I Have a Thesis

0:00 How the buy side spins dollars around Garrett steps up with that day to finesse selling it right no room for a guess the mics are hot, the beats hit hard Fans throw their hands high in the yard Deals and impressions

0:15 like a verbal spray we're here today

0:16 to learn how to give your opossum a proper massage. Breathe. And begin. Pull up on a rib cage. Oh, that feels good. And this is good for the opossum about, oh, three or four times a week. Sun, moon, sun, moon. Doesn't she look relaxed?

0:48 Say sun moon. Now the word. Did you find it?

0:57 What did we just see? I wish I had a screenshot of Gareth's face in zoom. That was amazing. I mean,

1:14 yes, hello.

1:15 Welcome to I Talk. Let me snap myself out of my stupor from seeing the opossum massage. I'm Gareth. I represent the sell side.

1:24 And I'm Adam. I represent the buy side. It's July 31st, Friday at noon and time for Ad Tech. ADTalk. We'll be taking your questions, comments, suggestions, thoughts and roasts.

1:37 Roast Amazing roasts. We don't get enough roasts. I think what I found so troubling about the video that we just watched was the appearance of the lady. She seemed like she was out of the Brady Bunch. And that was so at odds with the absurdity of the possum massage that it, I don't know, it warped me somewhere like some sort of horror film,

2:04 you know, it seemed like it was designed to cause a flood of questions. Do people have pet possums? Do they need massages?

2:12 I. That was a pretty docile looking. I would have assumed a possums would be biting and scratching in that scenario, but I guess they can be somewhat domesticated. Super interesting, but lots to talk about today. I want to talk about a few Gamera theses that I've been kicking around.

2:32 So.

2:33 So Adam, I do feel like your Adam loves ad tech post has been like, not unsurprisingly, some of the thinking in there is chalice informed thinking. Right? Like you wouldn't be building something that you didn't think was philosophically aligned with where the market is going. I really enjoy that and I have some philosophical thoughts for Gumro that I think will be interesting. Talking points to that are similar that are like, hey, this is, this is why I'm.

3:02 I'm approaching Gumro the way that I am and I make my compromises like everyone does, but I have a thesis, right? There's something down underneath that that is informing. this is why I think this is Going to be valuable and like going to affect things. Then we'll talk about model weights. Yeah. Which I've been reading a lot about modern weight models.

3:25 Yep. Yeah, we'll dug into that controversy so I can educate.

3:30 Oh, I would love, I would love

3:31 to be caught up on it.

3:33 I've. I've read a few articles on it and I've been educated by the AIs, but I don't know that I can trust the AIs to educate me on this one.

3:39 That's, that's why it's a tough topic because like people are. Yeah, we'll have to, we'll try to map the self interest and lack and what seems to be opposite self interest of the public postures taken on it.

3:50 Yeah. And then we have to talk about ads for AI. I know that you want to stay neutral on it, but it's a really interesting topic and I've been, I've been back and forth with the open ads guys. So the open ads guys are. Steve, listen, Michael Bishop and they were, they were really early on this. Right. They were writing about this in 2023. They have been talking about this for a long time and I think it's a fruitful topic.

4:17 It's a really fascinating one and I also think it's counterintuitive because most people don't realize why this would be a thing. So it's worth explaining what it is and why it would happen. So let's start out.

4:29 Thesis. Yeah. So I released my blog about three weeks ago. Three pieces out so far. I have two more in the can that will come out in the next two weeks. Try to keep going on that. But the thesis of the five part series is the two architectures and it's what I've decided after six years of leading Chalice that the scaled software for an AI application for enterprise advertisers is different at the core from what we call pool stack.

4:58 So there's two architectures is my theory. One for doing a bunch of advertisers at a time, it's best for small advertisers and one for doing one advertiser at a time, that's best for big advertisers. And I'm not saying they're incompatible. That the same company can't have both, but you can't use the same tools for both. Like the same company can house both, but it's like having, you know, a record department and a video department or something.

5:22 It's, you know, it's two different media. It's Two different architectures is what I'm saying. You cannot build one tool that does both of those. And saying I really want to educate the market about not only because it's underscores the value of Chalice, but because it's true. We have a lot of evidence of it and it's something that big brands should, should really understand. There's easy pieces of it. Like if you have your own data then you don't need a pool that should be obvious.

5:48 And then there's a lot less, less obvious things that are uncovered which is what happens to bidding when everyone is using a pooled stack and there's a lot of people in a category that end up all bidding on the same thing. What happens to the landscape? And that one's harder to explain because people tend to think about buying media in terms of buying packages of media instead of buying individual impressions, which is the way most digital media works today.

6:14 The actual, the way it should work. Yeah. And I really appreciate the thesis because is very striking to me that we think about advertising systems as monoliths and like there is the Google monolith and a lot of people think about it that way. And in reality Google services so many different types of customers with so many different types of needs and there are probably so many people who are not well served by the monolith and we haven't done a good job of stratifying it.

6:45 And this is I think a really solid, well thought through attempt to stratify the advertising market. This is like just the beginning. Like the only place I've really seen it is health because healthcare has such unique requirements around targeting and disclosure. I've seen like a few companies really stratify. Oh we're a healthcare media buying platform. Outside of that it's nothing. It's one size fits all, pooled solutions everywhere. And that's wild.

7:14 Yeah, well I mean you could see in creative like most, most brands will do, make pretty generic creative and a few brands step out and make really striking creative and everyone can see that's better. But one way to say this message is actually the same thing is true of media buying. Like if, if you're big and you have the ability to megaphone to say something unique, it's clearly better. And if you're big and you use your data to buy media in a unique way with unique value to cost estimations, it's also going to be better.

7:45 And of course if you could do both at the same time, you're going to grow like one of, one of the mysteries of business is how much growth has slowed down and how much brand equity has eroded since the traditional media era. And I'm hoping my piece goes some way to explaining that. But thanks a lot for reading it. I've gotten some great engagement and it's certainly been passed around places I wanted to be passed around without people telling me what they thought.

