The Sell-Side Strikes Back: Joe Root on AI, the Outcomes Era, and Rebuilding the Ad Stack from the Publisher Up
- May 19
- 51 min read
For the better part of two decades, the buy-side controlled the game.
Data. Decisioning. Optimization. Margin. Publishers? Commoditized. Intermediated. Squeezed.
But that era may be ending.
In this episode, Joe Root (Co-Founder & CEO, Permutive) returns to Signal & Noise with a sharper—and far more disruptive—thesis: AI is shifting the center of gravity of advertising back to the sell-side.
We unpack what happens when:
Decisioning moves closer to the data
Signal-rich environments outperform identity graphs
And publishers stop selling impressions… and start selling outcomes
Because if the most valuable data lives on the sell-side—and AI can act on it in real time—then the entire AdTech stack gets rewritten.
The Death of the “Buy-Side-First” Internet Joe breaks down why the traditional model—where DSPs optimize against thin, degraded signals—is fundamentally broken. By the time an impression reaches the bidstream, most of the signal is already gone. The result? Poor targeting, wasted spend, and a race to the bottom.
The Rise of Sell-Side Intelligence Permutive’s approach flips the model: decisioning happens at the edge, inside publisher environments, where the richest behavioral and contextual data actually exists. This isn’t just better targeting—it’s a different architecture.
From Curation to AI-Driven Outcomes What started as curation and probabilistic targeting is evolving into something bigger: → Real-time prediction → Continuous optimization → Outcome-based execution
We explore how AI turns fragmented signals into scalable performance—and why this unlocks a new commercial model for publishers.
The “Outcomes Era” Explained Selling impressions is easy. Selling outcomes is hard. Joe explains what actually has to change—technically and commercially—for publishers to move from CPMs to measurable business results.
Agency Business Model Reset As AI erodes the billable-hours model, agencies are being forced into a new role: → Investment managers → Principal traders → Outcome owners
We dig into how this shift is reshaping incentives, margins, and how media gets bought.
Agentic Trading & the Future of the Market If both buyers and sellers deploy AI agents, what happens next? Do auctions disappear—or get demoted? Does allocation move upstream—before an impression is ever served? And who wins when media becomes negotiated instead of auctioned?
This isn’t a conversation about incremental optimization.
It’s about who controls the advertising system in an AI-driven world.
Because if Joe is right:
DSP-centric decisioning gets abstracted
SSPs and publishers gain leverage
And the open web—long written off—may have its strongest comeback yet
The future of advertising won’t be defined by who buys impressions fastest.
It will be defined by who controls signal, decisioning, and outcomes closest to the user.
And for the first time in a long time… that might be the sell-side.
Read the full transcript bellow:
Brett House (00:01.098)
Hey everybody, welcome back to Siglund Noise. I'm Brett House joined by my co-host Rio Longacre. Today we've got Joe Root. And this is someone who's been thinking about the future of sell side long before it became fashionable in the industry. If you don't know who Joe is, he's the co-founder and CEO of Permitive, right? A University of Oxford and Imperial College of London grad. That's pretty impressive. mean, everybody, anybody can get into Oxford.
Rio (00:26.542)
It's impressive.
Joe Root (00:27.605)
Bye.
Brett House (00:28.68)
And I was fascinated by your background because it went from like sales intern to CEO. was like, I want to hear about what happened in that transformation, in that sort of trajectory. And we'll talk about that in a minute, right? But you were kind of the perfect combination. English, you know, we were kind of, you know, I was a comparative literature major. RIA was involved in journalism. We both had that sort of English side of things and computing with a master's in computing. Seems like the perfect combination in today's world.
Joe Root (00:53.397)
There we go. What more could you want in life? Yeah.
Rio (00:54.7)
Love it.
Brett House (00:56.988)
I know, seriously, like you can talk and communicate and you can build stuff, right? Which is super, super awesome. And he lives in New York, right? So, Permutiv, Permutiv arguably is a platform built to help publishers take back control of their data monetization. Sometimes it's been called a publisher DMP. And obviously we're at a time in the industry when it seems like a lot of the industry is moving in the opposite direction.
Rio (01:02.54)
And he lives in New York, he's got it all.
Joe Root (01:06.933)
But I feel so good now.
Brett House (01:23.976)
right with the buy side sort of grabbing control, you know, the open web sort of under threat by the walled gardens. You know, we've talked, we've all talked about this a lot. You know, really what I think makes this conversation super interesting is the thesis about how this is evolving in the age of AI. And I know that you've got some, you know, you've been talking about outcome outcomes era, right, which we've heard a ton in the last, you know, two years and a shift away from selling impressions towards delivering measurable results. So in sort of
which is pretty interesting from a publisher perspective. Stop selling impressions it seems like you're leaning towards. So, would love to hear a little bit about you. Did I get your bio right? How did you go from a sales intern to a CEO? Tell us a little about yourself, Joe.
Joe Root (02:10.001)
Yeah, you got the bio very on point. yeah, me and my co-founder met in our first week at university or kind of college over here. So we were kind of going through a computer science degree. Both have become like really interested in the same area of research, around kind of distributed systems and machine learning. And also stumbled upon like this guy called Paul Graham.
was writing these essays about starting a company and exactly why Combinator. And at this point in time, we just had no interest in going to work. And at that point in time in London, the only tech you could do was go and work at an investment bank. we just were looking at what you could do in America. It's just so much more exciting starting your own company. So at this point in time, why Combinator is only a years old and
Rio (02:40.001)
Y Combinator.
Brett House (02:41.555)
Yeah.
Joe Root (03:04.277)
kind of being a computer science graduate, I was like, oh my God, I'm not going to be able to in any way start a company if I can't sell something. So I met the guys at GoCardus who were kind of YC company in London and they said, hey, why don't you come along and learn how to sell things for a while? And yeah, I was picking up phones and wasn't the best salesperson in the world, but it gave me a good appreciation of the art of it. yeah, a few months later we were at Y Combinator and that was the journey into permittivity.
Brett House (03:33.8)
Yeah. And that's one thing I've taken away from a lot of execs and founders that I've talked to in CEOs is that when you can combine sales experience, whether you're really good at it or not, and product kind of subject matter expertise and like deep knowledge, that combination is just incredibly powerful, especially when you're a CEO, right?
Joe Root (03:52.373)
Yeah, I mean.
Rio (03:52.642)
Yeah. Plus it, plus it's funny, Joe, like that. And like what you said too, about having to pick up the phone and just call people. mean, that is such an important skill to learn. No one's naturally good at it. When you first start, just takes trial and error, dozens, hundreds, maybe thousands of phone calls to do that. Right. I was actually my first job in a market in marketing. We had a lot of it was market research. So been there.
Joe Root (04:05.428)
Yes.
Joe Root (04:09.619)
Yeah, I'm, yeah, it's terrible, but I went in thinking our sales is this kind of dark car. And I'm like this pure thing of computer science, but like, really got appreciation for like great salespeople have this process and methodology and the best ones are just relentless at following.
Rio (04:27.47)
Yeah, they're worth their weight in gold, a good salesperson, that's for sure. And you'll put up with a lot for a good salesperson.
Joe Root (04:30.195)
With that out.
Brett House (04:32.712)
Yeah, yeah, one of my favorite recent books is The Unsold Mindset. I give you credit to Mark Sabatini and it's all about, for reading that one, it's one of those great, like if you have a four hour drive ahead of you, you're going up north or something, out of New York City that is, it's just a great like three or four hour ride. And it's just all about how you sell by not selling at all. You know, there's a little bit of this sort of seven secrets to highly effective living kind of stuff going on there. All these business books seem to have it, but it's about like, how do you actually sell authentically?
Joe Root (04:52.465)
interesting.
Brett House (05:02.674)
and forget all the fucking mechanisms and techniques and, you know, be authentic, be your authentic self and people will buy that.
Joe Root (05:02.741)
Hmm.
Rio (05:10.008)
I think that's even more important now, Brett, right? With all the AI sloppits out there. I posted about this morning. I ran a campaign for this company. I remember we sent out thousands of emails. We got a bunch of opens, not a single click of response. People are so just done with the deluge of content. They're overloaded. They don't want a white paper. They don't want to be qualified, quote unquote. I think the way that people sell and the way you have to market has really changed quite a bit. It's fascinating.
