:15)

All right, let's call it. All right. So, Rajiv, thank you so much for being here. This is an amazing conversation. I really appreciate your insights about all this open this AI stuff and I'll probably see you on this weekend at IAB Leadership.

C (60:30)

Yes, I will definitely be there. And thank you, Eric and Ari, always so much fun to chat. I always learn a lot myself, so appreciate it. And yeah, I think this is going to be the breakout year for agentic in our industry and I'm super thrilled about where we're positioned and the opportunity to innovate.

B (60:47)

Love it. See you next week, everybody.

C (60:50)

Thank you. Thank you for subscribing to marketecture. New interviews are added every week at Market TV and your favorite podcasting app.