8:10 So nice.

8:11 Some private marketing that we'll have to figure out how it helped on the other side, if it helped.

8:17 Yeah, you know, attribution's tough. You have to build an mmm.

8:23 Yeah, yeah.

8:24 So I have a few theses that I'll preview here that people have heard me talk about before, but I really am passionate about them because we're at a weird inflection point now where this is so interesting. Actually. It almost mirrors our political system in that there is shaken faith in Programmatic in a bunch of circles. And I'm really seeing it in publisher circles. Publisher circles are really talking about direct sales and agentic direct sales as being the future.

8:59 And I'm seeing people talking about undoing the errors of Programmatic and reigning it back in, taking control back from Programmatic. And while I find them disturbing, I feel a need to explain them and examine Programmatic Programmatics failures because it does have some. And I spent some time thinking about this and people have heard me say a few of these things before in passing, but I really want to litigate them a little bit.

9:30 And the first one that I want to talk about that I threw into chat is the concept of trust in Programmatic. And there is a lot of trust. I remember when we first started talking, Adam, you were like, gareth, what's this field? And what's this field? What is this placement id? What is this? And I was like, oh, that's something that's just made up by either the intermediary or the publisher and that gets passed to you as an optimizable signal.

9:59 And there is no policing, there are no standards, there is nothing. This is true of IB categories as they're defined in most systems. This is true of placement identification. This is true of the vast majority of the data that we have about the inventory itself. That isn't the URL, it is seller defined. Hell, there is something called seller defined audiences where the idea behind seller defined audiences is that publishers are going to publish segments for people to buy against.

10:33 And don't get me wrong, I think that there is utility in some publisher first party data. The AI wouldn't be crawling publisher websites if there weren't insights it could glean from the content of those websites what users read has some sort of bearing on the value of the user. For certain types of advertisers, that is certainly a truth. I believe that however, when the architecture says that anyone can make up whatever they want and put it in these fields, the notion that these fields will have any utility whatsoever is a troubled one.

11:08 And. Right. Like if they. Because at the end of the day the buying systems are algorithms and that means that they're gamble. And so if you realize, oh, I can manipulate the contents of this field to manipulate bidding, not necessarily to manipulate outcomes. Because as you were saying before, pooled bidding systems really trend towards buying certain things and they trend towards certain behaviors in on mass. Which means if you as a publisher can figure out what it wants and you can give it to it, it will bid more.

11:40 It's crazy. Yeah.

11:43 I would just say that that that activity by publishers follows a failure of the market to produce reasonably high bids for a reasonable share of opportunities. Right. Like if from the publisher side it looks like for every 10 opportunities there's no bids on eight of them, some bids on the ninth and a ton of bids on the 10th, then I think it almost stands to reason that they want to get value for the others and end up pulling shenanigans.

12:16 And then I would just peel it back, back one layer and say that failure, the failure for value to be spread out across opportunities has to do with, you know, monopolies and, and just too much concentrated on the buy side that's causing too many people to bid on the same IDs.

12:32 I agree and I, I don't mean to make paint a moral picture here. I mean to paint a picture of. Show me the incentive and I'll show you the outcome.

12:40 Yeah.

12:41 And the incentive here for publishers is to game these fields because they will make more money that way.

12:48 Right.

12:48 And that's exactly right. And a, you cannot blame publishers for doing this. This is the prisoner's dilemma Pareto optimal outcome. This is what they have to do if they want to make money, because if they don't, other publishers will and they won't get those bids on those that 10% of users that, you know, the pooled monopolistic bidding systems have said that they want, they have to fight to get that money or they'll get no money at all.

13:16 And the answer here is, apart from moving away from these pooled bidding systems, is zero trust. So how do we move for these fields that are publisher defined to something that is defined by a Neutral third party and ad tech doesn't have a great history of neutral third parties. Like we're not really good at it. Now to be fair, I don't know what, what industry is again, neutral third parties.

13:43 Our, you know, our, our TV is governed by the iab. It's a neutral third party. And pre bid is a second third party that came out to govern header bidding and now also governs, you know, how, how requests are sent out. So you don't lack for neutral third parties. But there hasn't been much agreement in those bodies or much enforcement maybe more to the point.

14:06 Yeah, well, and they're not vendors, so they can define standards but they cannot do enforcement of the content of fields. So I think that in both of those scenarios now pre bid actually does a little bit of this architecturally. So pre bid, because of its nature, there are certain things that it detects itself. So because apart from being a third party, prebid is itself an auctioning technology. So there are certain things that it can actually do and enforce and be like these fields are here, I populate them.

14:37 Like sure, they could be manipulated, but you'd have to be a really bad actor to manipulate them. And the IAB has none of those teeth whatsoever. Right. All the IAB can do is publish standards and be like, hey, please adopt them. So when I think of neutral third parties, I think of verification vendors. I feel like DV and IIS are kind of supposed to be neutral third parties. Like kind of their job supposedly. Right. Like, I don't know whose interest I mean, other than their own.

15:06 But they're, they're, they're supposed to be neutral. Is that, would you say that's fair? I feel like DV and IS are supposed to be neutral at least. Yeah, theoretically. Right?

15:14 Yeah, they are between buyers and sellers. Yeah, of course, yeah. They're both their customers.

15:20 And so my thesis here is that I think DV and IS are zero trust. Right. As a very, as a verification vendor, DV and IIS do not necessarily trust what is told to them in a bid request. It is their job to sniff at it, see if it's legit or not, see if it's real traffic or not. And then if they're running one of their blocking or their detection tags, they serve to the page and it's their job to try and figure out what's really going on.

15:46 Right. So the process have no enforcement mechanism. They don't, they don't even tell the buyer what they found.

15:52 I mean that's, that's its Own sort of, sort of comic. I don't understand it still.