So Joe, we were really excited to have you on this. You were on my previous podcast, First Party Pod, which is defuncted. And then part of the reason why Brett and I decided to do this pod was we were looking for something to do and you were on that one. had a great discussion about where Permit was at the time. It's been about a year. So we're really stoked to get you back on this because all of the really interesting changes going on within AI and how that's impacting not only the ad tech, let's say, ad tech, mart tech stack itself, but how that's affecting the
balance of power. mentioned this before, right? I think historically balance, like, I think it's easy to argue, like the buy side had a lot more power over sell side and publishers really have not done well the last 20, 25 years. you know, I think the joke is like, no matter what happens, they always get hosed, right? So, AI is maybe changing that equation as decisioning was closer to data. And if a lot of the most valuable data is on sell side and maybe the center of gravity,
starts to shift and I'm starting to see some interesting things. So we're really keen to get your perspective. So you're a permitive. I really feel like you're probably in the center of this shift and really maybe accelerating a lot of it from centralized buy-side optimization to distributed sell-side intelligence that are truly closer to the user and closer to where things are happening. So the question we're exploring today, I think it's pretty disruptive is like what really happens as Brett mentioned earlier, when publishers move away from the race to the bottom and trying to sell impressions and start selling.
outcomes and if we're the age of outcomes, they should be anyway. So I'm looking forward to this discussion. We're excited to have you on and let's kick off the Q &A.
Joe Root (07:11.044)
Mm-hmm.
Brett House (07:15.454)
Yeah. So Joe, so what's changed? mean, you were on that first party pod with Rio. Seems like ancient history, right? A lot has changed since then. And you guys were talking about curation and probabilistic targeting, right? Yeah. Curation was the rage for a couple of years there. Now it's really on helping publishers package and sell their data more effectively. what has really changed since now and then? What have you seen?
Joe Root (07:15.615)
Perfect.
Rio (07:24.846)
Curation was all the rage,
Joe Root (07:37.237)
Yeah, so I think what's remained very true in that time has been this addressability crunch and the loss of signal is huge. So I think in the year between just more and more agency and advertiser folk are recognizing, hey, the data, the things I used to make decisions on aren't scaling and I'm experiencing that in my reach, I'm experiencing that in my outcomes.
And you can see kind of the market become much more literate around this. When we first went out two years ago, it wasn't really a thing. Like the fact that Rio now you kind of see this as such an essential part of the ecosystem, I think kind of shows the trend lines which are starting to play out and people just a lot more literate here. In parallel with that, I would say like the rate of disruption within kind of the agency business model has just been remarkable to watch from the outside.
We just had a really great agency exec who she joined our board recently. she was saying, like we were talking to the last friend, was like, yeah, the disruption model for agencies used to be like every seven years there would be a disruption. And the last couple years, was like, that's narrowed down to two years.
Brett House (08:32.744)
We talk about that a ton, yeah.
Rio (08:34.498)
Yeah.
Brett House (08:53.988)
I wouldn't be surprised if it's near done to six months at this point, if you think about it.
Joe Root (08:56.945)
Yeah, yeah.
Rio (08:56.952)
Well, feels like, yeah, the retail media thing just finished disrupting. Now suddenly AI, you're right, it's like wave after wave every couple of years now,
Brett House (09:04.392)
Yeah, now the FTE model is sort of being called into question, right? The buildable hours model, all that.
Joe Root (09:10.287)
Exactly. So that is such a tectonic shift that billable hours model to agencies because of AI that billable hours model no longer works with clients. And what we're seeing at the agency level is, hey, increasingly that means agencies now need to find ways to make money from the budgets they trade versus that billable hours model. clients see a lot of benefits to that and a lot of good alignment there.
But that now changes how media gets bought, how the agencies think about that. And when you pair that with kind of what's possible with AI, it's like, wow, we can rethink this whole ecosystem because there's such an urgent agency pressure to rethink these models.
Brett House (09:52.404)
Yep.
Rio (09:53.014)
So Joe, your recent conversations that a couple I really keyed onto was like one was with the AI Edge podcast with Miles and Shiv, great discussion. I love their podcast. Great guys. Yeah. In was one of the earlier guests in this pod until last year. One thing you touched on a little bit was with the shift towards AI as it's adopted and like really like looking at who controls the stack. When did it click that?
Brett House (10:03.655)
great guys as well.
Joe Root (10:05.045)
Very good.
Rio (10:22.178)
This is not just another gimmick. This is not just, you maybe not even another layer, but this is something fundamental in how it changes like the balance of power and the center of gravity. When did I start clicking?
Joe Root (10:30.825)
Yeah, I think there are two things which we've really noticed and observed as we've so the history of the business is we spent the first seven, eight years to exactly what you guys said. We went and built this data management platform on the supply side because we felt, hey, third party signals are disappearing. There's this abundance of signal on the supply side. And if you can build the right infrastructure and technology, you can make that available and decisions could be made on that instead of this kind of declining signal, which was available on the buy side.
Brett House (11:00.65)
So was that like a real focus on first party, enhancing first party signal, right? And capturing that in a much richer way for publishers?
Joe Root (11:05.823)
Totally. Exactly. like, really kind of publishers have this enormous volume of signal, but the pipes which sit in between the publisher and kind of the DSP on the other side, really don't want to take too much signal into them because it kills the cost margins of that business. So DSPs, SSPs, everyone's getting rid of inventory and bid impressions. They're getting rid of signal. So actually by the time an impression makes
Brett House (11:33.834)
And that's a compute-based cost, I'm assuming, from just the data passing. Yeah.
Joe Root (11:37.107)
Yeah, totally, right? It's a compute and bandwidth cost.
Rio (11:40.45)
Well, yeah, the amount of signal, they're only taking a fraction of what's available to actually, you know, for the bidding logic or any of that, right? I mean, a lot of it is just missed entirely.
Joe Root (11:47.605)
X.
Exactly. So by the time an impression makes its way over, often 70 % of them have been stripped away anyway, because there are these kind of limits, the DSP places on what can be sent to them. And then when you observe the impression, they'll have an ID, but most of those third party IDs are missing. And even things like the URL gets stripped of all the parameters. So you're just buying a domain. So now you're making decisions on like barely any signal. And on the other side,
Brett House (12:16.296)
Yeah.
Joe Root (12:17.641)
You have this abundance of signal every interaction you take, what you're looking at, what you're researching.
Brett House (12:21.672)
Which tells a little bit more about who you are and maybe what segments you belong in, Don't they say the average, isn't the average consumer in like 36 or 37 different segments and like a traditional segmentation schema? Yeah. Yeah.
Joe Root (12:26.271)
coating.
Rio (12:32.408)
Well, plus the contextual signals too, right? So there's like, there's both.
Joe Root (12:35.061)
So what we observe is this two to three X order of managing more data on the supply side, because exactly what not being leveraged at all. So you're talking about hundreds of thousand X more signal available on the supply side, which just all disappears as you go through kind of this supply chain today. And that data is really powerful if you can leverage it to make decisions. But today,
Brett House (12:43.486)
that's not being leveraged. Yeah.
Joe Root (13:01.417)
the way decision is done on the demand side kind of stops us from leveraging that signal. So we kind of recognize that.
Brett House (13:05.162)
Yeah. Well, no, and that has real world downstream impacts. like, just as a funny aside, when I was at Exolate, they had on their privacy page, you could actually type in kind of who you were from a profile perspective. I don't know if you threw in your email address or a cookie idea, probably your email address. And then it backed that hashed email into everything that it knew about you. And it was so far off. So it thought that I was a single female urban influencer living by myself in New York City.
I was a married with young child, aging white man, not an urban influencer. So was just, I always wondered like, cause you're selling that. Yeah. You're selling the power of one-to-one and programmatic and automation and, audience segmentation and specificity. And yet it seems so diverse from reality. And to your point, it's cause they're probably only capturing a small sliver either that, or they just associated me with the wrong ID. Yeah.
Joe Root (13:35.485)
Exactly.
Very different lifestyle.
Rio (13:44.984)
Why am I getting all these cosmetic sads, right?
Joe Root (13:46.569)
Yeah
Joe Root (13:57.62)
Yes.