15:59 Well, I guess there's an incentive issue. I mean the cynical person would say, you know, if they cleaned up the problem, no one would pay them anymore.

16:08 I've heard that. I mean this is similar to like when Augustine Fu posts, right? Augustine Fu does all of this research. He has these tags that detect all of this fraud. But he doesn't do any prevention. He purely does reporting. And if you were to ask him to do prevention, he'd be like, that's not my business. And it's like, well that's a great way to stay in business. I do nothing but find fraud. I don't prevent it.

16:32 So on this one though, I think that there is this concept of objective neutral third party on the buy side and I think that we need it for a few of the fields that we have going into the bidstream on the sell side where it's like who is the neutral third party who is describing the Internet as it stands today? And I mean you've seen it, right? Chalice uses URL content signals. I would argue that is a lovely objective third party neutral way to assess the quality of sites the way that crawlers do it.

17:02 And we just have a few more in addition to a crawler, right. We have signals flow that flow in on the page that replace a bunch of. Well, not replace but supplement a bunch of publisher defined signals signals. And the really interesting thing about this is that most publishers welcome this and the reason most publishers welcome this is most publishers do not have the horsepower to play the game of how do I manipulate these things on my page to maximize my revenue.

17:29 Because it takes real horsepower to do that. It takes engineering resources, it takes time, it takes effort to be like how do I manipulate these signals to maximize my yield?

17:38 Give an example of a signal that you manipulate to maximize yield.

17:42 GPID. Yeah, GPIDs are one of them to the people. It's a global placement id. It was invented by the trade desk trying so in trade desk defense they were trying to standardize this in the open sourcing pre bid to build a standard way to describe individual advertising placements.

18:03 Yeah.

18:04 And so they. I remember when Judd first introduced this he would there and you know they, they say it was to standardized placements. They, they, they built it around a Google identifier. So something called the Google Ad unit. So the Google Ad unit path and turns out the original use case was actually just to try and deduplicate traffic from adex because they wanted to stop buying from Adex create something that was like describing inventory. It was much more to as a, as a deduplicating tool.

18:36 But this is a publisher defined thing still, right? And in fact the original concept for this was rolled out by someone who was on my product team at night. It was called PBAD Slot. And the notion was that publishers should use this to accurately define their inventory. Because you can't accurately define your inventory with your other objects, like things you have to build in Google, like things like that. There are all sorts of problems and my reflection here has become this should be a zero trust object.

19:03 This should be an object that is defined by a neutral third party the way that inventory quality data is defined by a crawler. This should be something where someone says, okay, we're going to create a objective works the same way on every website. No human intervention identifier for inventory on a website.

19:22 Now a lot of efforts in this area have failed because publishers reuse those IDs. They have to be specific to a page as well. So you've worked on this problem.

19:34 Oh, this problem is solved. So yes, they don't necessarily have to be page specific because if you make them page specific, you fall into the trap of not having enough volume. So if you are actively optimizing to them. Some pages are flashes in the pan. What if you haven't seen very much traffic on this page Page, you still want some way to understand it. So we have tricks, right? This is, this is one of Gamera's core piece of IP is how we do this.

20:05 How do we build a standardized way to label inventory on the Internet?

20:10 And like a piece of the I.D. that's the position. Is there a piece that's a page and a piece that's the combination?

20:16 Yeah, there's a little bit more than that. But yes, you're smart to think about it in pieces because that's what it is. There are different components, different values based on a bunch of different attributes of the ad itself. And by the way, page is not enough. Different pages on different screen sizes behave very differently. And that is one of the hard things about a crawler. Someone who's running a crawler. You can't crawl every screen size. That's nuts.

20:48 It's going to cost you a bajillion dollars.

20:50 Also, some pages render differently for, you know, for other reasons besides the device.

20:57 You got it. Different users get different experiences. Publishers vary these things. Yeah, I mean, don't get me wrong, crawlers are really much, much, much, much, much, much, much better than nothing like so much better than nothing. It's indescribable. However, you know that that last 20 to 30% you can capture when you're on page. And, and that's. So that's what we're working to do on our on page signals is create a zero trust environment. So create an environment where the signals are there that uses the exact same method on every publisher.

21:29 We're not getting greased by anybody. We don't make more money when the signals are better or worse. And also really importantly, we are like hyper reporting capabilities to publishers on these. So publishers can track them in their ad server. Publishers can visualize what we're seeing for different placements using our ui. It's a, it's something that actually was built from the perspective of the publisher, not from the perspective of the buyer. To the point where we even have publishers publisher tools for them to use these for Things like building PMPs or direct sales if they want to.

22:00 Yeah.

22:02 From the publisher side about curation, my, my thought about curation was that it was, you know, like a pub LED solution to the problem of clustering of bids by ID is like that publishers and some other people were like, well there's other value here that's not being represented. I know the middle took all the value, but oh, it is.

22:25 So I, my thesis on curation and I wrote about this when it was first coming out and I still strongly believe this is that curation was one an invention by App Nexus where an enterprising trader realized I can use App Nexus's UI to optimize a campaign while sitting on the, on the sell side rather than from the D from inside of the dsp.

22:56 Sure.

22:56 The reason this works so well in App Nexus is because App Nexus, the way App Nexus was built was not like, oh, there's a DSP over here and there's an SSP over here. They were like, until they broke them apart. It was one system and the buy side of their ad server was also a dsp and it did both things at once. You could target your own inventory and external inventory at the same time. And they realized, oh, we can slap that targeting tool onto a PMP system.

23:25 And now a trader can go and say, hey, buy my pmp instead of saying, hey, I'll come work for you and run a campaign for $100,000 a year.

23:35 Yeah, but why didn't it work from the publisher view? I mean from the buyer view it worked.

23:41 So that, well, that was all that mattered. So the publishers, this was invisible to publishers. Publishers are still scared and skeptical of curation. And I just think that it's, it's the reason they're so skeptical. The reason it's so strange to them is because it's just not a publisher product. It is a product that is exclusively sold to buyers where it's mostly ad operations. People like you guys do real things with curation, right? You use containers.

24:08 But even the base curation is selection on inventory. So it is more publisher friendly because we're actually selecting publisher qualities, not just IDs.

24:18 DSPs supposedly do that. They have targeting by, you know, seller and domain and all that stuff every dsp.

24:25 So why did curation work then? Why did, why does curation make performance better if they already do that selection?

24:30 Because some traders realized. So I, my, my thesis here is that most hands on keyboard people, and I don't want to say that many hands on keyboard people are overwhelmed and they can only devote so much time and energy to each campaign that they're said the DSP does, doesn't do it by default. I don't know why it doesn't. By the way. This is still a mystery to me because I, I know people who have huge curation businesses and all that they do is like set up a bundle ID whitelist and the buyer goes oh, this works great.

25:04 And I'm like how, how does this work? What is going on here? You mean the DSP doesn't have a bundle ID whitelist in it that it can just buy against and it performs well? Like why, why does this work? I don't understand.

25:15 I would. I think curation works because it's more because it's per advertiser. Just like I was saying about two architectures. If a DSP did it, everyone would be on that white list and it drive the price right up and it wouldn't work.

25:25 That could very well be it. It could be as simple as the D you are optimizing per advertiser. It's an ad ops person who is building a list and reporting and functionality that's specific to that campaign and that will always beat the DSP algorithm just trying to hit go. That is a perfectly possible valid explanation for why curation would work because it's very confusing to me. I know how a lot of this, a lot of these businesses work.

25:51 What a publisher see, right? So publishers now have their domain selected, but they don't make any rhyme or reason of it. They don't see any money out of it.

25:58 They can run a report. Here's how much cure here's how much revenue came from curators in the past 30 days. So some exchanges have that reporting capability, but most publishers, it's just a mystery to them. The only place there's a really big opportunity for some publishers is publishers who have direct sales teams can become curators themselves. And that is something I am seeing more and more of where by virtue of having the direct sales team who is able now to pitch that per campaign thing to advertisers, the publishers become curators.

26:31 But that's only the biggest.

26:32 Right.

26:32 That's only like the. The ones who have sales teams, which there are not that many of.

26:37 Okay. So they didn't see, you know, could they. Why didn't they see it as net new revenue?

26:47 So in their minds, and they're right about this, most curators are going to take as much margin as they possibly can before the campaign starts working. So to a publisher, the net new there because the publisher has no transparency into this, their assumption is that most of the net new revenue gets eaten by the curator. And not for every curator, but for many curators. I think the publishers are right about that, where whatever incremental the curator can create, they're going to ease their margin up as high as they can to capture as much of it as possible.

27:24 Yeah. When AppNexus launched Curate, AppNexus takes a fee and it's set up so the curator can also take a fee. There might be an agency fee. You could really stack them up. So there's rebate systems now.

27:37 Yeah, they're literally in there. There's a rebate.

27:42 It's like the machinery to take spreads is built in.

27:45 Yeah. So I think that that's where a lot of the publisher skepticism comes from, where this was billed as something that was empowering publishers. And I think that that was a little bit of a disservice. It was, hey, we can use publisher data to make things perform better. I don't think that that's the reality of what curation is. I think the reality of what curation is what we were talking about. It's adopts people doing per campaign optimization that works better than pooled ds.

28:11 Right. But now Gammarate really is that value proposition of using publisher data too?

28:15 Correct. And here's the thing. Curation is a really nice layer to insert it because legacy DSPs do not want to ingest it. And I can. This is a tree that I have barked up for years. They just do not have the incentive. They do not have the horses. They don't live a life of trying to make everything work better all the time. They don't. They're not continuous improvers. Right. And there's only so many hours in the day and so many things they can work on.

28:46 And they're out there trying to conquest business from one another and, and keep their clients from in housing.

28:52 But the specific improvement you're talking about is advertiser performance, right? I mean, you're talking about publisher performance,

28:59 but no, you would think so you would think that they would really get out of bed for this. But only recently have I seen advertisers and people really starting to care about performance. And the, the lip service paid to outcomes aside, the assumption on the sell side for a long time prior to that was advertisers don't actually care about performance. They care about low CPMs and reach. And because there are no performance goals, there's no real incentive to use.

29:29 I mean, trade desk, you know, trade Jeff always talks about, he cares about the ecosystem and trade. Std7 has invested in both our companies. So that is a clear, you know, putting money down to make the open Internet better. But if advertisers don't care about performance, then, you know, who's the. How do you, how do you win? How does camera win?

29:51 This is a great question. If advertisers don't care about performance, I don't win. Or the way that I win is I go out and I find advertisers who do. And maybe that's not existing programmatic advertisers. That's a thought that I have all the time. It is. If enough of the market truly doesn't care about performance, then I have to go find another market. And the other market definitely exists because online advertising for performance is one of the largest businesses in the world.

30:20 It's just a matter of. I know ad tech and ad tech has all these lovely pipes to surface this data. The curation pipes are really nice ways to surface this data, this data, but it is hard to say which advertisers will care, which ones will not. So if you're listening to this and you're an advertiser who cares, you should shoot me or add them an email and you can talk about it.

30:39 Our audience is a bunch of advertisers and publishers who care. We're, we're gradually winning in the marketplace. There's a big shakeout going on and you know, a lot of us will be left standing when, after the, after some rough times ahead.

30:55 I feel that way too. I think I see it too And I think I see agencies talking about performance and talking about these things more than I've ever seen them talk about it. And one way to think about it is performance does provide them with an opportunity. Because all of the gross parts of principle based buying aside, I prefer that to I want this user ID at the cheapest CPM possible. I like that. More than that, at least there's a CPA where it's like, okay, yeah, I'm going to, I'm going to try and scoop up credit for conversions.

31:25 Who knows if they're incremental or not, but I'm hitting a number and I want to hit that number. That's, that's something. That's an optimization process at the very least.