Joe Root (14:00.981)
tiny slither and then you're you're desperately trying to recreate from that tiny slither what a very complex data sets on the other side. So what you end up doing is like people take an IP address and try and turn that into a deterministic ID, but these things like there's research out there, they're wrong 85 % of the time. So then you get categorized exactly.
Rio (14:19.98)
Yeah, it's not very precise. Yeah.
Brett House (14:20.412)
Yeah, yeah. Anybody in the family can be on that IP address watching a show or whatever.
Joe Root (14:26.441)
and they're getting recycled all the time and...
Rio (14:27.694)
Well, depending, yeah, you might be depending on what network on or machine you could show, show an IP in a different city entirely. Like I remember, like I think my work laptop sometimes shows up from Chicago. It's crazy, right? Because the way it's, the way it's routed. yeah, not precise at all.
Joe Root (14:38.812)
Exactly.
Brett House (14:39.154)
Yeah.
Joe Root (14:42.611)
Yeah. So yeah, that kind of loss of signal, I think kind of transpired Brett to your point, right? Like it has real meaningful business impact. And I suppose what we've seen and what's really started to shift our mind is like when we originally started, we saw the impact on media companies. So you're a publisher, 70 % of your inventory doesn't have a third party ID. So now the buy side, it's not even that they aren't bidding on it. They don't even see it. So you're getting paid nothing on this inventory. The other
Rio (15:11.086)
They'll prioritize IDs first, for sure, the way DSPs work, yep.
Joe Root (15:14.491)
Exactly, precisely that. But then the other side of it and what's been really interesting is, hey, as agencies go through this business transformation, which is like that FTE model is no longer how they make money, a key part to how they make money as they build out their operating system and the principal trading business is the data businesses they've built. And the impact this is having on their data businesses, which has been the really interesting piece, like those data businesses are losing scale because they're relying on IDs which have disappeared.
Brett House (15:31.922)
Yep. Yep.
Joe Root (15:43.293)
not because the data isn't amazing, but because kind of the current OpenRTB system isn't giving them what they need to be able to pen their data. And that's making it really hard for that transformation.
Rio (15:51.758)
How many, how many agencies, Joe, are actually like leveraging sell side signals? Right? I mean, it's because, buy side, you're right. They have massive, we talked about this in our last episode, you know, they, you know, I think a lot of the legacy data businesses, you know, these were database marketing companies. They've built to these massive databases, hundreds of millions. Some of them claim billions of users, right? You know, with all all of these different data points are capturing, but how many of them are leveraging like sell side signals, which I mean, I grew with you, like it's like,
Brett House (16:07.912)
Yep, the epsilons and the new stars, yep.
Rio (16:21.516)
not fully, not leveraged enough, certainly, but contains some of that rich information for targeting that they should be using.
Joe Root (16:28.573)
Yeah, so what we've really seen is these companies are leveraging sell side signals, but they're doing it in a tactical way, not necessary yet in a strategic way. And that's what's changing very quickly. So if you rewind a year ago, what we found is agencies are saying to their teams, hey, you need to go and fly X million dollars against our data sets with your client. And the client teams would be like, hey, actually, I just can't get this spend away.
because the audiences aren't scaling well enough. And when that happens, what a lot of those teams, the client and at the trading level realizes, hey, well actually there are a load of these curators who have access to Signal, which is more readily available. Let me go and transact with them. And that's kind of given birth to curation and that's been a really powerful kind of lever of growth for us and our customers.
Brett House (17:19.9)
And curation is aggregating various direct publisher relationship data signal into a package.
Joe Root (17:24.405)
Exactly. It gives a very neat way to start to kind of bring some of this signal together on the supply side into a package so you can now buy. Now, this look.
Brett House (17:35.598)
You can scale your buy and is it all premium inventory and you guys have direct relationships with like premium publishers?
Joe Root (17:42.353)
So that's where there's a whole spectrum of people approach this in a different way. Our approach to the market has been two things. One is we believe that the next era of buying really needs to focus on premium. When you focus on premium, a lot of the programmatic issues, fraud, non-human, all that sort of stuff starts to disappear. The second
Brett House (18:01.288)
Yeah, which is direct buy, right? Which is direct, but you have to be able to scale that considering the number of premium inventory providers there are, right?
Joe Root (18:09.381)
Exactly. So we're very well integrated into that premium market. So we bring those publishers along.
Rio (18:14.53)
You have a good, what percent of the, Joe, what percent of major publishers work with you guys? It's a high percentage, right?
Joe Root (18:21.199)
yeah, 60 % of media companies globally use permative. So we have kind of very, yeah, quite a large share across that bit of the market because, I mean, I think you made a joke about kind of publishers regardless always lose. It was quite a contrarian bet when we went to market to go after publishers, but I think we really believe in that model and the mission they're on. we kind of attacked a problem which I think
Brett House (18:24.735)
Yeah.
Joe Root (18:50.554)
not many people were thinking about at that point in time.
Brett House (18:52.776)
Yeah, and you made a really good point about how you're actually, it sort of filters out what is the fraud problem. We just talked to Dr. Foods. It was the second time he was on our show. And, you know, every time he comes onto the show, he just drops some data that he did and some controlled experiment with a client that just blows your mind, right? And a lot of what he was talking about the last time was sort of how you get this domain spoofing, right? And how SSPs will have spoof domains in the mix.
Joe Root (19:02.975)
Who gets this so well?
Joe Root (19:11.039)
potent.
Joe Root (19:17.171)
Yeah.
Rio (19:21.318)
Yeah, you explain how they launder the spoof domain and pass it on. Yeah.
Brett House (19:23.442)
Yeah. And once they make one jump out of the SSP to other partner SSPs or whoever partners with them, it's laundered immediately. It becomes invisible and you can't track it back. Even though they're actually getting paid for that impression, you can't track it back to a fraudster if you don't catch them the first time around. And so it sounds like you don't have that problem inherently when you're sort of managing direct buys in these curated playbooks.
Joe Root (19:32.308)
Exactly.
Joe Root (19:37.534)
Yeah.
Joe Root (19:41.991)
Exactly. No, so.
Joe Root (19:48.775)
Exactly. So our belief and the whole host of reasons why, but we think agencies really needs to get closer to the supply. And I suppose the offering we kind of take to market with agencies is, Hey, we take you closer to premium supply and we also uniquely able to make the premium supply signals available to you because we manage those. still today the SSPs and the DSPs and others can't absorb that scale of signal, but because we run in the publisher environment.
we can do your decisioning in their environment and then pass that through the supply chain.
Rio (20:22.776)
So as you're looking at agencies themselves, like you'd mentioned, you know, obviously with this, with this, the signal available from publishers, agencies are using it for some, for some curated deals for sure. But like, do you think like as there's, I mean, just, we've been a lot of talk about this push towards more value-based or performance-based contracts. I don't think a lot of it's happening yet, but there's certainly interest and there's more discussion than I've ever seen, whether that becomes a reality or not, I think is still kind of to be determined. But I think if someone could crack that code,
it would, they'd have a huge advantage. Like thoughts on like what you're doing, how it relates to that potentially and any trends you're seeing in agencies trying to move their businesses in that direction.
Joe Root (21:03.637)
Yes, totally. I think what we're seeing is clients are putting all this pressure on the FTE model agencies. And what they're really saying to agencies is, hey, you should make your money by being the best possible investment partner for us. And what does a really good investment partner look like? Well, there are a couple of pieces. One is like, hey, you need to invest my marketing dollars to drive the best possible outcomes for my business. That's a really key part being good investment.
Brett House (21:12.03)
Yeah.
Joe Root (21:33.205)
You should leverage the scale of who you are in order to go and negotiate really good rates as a good investment partner for me. You should identify where opportunities in the supply exist for me and go and negotiate those. So there's all these parts which make a great investment partner, but almost all of them are in that agency wheelhouse and strength of like, hey, if they can get really good signal, they can make really good decisions.
They're really good at negotiating and the principal trading business, used to be a dirty word, I think is very much back and vote because clients want that as well.