31:35 Yeah. The 2012 solutions have run their course like they're, you could still use them and they'll still work. But for people who want something better and who've been paying attention since 2012, like it's. That can't be it.

31:47 It can't be it, man. And it's actually a really interesting segue into something I spent some time thinking about. So this is like a half baked thought, but I think it's worth talking about, oh, did I move? Am I better now? There you go. So my half baked thought on this is that there's a distinction between data and optimization and sometimes that gets blurred by some companies and measurement.

32:19 And measurement.

32:20 Yeah.

32:21 And I like attention kind of. But I'm going to go after attention for this. And the reason I'm going to go after attention is because I think attention is actually two businesses. I think that one business is a measurement business where they're like, these are the things that we measured and as a part of your campaign. But the other half of the attention business is an optimization business, which is we have a data set of conversions or of outcomes or of something.

32:56 And in the objective data that we measured, these are the things that led to outcomes.

33:04 Yeah, I don't think anyone does that except challenge. I'm going to cut you off right there. Oh really?

33:07 No one does that.

33:08 I think we're the only ones who do that. Well, we'll take the metrics. Amazing. The empirical metrics. This is what our Hyundai case study is about. Taking the empirical metrics of pages like how many ad loads out of view, how many network calls, the ad to content ratio the refresh rate and score those against an outcome and then find the optimal prices for all of that. Every, all those features and combinations of features.

33:35 So it works like, oh, I, I mean I'm sure it does. But you guys are doing that right? On a valuation basis, my understanding is that the account.

33:47 Yeah, cost for value.

33:48 Yeah, cost per value, like really robust for a given advertiser, et cetera, et cetera.

33:54 It took all six years. Like containerization and RTF had to be built, tested and deployed before we could actually really do it.

34:02 So the attention that I'm talking about is you have an attention measurement tag and you have what it does right time in view, like maybe something for ad density, I don't know. But all you have is the measurement and the outcome. And when you're deriving an attention score, you use the measurement and the outcomes to inform the attention score to be like, what components of this created attention? Because in order for attention vendors to sell, what they have to sell is that attention correlates to outcomes.

34:33 Because if attention doesn't correlate to outcomes, what is attention worth? It's not worth anything.

34:38 Yeah, what you said isn't different. The score is correlated to the outcome. Then it's not a consistent score. Like, all you can do is break it down into its components. Like you can't.

34:48 Like, I think attention is, is another form of pooled algorithm.

34:51 Yeah, of course.

34:54 And it's like, it's a similar to like a pooled solution where it's like, oh, these are the things that make good attention, but the reality is the things underneath them are actually just a half optimization process. Because it's only half. Right. Like, there's no bidding. There's no like, evaluation of other opportunities. There's, there's, there's none of that in attention costs.

35:14 Right. Like cost per, you want to, if you, if you do, even if you do have a stable definition of attention, you'd want to optimize cost per attention, which you run into a problem as soon as you aggregate a bunch of customers under the same attention. Like, you're making it more expensive.

35:29 You got it. Plus, if the, if the attention data is pooled, in theory, each advertiser should have their own attention data set because these are the things that correlate to outcomes for you. And this is why I draw the distinction between an optimization product and a measurement product. The measurement product should be responsible for getting all of those parameters to the optimization company. And the optimization company is then responsible for ingesting them, correlating them, doing cost of value analysis.

35:57 And I think that the attention companies are a little bit of a blend. It's a little bit of both that they do where they. And because of that they are not as efficient as they probably could be if they were doing one of them.

36:12 Yeah, I like attention companies, but I don't. I don't meet a lot of clients who are interested in it as a metric or a signal. And when I do, it often feels like we're stuck on a proxy that we don't have to be stuck on.

36:28 It's really interesting. Well, because the. And I think, like, I feel pretty strongly about what I said, which is the only way that it's valuable is if you can prove that it correlates to something.

36:37 Yeah.

36:38 And if you're proving that, why not just correlate the things that created attention to the thing? You don't need that part of it.

36:49 If you have like a sexy model, you get a lot of attention. So it's like, what do you do with that problem? It doesn't necessarily help the brand. Victoria's Secret ads get more attention than Black and Decker. But.

37:00 Well, I'm. Now they got to do the funny little optimization thing, right? Where they got to be like, oh, well, this is bad attention. This is good attention. Because without the outcomes, you can't discern. And yeah, so I think per advertiser, right.

37:13 It should be per creative. Like some. Some creative works in two seconds. Like if you have a creative that works in six seconds, then you need six seconds of attention.

37:21 Plus, I feel like this is a weird thing where now if you plug an attention signal into a custom algorithm vendor, which I'm actually. Have you ever been asked to do that? Has anyone ever been. Oh, really? Optimize to attention. There's a little noise in there because now you have something that's doing math over here to calculate attention. Something that's doing math over here to basically predict the math of this thing. It seems noisy to me, and this is one that I.

37:49 I hang my hat on, where I actually think attention is really better than viewability as a goal. And I don't mind it if people really insist on having something better than that for these things. But I think that they eventually need to either be a measurement company or be an optimization company. Pick one. Like it's very. And if you're going to do both, do it the way DV does, where DV has an optimization side of the company and a measurement side of the company.

38:19 Right.

38:19 There's cybids and there's dv.

38:21 How's that going? I have a lot riding on that.

38:24 In a good way or a bad way? I don't know, man. I was on a Panel with someone from Cybids once. That's the only time I've heard of Cybids. That's the only time other than when it was, when it was acquired.

38:38 Again, what I know now about the cost landscape is like you must not use any kind of pool detention solution. Like right, the best, the best value to cost ratio is it can only be achieved by looking at things, looking at signals other people don't look at.

38:54 That makes perfect sense to me. Or looking at the signals and deriving your own cost to value ratio. That makes perfect sense to me too. And one of the things is people get real haughty about viewability. And someone commented me on LinkedIn, viewability doesn't matter. And I was like well it matters. If it's average 10% that matters. That means that it's probably not viewable. So like we'll just start there. You don't want to buy things that aren't viewable.