Brett House (22:09.344)
interesting. Rio and I have kind of been arguing against that in a sense, you know, simply because of the fact that, you know, with it's a pure reach play. Right. And so so they're really going like we need to fulfill these budget dollars. And so we're going to buy maybe more inventory that actually exists. Right. Like as in real human inventory. And we're going to buy as much of this at scale as possible, which is kind of a race to the bottom.
Joe Root (22:14.485)
So I...
Brett House (22:37.192)
because you start paying cheaper CPM rates, you're getting kind of fraudulent and shitty inventory, right? How does that support outcomes?
Joe Root (22:42.995)
Yeah. So I think, yes, I think outcomes are the counterbalance to principal trading. So the leverage the client has is, hey, here are the outcomes which you need to drive. How you drive those outcomes is down to you. The principal trading part is really the agency saying, hey, you don't want me to execute on client model. You want me to execute on outcome based model. I need to be able to leverage my own strengths to drive profit in this. And clients aren't
Brett House (22:58.472)
Mm.
Joe Root (23:11.827)
like naive or stupid, like they realize the agency needs to make money. It's how do you align the way the agency makes money to what we're doing? And I think like the agency takes a lot of risk in that. Like the agency is pre-buying inventory, huge.
Brett House (23:24.286)
Yeah, it's like they're sort of leverage, in a sense you're leveraging debt to invest in this because you haven't been paid technically by the client, right?
Joe Root (23:31.157)
precisely.
Rio (23:31.31)
It's funny you say that too, because like when we like we've done some in housing exercises with big brands who wanted to bring maybe not the buying itself, but the platform licenses, you go and paper direct with them some of the analytics to measurement first party data. We've done exercise like that. The hardest part in one of those in one of those activities is the contracting because you have to explain to the brand. You need to pay the platforms net 30. That's non-negotiable. Meta Facebook.
Joe Root (23:52.778)
Yes.
Joe Root (23:56.244)
Yeah.
Rio (23:58.574)
Google, they're going to shut you down. I say, well, we normally pay all our vendors net 90. That's just, that's non-negotiable. it's like when the unstoppable forces, the immovable objects, something's got to give, right? And then guess what? Like Google and Met are not going to give. So that's the hardest part. funny you say that.
Joe Root (24:06.566)
It's...
Joe Root (24:11.669)
Hopefully.
Joe Root (24:15.367)
Yeah, so I think really what we believe is that agency model is moving very quickly and actually AI enables them to do that and do that at scale if they can plug themselves into as much signal as possible to make good inventory-based decisions and to make good outcome-based decisions. And that's what we're...
Brett House (24:36.2)
And what does that architecture look like? And how do they, I'm assuming this has got something to do with there's a clean room in here somewhere. The reason why I say that because the publishers want to completely anonymize or pseudonymize their data because that's their golden goose. You can't just put that into another system. So you've got to do it through some sort of intermediary, which would be, is that a clean room in this case?
Joe Root (24:44.905)
Yeah, yes, yeah.
Rio (24:53.806)
So it's our first party data, yeah, for sure.
Joe Root (25:01.941)
Exactly. So what we see is you kind of need these two parallel pathways. You need a clean room for the agency to be able to bring their own data sets and pair that with all the publisher's supply-side signal and identify what supply-side signal best represents the data in the audiences the agency has. So if I've gone and built an auto-intender audience as an agency, I now need to go and find where that lives across the supply-side signal which exists.
Brett House (25:22.484)
Yeah.
Joe Root (25:31.413)
So that's kind layer number one. Layer number two is you then need to be able to go and activate that. And today, curation pipes and the way primitive integrates with that then allows them to take this insight from the clean room and very quickly turn that into a single thing which they can go and activate in the bid stream. So kind of the collision of these two.
Brett House (25:31.73)
Yeah. Yeah.
Brett House (25:48.011)
And is there an onboarding step in there in terms of the dirty word onboarding? Where you have to take that signal, you've got X number of IDs for X amount of reach, and then suddenly that's cut in half because you're going through onboarding pipes when you bring it over to the DSPs.
Joe Root (25:54.493)
Yes.
Joe Root (26:05.787)
Exactly. So I think this concept of onboarding is changing very rapidly. And I see this in our clients and our customers today. The old world of onboarding is, hey, I want this one-to-one match and I'm going to go and hit that onboarded audience. Now you're onboarding that audience to go and find insight in which signal should I hit. So, hey, I've got this auto-intended audience. Here's all the contextual content which they're consuming as they move across the open internet. Here are the audiences which they belong in.
Brett House (26:19.923)
Yep.
Joe Root (26:35.081)
when they're on all these different media company properties. the onboarding is a much smaller part of what happens.
Rio (26:42.766)
So it's not a one-to-one deterministic match for onboarding, right, anymore. It's more complex than that.
Joe Root (26:48.117)
Gone are those days. Because as soon as you do that, you have two problems. Your reach gets massively cut down. ultimately, can then only address somewhere between 10 and 30 % of consumers have an ID. So all of a sudden, you lost all this reach. And problem number two is that onboarding methodology, when you do it with Facebook, Facebook could pair it with their own signals and then finding your next set of consumers and your next set of consumers. But onboarding in
Brett House (26:54.6)
Yeah.
Joe Root (27:15.669)
programmatic and open internet has always been to retarget those consumers. And when all you're doing is retargeting, you're not growing your brand anymore because you're not finding kind of the next set of consumers and the next set of consumers. So that's kind of the second piece is like you solve this reach problem, but you also solve an incremental sales problem.
Brett House (27:19.305)
Yeah.
Brett House (27:33.022)
Yeah, and so it sounds like you guys have got a real-time signal of data. So a data flow that's going into the client's architecture, right? So it's going into their cloud data warehouse, whatever. On the agency side, maybe on the brand side? Is it generally on the agency side?
Joe Root (27:46.377)
Yes, so it's primarily on the agency side. We do this with brands as well, but agencies, I think, is where we found the biggest pain and business model change in need. that is, which I wouldn't have made the bet on five years ago, but that is exactly where the majority of growth comes from.
Brett House (27:53.331)
Yep.
Brett House (28:00.789)
Yeah.
Brett House (28:08.35)
So if that stream and so it's so let and tell me if I'm describing this right. So it's pseudonymized. You get the stream of signal that's much richer than they have now because it's cell side signal, right? It's contextual. It's got all that stuff packaged into it. And then they and then there's you guys I'm assuming are powering kind of a predictive analytics model of some sort that saying, hey, you know, based on specific outcomes, whatever the outcome is, we can start to find incremental audiences that you may not have seen that don't fit the sort of, you know,
apples and orange description of like auto in Tinder, but it might be somebody that's doing, which kind of lends itself to almost automated second party linkages, right?
Joe Root (28:43.495)
Exactly. Exactly. So it's like, hey, Tesla Intender, well, five weeks ago, they were reading a review of Tesla across her store on Tokyo, wherever it might be. Now, you aren't targeting the 70 % of consumers reading that content who don't have an ID. We need to flip this principle. We need to transpose this. And actually, you want to go and hit the consumer, the intent signal.
not at the ID signal.
Brett House (29:13.95)
Yep, yep, which is very contextual in a sense, right?
Joe Root (29:17.961)
very contextual, very behavioral in nature, but a whole lot of data which today the programmatic pipes just cannot tolerate without destroying their cost margins.
Rio (29:30.062)
So Joe, this is the probabilistic targeting that you've been talking about quite a bit, which, you know, it's pretty powerful, right? I mean, and you're right. Like people have not been taken advantage of this, but like, if, let's say this does become more of the norm, right? Like, what does this do? Like looking at SSPs, DSPs, does the balance of power shift? Like, what does it do to, I mean, if decisioning, a lot of this is starting to happen closer to sell side, it is taking away functionality that DSPs do today to a certain extent, right? So comments on that or thoughts?
Joe Root (29:57.525)
Totally. So, yeah, so kind of four players we think about, right? For the media company, well, now the 70 % of the inventory which was ignored is being bought and they see CPMs go up. So when we see these deals being done, they typically see a 2X increase in yield. For the client downstream, now you're reaching a bigger audience, you're seeing sales go up and you're seeing CPAs go down. So we're seeing like somewhere between kind of 100, 120 % sales uplift.