39:19 That makes sense to me. But you haven't won the war by buying something that is viewable. You have simply entered table stakes. Right?

39:27 Like, well the hard part is splitting all the viewable inventory into the pieces that are worth it and versus the ones that are gamed.

39:34 That's, that's should be what the game is like. Like the, the notion that we have to.

39:39 The non viewable stuff is cheap.

39:42 Oh God, that's cheap. Cheap CPMs for user IDs non viewable inventory is great. God.

39:50 But a lot of the DV viewability metrics are misleading because they, they penalize everything they can't read and there's a lot they can't read. So it might be perfectly good.

39:59 That's why they don't like to give logs. As someone who has a tag that's measuring some of this stuff right now in market, I can tell you that it's the wild ass west out there. Also like the metrics that you get from DSPs, right? So whenever you have a measurement tag deployed it has little macros in it, creative macros that populate with data. Sometimes I can't get URL, like the DSP can't tell me the URL that they bought.

40:23 I can't reach out of the iframe to figure out what URL I'm on unless it's someone who's in my network. And like where did that impression go?

40:31 Could be anything.

40:33 Could be anything. It's crazy. Like I have JavaScript running in there to figure out where it was and I don't know.

40:38 So like on the Pub side, you don't know where it showed up.

40:41 Only. Only if I have my JavaScript on the publisher, then we can close this loop and the publisher can get proper credit for what they sold. But without that, like what about the

40:50 pub side ad server that has every URL. Everything that shut on every URL, right?

40:54 Oh, they don't pass. Well, a, there's no. In their logs. They would. But you can't report on URL and publisher ad servers. It's not a thing. And two, like the ad server doesn't talk to the programmatic pipes. They don't talk to each other unless it's going through ad.

41:08 You don't know what it went for. Right. There's no price on it. There's no timestamp.

41:11 God, you got nothing, man. It's this. These systems do not speak to one another. This is why having objective ways to define a lot of the data that's used as the currency in a lot of these things. And I used the C word. I don't like the C word. There's only one currency. It's cost per thousand impressions. That's. That is the currency that we operate in. I don't see CPM going away anytime soon.

41:33 Nielsen ratings. There's a lot of buys against Nielsen ratings in digital, in CTV though.

41:38 But so they. What are they paying though? What is the. What is the money that transacts? So you pay a dollar per Nielsen

41:47 rating, like per Nielsen demo.

41:51 Interesting.

41:52 That's how make goods work. Like if the, if the show didn't get the ratings it was supposed to get, then you get a make good on those. On all those other eyeballs. And a lot of times they pay

42:00 it off in digital and they rolled that off in CTV really well. Nielsen, good, Good job, Nielsen. I give him credit for that, man.

42:12 Like them as standards. Like, it's just like, you know, we talk a lot, we're encouraged. Things are changing. Advertisers are getting smarter. But the entrenchment of those guys as the standard bearers is not. That hasn't budged. And it's a real. It's definitely a barrier to progress.

42:27 It's really impressive. That man like they, they really did became. They. They became the way that things are defined. Not just measured, but defined. They are the market.

42:42 And everything has an interest. Every incumbent has an interest in keeping them there because they've built around them and you know, they pay them a lot of money. So it's like they're like, yeah, like no matter if the members of the Cartel can change and they'll have the same interest in supporting DV and Nielsen to hold up the standard and make it harder for things to change.

43:02 That's crazy.

43:04 Our market is. Starts to feel really sick when you look hard at those.

43:09 Well, it's like people look back for the glory days and it's like. Well, the, the rules of the game were simply defined at the very least. In this, in this. Right? The rules of the game are simply defined. Nielsen says this is worth this.

43:24 And Prime Demo probably made a lot of sense when they, when they created it.

43:30 Yeah, I'm sure it did. I'm sure it was a tremendous innovation. But yeah, man, I think objectivity and signal and then sophisticated systems that interpret those signals is the future.

43:43 What's your business model?

43:45 Yeah, so the business model, I mean this, this is something that's constantly evolving because we're a startup, but we're hard in the data licensing business now. It's really my focus. I, I like the data licensing business and I like data licensing and measurement. There's. That's really direct. The direction we've gone. The only time that we do anything else is normally if a DSP wants us to do something with them and they need a way to transact.

44:10 That's. That's not that but we are, I am, I would say 75% of my conversations now are data licensing conversations. It's, it's. How do we get this into. How do we not become a currency but become at the very least the way that these things are discussed?

44:29 Do you, do you rev share with publishers or do they enjoy the benefits of.

44:33 Of better data Pubs enjoy the benefits of better data. We have a free product for publishers. So every publisher, they get tons of reporting from us. We have paid tools they can use as well. But this is one that gets me like every publisher should install this. I know. I cannot believe that publishers would consider not installing this. It's crazy.

44:52 And they installed the free tool, what do they get?

44:54 They get super granular reporting on how all of these metrics. Right. So like you were talking about earlier, right? Like refresh rate, add to content ratio, ad density, work down to the placement level on every single page and every screen size in their website and they get active monitoring. So not only do we tell you a page has a problem, we tell you we observed this problem on 6000 impressions that rendered yesterday. And like this is, you need to go look at this and make this better.

45:23 And then buyers will buy more from you.

45:26 You help them understand what CPMs they're getting associated with all those things and how those differ by ssp.

45:32 So we have that data, we don't push it into the UI right now. But I'm starting to think more and more and more, like I've been having. A lot of publishers are asking for brand safety stuff, by the way. So this is a new, new axe to grind. Publishers keep getting deprioritized for brand safety reasons and they don't know. And there are apparently basically no tools for them to install that actively monitor and say, hey, you have a brand safety violation.

45:58 Like, apparently that's not a thing. And how is that not a thing?