And usually CPA is coming down by half. For the agency, now all of a sudden, the scale of your audience has gone up, you're reaching more consumers. So you're able to make more money on the budgets you trade because ultimately Facebook's not taking a principal trading deal with you. You're only making it when you're going on the open internet. So those three parties all benefit. You then have ad tech, which is kind of some players benefiting, some losing. So right now I would say like objectively.
Brett House (30:42.686)
Yeah.
Joe Root (30:53.001)
the DSP is losing in curation and the SSP is winning, right? It's just a hard fact. like the trade desk gets a lower take rate when this is being done in a deal ID and in a private marketplace and the SSP is able to charge more fees when it's being done in a private marketplace. So like we're seeing this transitioning of as decisioning moves from buy to sell side, we're seeing a shift in effectively who is able to kind of take margin.
But I think if you re-zoom out, re-agencies under hue. yeah.
Brett House (31:26.026)
A shift to the SSP. Hold on. on. Let's have a look. So there's a shift in who's taking margin towards the SSP side. So the trade desk, for example, has open path and they've sort of gone to that play. Do you see a future where DSPs in the traditional way we define them kind of go away? Because it kind of reminds me of the take rate conversation we had with Dr. Fu Ryo, where he was saying that what's happening in the buy side is that the take rates are for every
Joe Root (31:35.26)
very yet.
Joe Root (31:49.823)
Yeah.
Rio (31:54.99)
Yeah, they're just cranking it up.
Brett House (31:56.235)
Yeah, they're cranking it up for basically every incremental penny on top of the ask price. So let's say you ask $1.50, sell for $3.50, and they take 99 % of the difference between the two.
Rio (32:07.022)
Yeah. The insinuation is like they're having like, new quarter, we got to our numbers. You know, like he was in, was saying some of the major DSPs were able to just like raise the take rates dynamically, you know, to your point, Brett, to, gobble up the incremental bids as they're, as they're, as they're moved up. So yeah, it's not good for brands or publishers.
Joe Root (32:16.351)
Yeah.
Joe Root (32:22.965)
Yes, mean, I think, yeah, I mean, you're seeing DSPs are going direct to supply, SSPs going direct to agencies. You're kind of seeing that, I think the dichotomy, like this kind of dichotomy or this kind of biification of buy and sell side, that's kind of disappearing, right? And like, actually we're just seeing, there should be a single direct connection between the agency and between the publisher and both DSP and SSP are trying to do that. But they're all like quite founded.
Brett House (32:23.816)
Yeah.
Brett House (32:41.385)
Yep.
Brett House (32:50.856)
Yeah, you've heard it here at Signal of Nose, Joe Root is calling for an end of the middlemen, right? A lot of the middlemen, right? Direct connection. That's a topic that keeps coming up again and again and again. It's like the middlemen are taking a cut. And 40 cents to every dollar, 50 cents to every dollar, whatever it is. Yeah, they're toll takers.
Joe Root (32:56.623)
Yeah!
Rio (33:00.62)
Makes sense.
Joe Root (33:05.173)
Yeah.
Rio (33:08.462)
You're toll takers, right?
Joe Root (33:10.037)
Yeah. So I think, yeah, I think this is the piece which is really happening. on one part is it's not like the DSP serves no value if it's not doing decisioning, like to the point on net 30, right? Rio is like, actually the DSPs are like lending to an entire ecosystem because the publishers getting paid in like 30 days, the DSP is getting paid in 120. It's really the DS. Exactly.
Rio (33:33.164)
And the SSPs play a role too, right? mean, they all play a role in delaying payments or facilitating payments. They're the market makers, right?
Joe Root (33:39.893)
Exactly. So I don't think we should underestimate like the non-technical role these folk play, which has a lot of value. But yes, the value equation is changing. And when you zoom out, agencies are under margin pressure, media companies are under huge margin pressure. And then objectively, there's a 70 % margin take in between those two. Both sides of that are saying, hey, how do we get more direct to each other so that we can find a margin opportunity ourselves?
Brett House (34:09.748)
Yeah. you see a future where the agency is moving up the value chain in terms of the services they're offering? Because I think there's a certain commoditization play when they're just doing the principle-based media buying, because a lot of that is being automated. And so to me, I'm seeing this picture from your mind at your company, where if you're providing more signal,
Joe Root (34:10.097)
it's just a reality.
Brett House (34:37.322)
So it's not, you're not stealing the publisher data to the agency. The agency has, I can layer on a lot of analytics and decisioning and strategic value on top of just the media vibe, right? Which, increases their value as a partner. Yeah.
Joe Root (34:49.598)
exactly so.
Rio (34:52.14)
I think they have to do that,
Joe Root (34:54.645)
Yeah, I mean, I think kind of we really see agencies needing to take a lot more control over kind of how they invest and allocate that media and they need to make take more control as kind of custodians of their customers dollars. So these principal trading deals and negotiating rates like you're effectively looking for a really good investment partner.
Now our belief is AI will enable them to do a lot of this stuff better. like before, I need to staff a team of a hundred people to manage all these publisher relationships, but I could staff a team of 10 to go and execute in a DSP. Like that made programmatic just so much more scalable in that FTE model type system. But now, if you really look at what's happening, it's like, hey, agencies,
can leverage agents to do a lot of the direct buying. And maybe we can get onto what we're seeing on the supply side with publisher agents. But I think agencies are going to start to build all their own agents. And you're seeing that in their operating systems today as they all roll those out. Our job and our belief is we should provide good signal for that.
Rio (36:00.578)
Yeah, je-
Brett House (36:01.053)
Yeah, we gotta tackle this one Ria. is yeah, this has been a hot topic
Rio (36:04.278)
Yeah, Joe, that's so yeah, good pivot here. I love this topic. So we've talked about it a lot here. I actually wrote a series of articles about, you know, agentic trading. Is it happening? What's, how is it going to affect the ecosystem? Is it good or bad? I mean, it's, I've spoken with people at several of the big holdcos and I don't think anyone's actually doing it. Certainly not at scale yet, but there's a lot of testing. There's a lot of enthusiasm. I want to get, you're taking a couple of things. We, number one, like state of the union, like, it happening? Do you think it's going to timeline? I think number two would be.
Like if this really happens, like what does this do? does, if agentic trading is going to really focus on allocation, does this mean that agents are actually hammering out deals by themselves? Does this mean that, what does this do to the auction? Does the auction, does this happen pre-auction? They're just planning? Is it more planning exercise or are they actually making deals that are, know, with more like the old way with price sheets, but are happening at scale because agents are doing it.
And maybe the auction gets relegated to more like remnant inventory. don't know. I mean, it could really change things dramatically or could change things very not much at all. I posted something the other day about, you I was kind of joking like, well, if this, if agenda trading happens, what happens to DSPs? Will they be around? And I think Adam Hyblick from Chalice, he replied and said, well, you can't bid on something if you don't.
know, if you don't know the price, so which is a valid point as well. So like his point was the auction is going to still firmly be there because you need it right in the DSP and all of the bidding happens at a DSP level. So thoughts, I know that was a lot of thoughts on this Joe, like you know what will happen and then they'll impact.
Joe Root (37:36.979)
Yeah.
Brett House (37:38.142)
You can tell Rio's passionate about this topic.
Joe Root (37:40.175)
Yeah, so I wish I had a crystal ball which predicts it perfectly, but obviously we have a company strategy which we make our investments and decisions on. So we have like a very strong POV. I suppose in terms of like the story is like two and a half years ago, we're broadly speaking a supply side business. All of our customers are primarily media companies. Like it's advertised that business is rocketed, but you know, the two year thing.
So two and a half years ago, we started to realize, hey, we can make these like quality of life improvements using AI within the platform. can like allow you to search better within the platform because it kind of identifies kind of the actual meaning behind, for example, an audience or a segment name. That was really interesting. And all of sudden our customers were able to do things a little bit faster. Then around a year and a half ago, we realized, hey, actually like these agents could pretty intelligent now.
and can actually surface decisions back. if we take this AI and turn it into agents, what can be done differently? So we started to build agents for audience building, for campaign planning, for campaign optimization. We before last kind of speaking with one of our customers who now uses permutative Halo exactly. So we have this suite we call Halo, which we released across the cell site. They've gone from RFPs taking them five hours
Brett House (38:51.828)
The halo, right? Is that your agent? Yeah, yeah.