46:03 Well, there was this company called Moat and then Oracle decided they would deprecate it. No one kn. Ever since.

46:10 It's awful. Like news publishers, especially because the news publishers, right, they can't keep an eye on every single article that gets published across 60 local news stations. And all of a sudden they hear from a DSP like, oh, we ran on this page that, you know, had this thing that the buyer didn't like. We're not buying from any of your sites anymore. And the publishers are just like, what? Like there's no warning. Like, like we didn't do it on purpose.

46:36 Like, yeah, just let us know.

46:37 Keyword they didn't like. And. And don't be wrong. Marley, Marley, you chime right on in. It is complex. It is complex because, you know, if I'm a brand, I. There's a. There's a lot of Internet out there and if I can avoid taking any risk at all, I would prefer to do that. But at the same time, I think that there just needs to be proactive reporting for publishers. That's all that they want, right? Just tell them, hey, this one might be a problem.

47:07 Like, think about that. Maybe edit that word out. Or if you can't edit that word out, I don't know, label it somehow quarantine it. Some of the more sophisticated publishers I've talked about it, yeah, like, hey, this is a sensitive one.

47:19 All the language can be processed and labeled.

47:23 It really can. The robots are really good at this. This is one. Every time a publisher talks about this, I'm like, you know, the robots to crush this. There's one thing that they're good at,

47:33 as we're finding out. It's pretty expensive, but I mean, for us it's not exp. Not expensive for dv. You think they would have done it?

47:42 I. You think they would have done lots of things, but they. Yeah, you know, man, they're Fat businesses, they've been running a long time. They're not trying to shake the boat. I don't. I get it. But yes, I think that we're moving in a nice direction on all of these. I think objective data fed to proper algorithms is. Was the original promise of Programmatic and I think that we missed it. Right. For 15 years. It's just not what we've been doing.

48:08 And I think that's why, to bring everything full circle, that's why we have this huge turn away from Programmatic is because this initial promise of we're going to atomize all of these things and apply algorithms to them to do bidding and valuation, that's super accurate. That leads to great outcomes. That shit just didn't occur like it. And because of that, publishers are like, well, wait, now I'm just making less money. And this is not what I was told was going to happen.

48:37 Yeah.

48:38 And there is a reversal that's occurring. Right. What Chalice is doing will inevitably lead with. If Chalice succeeds will inevitably lead to a reversal in that fortune. Because there is a tight feedback loop for publishers as long as they're losing something like Camara on. These are the things that have lead to good outcomes for buyers that are a priority parts of running your website. Do them more and you will make more money because more and more buyers want to bid on your website.

49:13 Yeah. It feels, from my experience there's like better opportunity to work together with publishers on that path in the UK than in the us I don't know why. Maybe just because it's smaller and we're closer together. But that's so interesting because newspapers are still powerful. Maybe.

49:27 I have tremendous adoption in the uk. They use my direct sales products. They use all of these things. And the US is lagging, which is really interesting to me. I work with, don't get me wrong, I work some really big publishers in the us but the UK are fast adopters of this.

49:42 I don't know, newspapers and magazines just got crushed here by Google. And I think in the UK they did a little better. It survived.

49:50 Yeah. I mean, they have nice big sales teams. They're big companies. Yeah. I love my UK news pubs. I work with a bunch of them and I love them and they, they get this. So, yeah, okay, we'll do some stuff in the uk. This. I, I really think that once you, Once we, Once we demonstrate, right. Like, hey, this is the future and your revenue is going up because of this. It should fall like dominoes where every brand, every big brand at Least realizes they need to do this.

50:22 Somebody does something about the SMB market, we'll see one day maybe.

50:26 Yeah.

50:27 And the agencies will have to jump on board somehow, like at some point. Because if they're, if they keep losing to in housing, they're gonna have to, they're gonna have to think of a reason. Well, this is why we're better at this than in housing is they're gonna have to come up with a solution.

50:41 I hear they're working on that, I guess. Let's go to news.

50:44 Let's go news.

50:50 Emptying my pockets. Yes. Just talking about agencies, the Atomicom's earnings. They got a question from an analyst who asked how the improvements for AI are going to show up in Omnicom. Is it savings or are they returning the AI savings to advertisers who will now see more for their money? And CEO John Wren kind of punted on the question.

51:23 Didn't answer either of them. And to me, this was an example of how they've kind of kicked the can down the road as far as the value of AI. And the question indicates that the bill is starting to come due. Like they've been telling the market for many years that they're investing heavily in AI and developing in house AI tools that are going to bring tremendous value to their clients and the agency. And you know, they're starting being to be asked to show this.

51:53 It's so dangerous. Right. This reminds me of Sam Altman talking about being a hundred billion dollar advertising business.

51:59 Yeah. It's like when you say something like that, it's great for a while. Everyone celebrates you for a week or maybe for a year.

52:08 Maybe for a year. Yeah. I think he got a full year out of it.

52:11 And, and if you're elon, you could, you could keep the plate spinning indefinitely. Amazing. He's.

52:17 He's a master, right. At stringing things together and because since he has like three or four different projects, all he has to do is like bring a new shiny object and be like, well, look over here. This is the one that matters.

52:30 Yeah. My friend Nick said that that's the definition of a Ponzi scheme. I was like, only when it fails does it become unconscious.

52:43 I think that that's.

52:44 While there's a chance of everybody getting paid out. It's just investments.

52:47 It's just investment that looks like it's performing well, by the way. Yeah. So I think that the agencies are really, they're running into this because they've been shouting about AI for a long time. And I am very sympathetic to them because all of the AI startups in ad tech, Right. That aren't like of a previous generation are pivoting left, right and center because they're trying to figure out how do we apply AI to ad tech. And it's not trivial.

53:13 And AI is moving so fast. It's really hard to build a business on top of something else that's moving so quickly. And I think the market wouldn't have liked to hear we're going to wait and see. But the answer was probably we're going to wait and see. AI is moving so quickly. We want to, we want to be really thoughtful about it. And there's a chance that anything we build today will just be like totally worthless in six months when they're like, you know, some new thing comes out from anthropic and it's a machine and it just makes it happen.