Joe Root (39:02.557)
sometimes often kind of a couple of days to 15 minutes. So all of a sudden now they're able, that's just the real thing.
Brett House (39:06.314)
Yeah, yeah. And that's not hyperbolic. That's like the real thing, right? Because you've got a memory system that you've built around how you, there's only so many ways to slice an RFP pie, right? And you just train the system, which is hugely effective from that perspective. Yeah.
Joe Root (39:16.552)
Exactly.
Rio (39:19.63)
What a time saver, yeah.
Joe Root (39:21.141)
Exactly. So like even this morning, I got a video from, from one of our engineers who kind of built out a series of chord skills for our clients to respond to RFPs. He sent me over the demo, like start to finish. It was 10 minutes. I put it on 2X, get through it quickly, but it's like, it's crazy. Like he had a fully generated RFP response 10 minutes after kind of entering in all this detail and chord is basically just going backwards and forwards with permutative to assemble all this. So
What it allows is like, another really good example is like this client was only optimizing 10 % of their campaigns because that's all the AdOpps resource they have. We have these optimization agents, which now surface up a load of decisions for them to make on a campaign by campaign basis. And that AdOpps seems to using that for optimization. So all of a sudden,
Rio (40:05.23)
Right, so it's basically taking something that would have taken, like the limit was people doing it, right? Because it's so time consuming, it's so labor intensive, but you're saying with AI, can just, instead of doing 10%, you can do 100 % or something close to that.
Joe Root (40:15.573)
Exactly. So now like the throughput, which you can put through direct pipes, like all our publishers teams are dealing with this primarily like direct sold stuff. And all of a sudden you can put 10, 20, 30 X more through that for the same cost of headcount. So that means like on the sell side, you've got this capability to transact directly at a scale you would never thought was possible. And on the buy side, what agencies are doing with their operating systems, like
You play that for another six, 12 months from now and like, hey, you can meet these two worlds. And now you've bypassed programmatic pipes and you're able to trans that very directly.
Brett House (40:55.944)
Yeah, and you're-
Rio (40:56.088)
Joe, that's interesting. So sorry, but I wanted to point that out. No, pass over to you. So that's interesting. Cause remember like food, bread, if you remember last time, like when we were taught, yeah, we're, talking to him. mean, like, and he was saying like his, his thesis was these advertisers should be going directly to an agency should go in directly to the publishers through their portals, not through, like he was basically saying, don't go to these DSPs, go straight to the buying platforms themselves buys. And, I kind of, counter to that was like, then you can have 15, 20, 30 places you're doing your buys. That's.
Brett House (41:03.498)
I was just going to, yeah.
Rio (41:26.03)
I mean, that's really difficult, all the swiveling and managing of that, and then getting people to do it. And he said, well, an agency should be able to do it. I don't know. mean, that's a lot.
Brett House (41:26.334)
Well, yeah.
Brett House (41:34.761)
I mean, you can, you can argue that 70, 80 % of your buy is going to be across five. Like the number of publishers is like, can count on two hands, right? But I think his, one of his big points was that if you plug AI into basically a polluted open web supply, right? Which he's claiming that the open web supply outside of the signal that you're talking about. And I kind of want to lean towards that to say, Hey, you've got a lot of authenticated traffic, right? People that have gone to these publishers, they have, they have a
Rio (42:03.106)
From premium publishers, yeah.
Brett House (42:04.296)
Yeah, they're premium publishers. Some of it's probably open. Is it all premium or is it all authenticated? Okay. So you don't have to worry about all of the fraudulent bullshit that's out there in terms of supply that are just like domain spoofs and all that other stuff. Because if you plug it, if the agencies plug their agents into that sort of source of truth, it's polluted. Right? And you're just going to scale nonsense. You're going to scale just empty impressions. Whereas you're saying you guys have an argument because your data source
Joe Root (42:04.681)
Yes.
Joe Root (42:08.551)
All premium, yep.
Joe Root (42:18.3)
Exactly.
Joe Root (42:23.118)
sizing. Yeah.
Joe Root (42:29.097)
Yeah. Yeah.
Brett House (42:34.403)
is authenticated and rich and deterministic, right?
Joe Root (42:38.811)
Exactly. So I think there are like a number of these factors which are converging on, you want to go direct. So like our role in this and our bias is, hey, we can provide the signal for this next layer of kind of AI agents. And we can also provide an interface which connects you directly into the supply. So I have a bias POV. The other side of this, the of it, It's like, think open
Brett House (42:57.392)
No. I mean it sounds really compelling. Where do I side the dotted line here?
Joe Root (43:08.613)
Open internet is very rapidly changing and actually kind of a lot of these challenges were as we built these systems to buy across thousands and thousands of like publishers with search, just sending so many people to random blogs and web pages. And it created like a very perverse incentive structure, which fed a lot of fraud and all these other pieces. The reality is like AI search has killed the long tail of the open internet, right? It's like.
you will go to your trusted sources. So you'll go to the Financial Times or the BBC or whoever it is for the news and the content which you trust.
Brett House (43:44.328)
Yeah.
Yeah, but everything else is happening directly through a chat experience.
Joe Root (43:50.832)
Exactly, search is just...
Rio (43:51.16)
definitely whittle down the number of websites I visit. Yeah. I mean, he used to have all these tabs open. Now I don't. I'll just have one or two. And you're right. The ones I always go to, still go to, but like the most of the other ones I was just randomly going to because of a search I'm not anymore.
Joe Root (44:02.747)
Exactly. So like now, if I'm a media company and I want to get reach in any given market, they're probably in 30 to 50 destinations who I need to refocus my relationship on. And hey, the side effects of that as well is like, one, can negotiate principal trading rates. So there's an opportunity for me to make money. The inventory is much higher quality. So I'm not having to deal with fraud and all these other things. I get really rich signals so can optimize towards better outcomes. Like there are a lot of very good
things when you do that, not because like programmatic and the way we did things was bad. made sense at that point in time, but in the modern world, the way things have changed, it doesn't make sense to buy in that way.
Rio (44:48.206)
In this new model, who, like, where, like, so it sounds like you're working a lot with agencies, Um, um, and it sounds like the holdcos, indies, all the above.
Joe Root (44:59.529)
Yes.
Rio (45:00.972)
And then you have these relationships with publishers. Are you working with buy side at all? You work with marketers? Has that started?
Joe Root (45:07.124)
Yeah, so we have kind of two sets of client on the buy side. One is agencies, the other is the actual advertiser. The reality is though, if the advertiser has the direct relationship with us, it's still the agency who are going out and executing all that stuff day to day. So our paperwork may be with a different team, but nine times out of 10, the kind of folk who are working with all the agency folk in the day to day of it. And I think kind of
What we see is, think we are most interested in how does AI transform business models because we think that's where there's the real opportunity to go and build. And this transformation, the agency business model is just so intense that it is a great time to go and build with these guys.
Rio (45:53.838)
Okay, so questions. So the agencies are trying to build these like AI driven workflows that can aggregate, like say, like their scale demand across the different places they're buying media, right? I also see like the platforms are trying, they're also trying to build their own AI, like agentic layer on top of this. And then you see some brands are building their own agents. Like who wins the battle? Like where does this actually, where does the buying eventually take place or does it not matter?
Brett House (46:18.655)
Yeah.
Joe Root (46:19.517)
Yeah. Yeah. So I suppose state of play today, as of this very moment, agencies have gone and built these kind of operating systems for planning all the way through to activation, because it's part of an efficiency drive within these organizations.
Rio (46:37.038)
I mean, they're doing all that stuff today. It's just automating it, know, automating what they already do, right?
Brett House (46:39.742)
Yeah. And to wear the black hat, some people argue that they're not really operating systems. The word OS is taken on a bad word. In some cases, they might be further ahead, but that it's just a collection of point solutions with an AI wrapper on top. That's why black hat counterpoint to the operating system argument, which I think makes sense from a positioning and slide where like, we're bringing, it's kind of like the marketing cloud talk like 15, 20 years ago, right?
Joe Root (46:39.994)
Exactly. Exactly.
Joe Root (46:49.502)
Yeah.
Joe Root (46:55.594)
Yeah.
Joe Root (46:59.251)
Joe Root (47:06.376)
Yeah. Yeah.
Rio (47:08.206)
We'll buy these companies and we'll smash them together.
Brett House (47:08.308)
We're bringing it all together. Yeah, but is it, how far do you see and how mature do you see that from your agency relationships?
Joe Root (47:11.06)
Yes.
So the most mature bit we see within this kind of operating system stack is the data business, Is in the hold codes, they've all gone and built them or they've merged to kind of bring the assets in like Axiom into Omnicom. Or alternatively, what you see in the indie landscape is kind of these partnerships like Horizon with TransUnion for Blue. But regardless, kind of that is very real and exists today.
Brett House (47:20.82)
Yeah.
Brett House (47:28.458)
Yep.
Brett House (47:37.043)
Yep.
Joe Root (47:42.197)
And I think a lot of those workflows are being improved and made better by AI. But the output of that is basically a load of IDs, which the agency then needs to go and activate. And the challenge they have is back to like the old, but the problem of addressability, right, is they go to activate it and it doesn't scale. And then their client teams get frustrated and then they have to go and use data sets, which are not part of the operating system and the operating system loses adoption.
Brett House (47:55.77)
Back to the problems we've been discussing a lot. Yeah.
Brett House (48:08.562)
Yeah. Yeah. So it goes, almost goes back to the same onboarding problem. We've got all of these ideas. We've got a match to, wherever our DSP, our sort of ad serving is, and you lose all the scale. Yeah.
Joe Root (48:13.158)
So that's
Joe Root (48:19.217)
Exactly. So that's the problem we solved today. It's like, hey, we're just going to triple the reach of the output of your operating system by mapping it onto all the supply side contextual signals and behavioral signals so that you can keep all this working through that one operating system and flow. And that's kind of our value prop back and that enables them to make those businesses much more robust and bigger. But it also gives them the foundations as they start to build out
their own AI agents, as you see principal trading start to play a bigger role, as you see them trying to go directly, well, hey, those same data foundations which are feeding your operating system can also feed your AI agents. And that's kind of a broader belief we have is like, actually the agencies can really win if they start to do this. And I think the reality is, is like, they'll probably do so at the expense of some ad tech players.
Rio (49:18.776)
So looking at sell side, like looking at publishers, we've talked a lot about how they've, know, whatever's happened in media, they've gotten the short end of the stick, right? I think that's an easy argument to make, right? When the Craigslist launched, they lost what? Up to 40 % of the revenue, just from classified ads. And then the shift to digital, mean, like, you know, the quip was always traditional dollars to digital pennies, right? I mean, it's been a tough 25 years and now with AI, right? I I have clients who've had seen
Joe Root (49:28.424)
Yeah.
Joe Root (49:36.01)
Yeah.
Joe Root (49:43.146)
Yeah.
Rio (49:48.494)
productions of 20, 25, 30, even up to 40 % of the traffic's collapsed. And a lot of people who don't know the publishing business well don't realize that's actually money in their pocket that's being taken away. As the traffic goes away, you know, it's a big revenue source, right? Is the monetization they have from their properties digitally. like in this new paradigm that we're entering and what you're describing, what you're trying to build, right? Like, will this be better for publishers? Should publishers maybe, is this a golden opportunity for them? And
Joe Root (49:58.772)
Yes.
Rio (50:15.724)
You know, like, or could they lose out again to another, like maybe new kind of toll takers who swoop in and, and, and begin to, monitor, just take away from their monetization. What do you, what do you predict for publishers and like, what would be your advice to them?
Joe Root (50:29.589)
Yeah, so broad rules done for a publisher today. If you're getting bought programmatically, you're probably getting bought at a dollar. If you're getting bought directly, probably, yeah, you're getting bought between 10 and $20. Now 80 % of your inventory is likely being bought programmatically, 20 % of it is likely being bought direct. In some cases at 90, 10, in some cases 70. So every publisher needs to grow directly. And that is the number one strategy. So publishers...
Brett House (50:37.15)
to written in inventory.
Brett House (50:49.684)
Figure out a way to turn that, switch that.
Joe Root (50:58.031)
are proponents and supporters of, if agencies want to buy us more directly, we want to enable that. Now, for a publisher looking to take control of this, why is an agency going to buy you more directly? They're going to buy you more directly if you help them with their data business. So if you can bring their data supply side and scale it, well, hey, you can plug into that. They're going to buy you more directly if you deliver outcomes for their clients.
You need to start focusing on driving outcomes within your own direct campaigns. And what do you need to kind of support this? Well, you need all the right signal available to better deliver on those two value propositions back. And you also need an agent layer to support it.
Brett House (51:37.535)
But you also need to prove outcomes. That's a caveat here because publishers have always struggled to prove outcomes because they don't control the point of sale. Once an audience bounces off of their site or their properties to somewhere else, those are all living in different worlds. Is that a partnership play where they're working with incrementality, MTA, whatever measurement providers?
Joe Root (52:03.573)
So the unfair advantage Google and Facebook have is the conversion API. So Google and Facebook get fed this measurement data and then they optimize all their systems towards that. Publishers have been in this really unfair position that they...
Brett House (52:09.193)
Yeah.
Rio (52:17.551)
Yeah, you could argue they fudge, they fudge attribution a little bit, but keep going, Joe.
Brett House (52:21.074)
Yeah, well, then the retail will add the retail media from a walled garden perspective. I'd like the Amazons where they actually control in lot of cases the point or Walmarts, the point of sale. So outcomes are easier to prove for that type of advertising, but keep going.
Joe Root (52:21.449)
Yes.
Joe Root (52:31.551)
Excellent.
Exactly. So the reality is publishers, as they go direct, the big change a publisher is really trying to make with their outcome story is become more strategic. So for a publisher, they'll have a series of endemic clients who they already have a very good relationship with them. The first port of call is, hey, actually, let's start talking about the sales outcomes. Let's start talking about kind of the overall strategy you're trying to execute. And then you plug our conversion data, your conversion data in.
and we'll start to optimize. And that's what we're seeing with kind of the outcomes, the optimization agent, which Pimative has, it's like, you can feed that data in and now you can optimize towards an outcome. That requires a change in like the sales team and the way you go to market and all those pieces. But for as long as you don't get that.
Brett House (53:05.95)
Yeah.
Brett House (53:20.156)
Yeah, maybe their tech, their tech team, in a sense. And you're you're sort of saying it's almost the platformization, so to speak, for lack of a better word, of the of the open publisher space or the publishers base. Right. They've got to compete against Mehta. They've got to compete against Google. Right. And there's some people that argue that the open web is just doomed, that everything's going to be walled garden, you know, publishers or platforms. But you're saying, hey, if they yeah, if they start to replatform, which goes back to our very first podcast.
Joe Root (53:29.343)
That is, like, I had that exact.
Rio (53:44.248)
Well, the gardens are hedge gardens,
Brett House (53:49.832)
with Stephanie Laser, it was called the great replatforming. And this is kind of leaning down that path.
Joe Root (53:53.865)
Yeah. Yeah, I had a, I was on with one of our CROs last week who used exactly that language. He's like, Hey, we're building a platform here and it's a platform we need to deliver on outcomes. So I think that is totally right now. Five years ago, we were saying, Hey publishers, you should execute wall garden strategy and publishers saying to us, Hey, like this is good. We're selling more directly, but these agencies don't want to work with 50 walled gardens. want to work with one AI.
Brett House (54:21.716)
Yeah.
Joe Root (54:22.483)
and agentic buying really changes that because like you didn't want to work with 50 because that's 50 different wall garners you have to manage.
Rio (54:30.818)
But now your agents can. They can do it at scale.
Joe Root (54:34.097)
Exactly, and even better now also if I'm the agency, if I can negotiate the right kind of training agreement with those publishers, actually those relationships can start making me money and they're much more efficient and effective to run. it's kind of that business model change for us means that we are really invested in this agentic future.
Rio (54:56.184)
So I guess, you know, that's basically answered the question from before, right? If we go to agentic trading, could be very good for publishers, right? Because it enables them to have more direct sell deals and allows them to be able to counter that argument of like, okay, you don't need centralization anymore because AI can automate this and do it at such a scale and such a pace that humans could never do it. I you can never have enough people to do it. They couldn't react fast enough. So AI could do this.