53:53 Yeah, tough spot.

53:54 Yeah, it's a, I mean it's a six year bill. The agencies are talking about things that are five, six year builds and they've never given an update of what year of the build they're into. Right. They should be pretty far along in these product builds but they are multi year builds. They haven't told the market, they haven't shared much of a timeline with the market and I think at least some of them are not really building internally any tools.

54:25 I find your lack of faith disturbing.

54:28 Gonna rely, you know, relying on Google, Microsoft and some others to give them something that looks unique but you know, just won't be, you know, won't be custom built for the agency or its clients.

54:42 Yeah, I mean I definitely think a lot about, there's a whole strain of companies out there that are like model management companies. Right. They're, they're layers that sit on top of all the frontier models and everything else and dynamically allocate things within them. There's got to be something like that an agency could do.

54:59 That's what they're hoping. Yeah, that there's going to be something. I mean that's what I think cynically an agency exec would say to Chalice is like, oh, you've done this five year build but Claude's going to be able to do the same thing next year. And like, let's see, like I don't think, I don't think that's a very knowledgeable standpoint because the number of parameters we're dealing with and the execution pieces of it are way beyond anything Claude does.

55:21 We're just in a different family problems

55:23 It's a different family of problems. Claude and LLM. Yeah, Claude's an LLM.

55:27 You can write code too. Yeah, but it's like, yeah, yeah, sure.

55:32 It's not a data scientist. Right. It's not a, it's not going to build, build your data pipelines in a reliable way. Yeah, yeah, I, I'm with you. It's a long build for Chalice.

55:46 Yeah.

55:48 Next. Oh, our friend Ian Whitaker. I love this guy. He's talking about the Meta and Microsoft earnings calls. They both have enormous capex investments. Meta stock went down and Microsoft went up. And Ian here is pointing out in his post that the difference as the market reads it, is Microsoft sells compute. So every time they invest in Compute, it's like, hey, we bought more widgets and we're going to sell the widgets.

56:18 And the market's like, yes, we understand that. And Meta is saying, we invested our 200 billion in compute, but we're not going to sell compute, we sell advertising results. And it is working great for Meta, but the market is sort of saying, we're not so sure about that. So their stock went down because it's not clear to the market how investing in COMPUTE continues to make Meta's advertising results better in a way that translates into more money for Meta.

56:47 And I think the market is here, the market is being rational. Right. The market is saying, we don't really understand what Zuckerberg is saying about this growing indefinitely and producing moat that nobody else in advertising can cross. So we're hedging our bets here. It's rational.

57:08 I mean, I was someone who was very sympathetic to Zuck's bet on the Metaverse. Like I was very sympathetic to it because I appreciate a CEO taking a big swing like that where if he had been right, it would have been a magnificent home run and he would have caught the wave of humanity's migration to augmented reality. It turns out he was wrong. And in theory, there may have been lower risk ways to test the thesis other than how much money he put into it.

57:41 Changing the company name.

57:43 Yeah, and there was, there's just the amount of investment that went into it where he must have built some model where it was like, well, we need to invest at least this much if we are going to be the ones that make it happen. Like they wanted to make fetch a thing. And you know what, if you're gonna make fetch a thing, they are going like, you got to put some money into making fetch a thing.

58:06 But it didn't happen. I, I don't think the market's are going to forget that anytime soon. So.

58:11 Anytime.

58:12 Anytime. Meta's making a big bet now. They're like, we're going to hold our breath on this one. Yeah.

58:21 So they're getting pressure from the market to just sell compute because now they've built so much capacity and sucks like things hasn't agreed to that yet.

58:30 Yeah. Oh that man.

58:33 I want to buy some Meta computer.

58:36 I just can't get over the scale of Meta's business.

58:40 Yeah. They built so many servers. They would be like the third biggest server company in the world if it wasn't all for themselves.

58:47 God, it's social media, man.

58:49 It's just a website. Social media.

58:51 It's just a website just to have.

58:53 Just to scroll.

58:55 Just to scroll and look at reels and influencers.

58:59 Oh my God. We had to talk about influencers some week. It's getting so weird.

59:06 If everyone's an influencer, is no one an influencer?

59:09 Plenty of people still who aren't. We have another news. I think we have one more. We didn't do open weight models. We could do that next week.

59:18 Yeah.

59:18 That's all the news I got.

59:19 This is so interesting.

59:21 Yeah. All right. I'm not gonna do it in a minute.

59:24 No, we definitely. It deserves its own whole discussion, man. Open weights are so interesting.

59:30 Let us know if you think the open weight controversy is interesting enough for us to unpack. Very shy.

59:36 Most people don't know what it is. To be fair. I like, I like didn't understand it super well until you brought it up and I started reading about it. So.

59:44 Yeah, open letter. Yeah, it's got. It's, you know, it's like Kaiju, like the. It's giants are fighting. So it's interesting to know while we're on the ground and have a chance of being stomped by a Mechagodzilla or whatever. Like when are they fun?

60:00 Yeah. Oh, I. I have. I have thoughts on it too. Well, we'll talk about it. It's. It's a really interesting one because there's also like a cried wolf scenario here where we might get bit by that like, like Sam and Dario crying wolf a few times. And then we oopsie into the singularity.

60:19 Or a law. Or a law that just outlaws, you know, everything we do.

60:24 Yeah. Possible can't be trusted.

60:27 It's rough.

60:29 Yeah, man. We'll see for next week. Thank you for joining, everybody.

60:32 Yeah. See you next week. Have a good week.

60:34 See you next week. Have a wonderful week. We'll talk to you soon.

60:46 Yo. The crowd is back. The lights are dim, front rows buzzing. It's about to begin. Ad tech. Ad talk to taking the stage. Adam.