Brett House (54:56.33)
That's awesome.
Joe Root (55:20.565)
So coming back to that.
Brett House (55:23.528)
Yeah, it's going to go to where the signal is, right? It's going to go to where the rich signal is and gather as much Intel that it can gather, right? And you guys are sort of, how do you make it visible? I how do you, mean, because it's all within your system, right?
Joe Root (55:27.861)
Great way of framing.
Yeah.
Joe Root (55:38.165)
So the signal, to your point earlier on clean rooms, we surface the signal in two places. We surface it today inside the clean rooms for them to go and identify what signal they want and where. And then today we make that signal available in two places. We make it available in the programmatic pipes, and we make it available in the ad server. And the question is, are the agencies going to continue to transact by those programmatic pipes?
or are they going to start to build agents which allow them to transact via those direct solve pipes? And time will tell us which path we go down, but I think there is a lot of financial incentive for both the publisher and for the agency for that to be in more direct relationship. And I think those financial incentives can drive a lot of behavior.
Rio (56:29.624)
Show me the incentives and I'll show you the outcomes. Love that quote. Charles Munger, yeah.
Joe Root (56:32.371)
Yeah.
Brett House (56:33.512)
Yeah. Yeah. Why do we rob banks? That's where the money is. Sorry, bad jokes aside. So should we go to quick hits? Because I know, Joe, you've got a meeting coming up right before possible. So I'll start. Who actually controls the AI ad stack today? And we kind of talked about this. You have your vision by your sell side today. And where do think that's going?
Joe Root (56:41.127)
I won't use that one for the analogy.
Rio (56:43.47)
I love it. Let's do a quick hits.
Joe Root (57:02.229)
I think we bifurcate agencies on one side, publishers on the other, ad tech not really controlling that.
Rio (57:12.206)
Joe, what's the most overhyped and conversely the most underhyped use of AI in advertising right now?
Joe Root (57:22.469)
Good question. Most overhyped is probably the use of a code in these LMS overhyped because no company we work with has that code set up in a way in which they can actually leverage it properly. Underhyped, like they don't have a core strategy underhyped in the sense that this is transformational to business workflows. And I really do believe like
Brett House (57:39.838)
Yeah.
Joe Root (57:51.051)
10x is the capability of what these organizations can do.
Brett House (57:54.389)
Yeah, 10x, that might be underestimating that. think it's, yeah, like I think as you get more sophisticated with these things, we've had a bunch of these conversations, you start to get to like, hard to quantify exactly, but anecdotally you can be like, no, it's not 10x anymore. It's more like 30x or 40x, right? It's incredible when you set up the systems, because it can be repeatable as long as it's governed properly, right? Which I'm kind of learning the hard way. I'm trying to get as technical as you were back in your computer science days. It's hard though, man. I was an English major, you know?
Joe Root (57:57.245)
Yes, yeah.
Joe Root (58:14.58)
Yes.
Joe Root (58:20.277)
Ha
Brett House (58:23.37)
I'm change the way I think about things. Full circle here. So, so.
Joe Root (58:24.437)
Turns out computer science degree has no benefit with this either.
Brett House (58:29.546)
No, you're just as like...
Joe Root (58:32.65)
Yeah. Exactly. Yeah.
Rio (58:32.974)
Well, a lot of code now is just English now. mean, just your code's total, the concept, what is code? I I'm glad I studied computer science like up until I actually went through a couple of years in college, right? And like, I'm glad I didn't finish the degree. I regretted it over the years, but now I'm actually kind of glad. I'm glad I think being a journalist and learning how to write well and think well and reason and argue is probably as useful as anything these days.
Brett House (58:36.596)
Yeah.
Yeah, I've just, yeah, I've gotten to this.
Brett House (58:57.706)
I tell you, and I think, and I always talk to like product leaders and technical people and I'm like, you just learned that three months ago. Right? Like I've been no time of my life until now. I've I had my terminal open, right? GitHub, right? You know, codecs, Claude, right? Maybe like a Hermes or open collage. Like it's just changed. It's just a completely different world that we're kind of working through. And it just gives you all this ability and control that you.
Joe Root (59:03.241)
Ha ha
Brett House (59:25.084)
and buildability, you're less reliant on other people to do some of this stuff for you, right?
Joe Root (59:28.211)
Yeah, yeah, yeah, I think we see the role of engineers is just changing so radically, but actually once you can think like an architect to like 10x more effective impactful.
Brett House (59:40.658)
Yeah. So next quick hit, is curation still relevant in an AI driven world? Or is it going to be abstracted away, so to speak?
Joe Root (59:49.043)
I think curation is a technological pipe, hugely valuable if programmatic continues to be a thing. Curation in its first generation of, I'm going to package up inventory and resell it. I don't see that model as a long-term model.
Brett House (01:00:08.275)
Yep.
Rio (01:00:10.03)
Joe, if you were to bet, would you say open web gets stronger or weaker because of these trends that we've been talking about and what's happening with AI?
Joe Root (01:00:19.125)
I think premium end of the market becomes stronger. If they can just double the amount they direct sell, they grow revenue enormously through this period. Let alone if they can turn that 20 % into 60 or 70%, that is huge, huge growth for these businesses, given that it's replacing primary $1 programmatic with $10 direct. I think long tail open internet, I just don't see how it survives this era.
Brett House (01:00:42.506)
Yeah.
Brett House (01:00:47.154)
Yeah, so move towards subscriber based and then all if they want to survive, they've got to authenticate and have subscribers, right?
Joe Root (01:00:54.301)
Whether it's authentication, subscribers, ultimately, if you've got a depth of relationship with the consumer where they trust you and you are part of that consumer's habit, you are in such a strong position. And I think so many media companies are part of the habit of the consumer.
Brett House (01:01:07.593)
Yeah.
Brett House (01:01:11.272)
Yeah, and back to what we talked about, think this is a concluding remark is that our media consumption strategy or approaches have changed, And we tend to, all that long tail stuff is now being siphoned off by the LLMs and you're going direct to a handful of publishers, a handful of CTV providers for basically all of your content, right?
Joe Root (01:01:22.569)
Yeah. And I mean...
Rio (01:01:26.126)
Yeah, the aperture is closed quite a bit. It's narrower.
Joe Root (01:01:29.973)
And also like for a lot of our media companies, you may spend 10 minutes with them each day getting caught up on the news, but you're spending three hours with them on podcasts. So our media companies are just diversified so heavily into video, into audio. The large players are everywhere. They're not just kind of web.
Brett House (01:01:54.185)
Yeah, yeah, the old editorial model is can't stand up on its own. Well, this has been phenomenal. mean, any additional questions around? Yeah, this is great. Joe, you are a blast. The real super happy to meet you. was just just ton of fun. We should definitely meet up in the city sometime and we're to get real to move back to New York City out of Denver. Yeah, it's I'm telling you, it's it's inevitable. I've been here for 25 years and I may never leave. I'm like, I take the old Lou Reed.
Rio (01:02:00.12)
Great conversation. No, no, this is, yeah.
Joe Root (01:02:01.663)
Yeah, loved it.
Rio (01:02:12.688)
My wife wants to, it's funny. She keeps trying to push me on it.
Brett House (01:02:22.058)
He's like, he's like, why would I ever go anywhere else? And I was like, yeah, that's kind of why I moved here. It was that new, that Lou Rienop of New York, right? With him and his, I was like, I like that guy, man. I want to go live with him. It is a great city. That London. Yeah. Yeah. Well, hey, thanks for having you on. And thanks to the audience for joining us today and making it through the end. If you made it this far, join us at debdebdeb.signalanoise.ai. That's where we publish all of our editorial.
Joe Root (01:02:29.798)
It is so.
It's a great city. I haven't seen the Eurip yet.
Brett House (01:02:50.664)
video, podcast content. We've got a bunch of executive contributors that contribute content as well. And we are on Yahoo. We're not on Yahoo. We're on YouTube, Spotify, and Apple Podcasts, wherever you listen to your podcasts or watch your videos. So thanks, everybody.
Rio (01:02:59.369)
We're on YouTube.
Joe Root (01:03:10.303)
Thank you.
Rio (01:03:11.918)
So long.
Comments