Episode 41: Adam Heimlich on custom algorithms, we pour one out for Sizmek, and BonBon justifies its existence.

You

to the Architecture Podcast. I'm am Paro. We have an exciting episode today. We have Eric Franci, our usual co-host, along with Adam Heimlich from Chalice ai. Before we get started I wanna talk about our new justify your existence segment. So today we have Bon Bon, which is Elliot Easterlings new startup that many of you have probably heard of. We also have an exciting conversation about seismic. We're gonna pour one out for seismic. So let's just in with Adam. Adam, thank you for being here.

Yeah. Thanks for having me. I'm excited. I listen every week.

Alright, good to hear.

Amazing.

So as disclosure, I think aerium is an investor in Chalice. Is that right, Eric?

Confirmed.

gonna take it easy on him or are you gonna give, this is gonna be like a board meeting. You're gonna grill him.

Adam can take anything. I think we've seen his tweets and his LinkedIn

I prefer to argue.

Exactly.

He actually does.

Alright, let's, Hey, let's turn the tables. Eric, why'd you invest in Adam

yeah, absolutely. So we invested in Chalice, Adam. Was it 2019?

It was

did you do the seed? 2020. Yeah. Yeah. So, so it was, it was before the, the big wave of, of ai. But you know, Joe and I had, had sort of long had this thesis that one of the next big things would be applications of, of ai. So so it was, you know, had a, had a thesis on the, on the sector. Had known of and known Adam A. Little bit through his years of working in the space. So for us it's like, you know, we, we sort of take a, a, a thematic approach in, in sectors that we like. And then oftentimes you know, sort of not, not a rule but if we, if we know the team and, you know, feel great about the team, we'll invest super early just on, on that basis. So that was the reason.

Eric introduced us to most of the people who are still our angel, angel investors. Maybe half of them came from, from Eric and Joe Intros. So hugely valuable and appreciated.

That's great. So so what is jealous?

Chalice is custom algorithm software. So I had a 20 year agency career mostly working for large advertisers and I noticed they weren't really benefiting from the algorithms making decisions on their behalf. Right. They're usually doing a lot of manual work to kind of steer against that autopilot. 'cause the autopilot is driving to clicks and conversions and they care about things like lifetime value and incremental sales. So I. I knew that it's better if you can have the algorithm driving for your actual outcome, right? Custom algorithms have always been part of programmatic. AppNexus was built with an endpoint for custom algorithms. Beeswax was built with a great endpoint for custom algorithms and thank you Ari. That was buoyed my spirits when that came out and helped convince me that there should actually be a company that makes using these endpoints easier. So we've been in market three and a half years. We have scaled software. You could think of it as a data science accelerator. You can use any data you want. We don't mingle your data with other customers' data, and you could drive to any outcome you want. It's extremely effective. We, we thrive mostly on same store sales. We probably lose four out of five pitches, but the only reason people don't adopt. Chalice or software like it is, is cultural. It represents a different way of working. The person who might be sitting around making up bids has to stop making up bids. Learn a lot about why the machine chooses the bids it chooses.

So, okay. The lot to unpack there. It feels like you're kind of going against a trend, so you, all we hear about is P max and AI and automatic results. Is, is that a, is that accurate that you're going against that trend or are you trying to reinforce that trend? I.

Well, I think what's different about childless and maybe unique is that we, we foresaw that there would be more rebellion or. On the measurement, right? So a, the P max and most AI tools in advertising are very, very powerful. They're extremely predictive, but all they predict is clicks and correlated conversions. So it's like having an enormous gun pointed at yourself,

Well, hold up. Hold up. What does that mean? Don't most customers want clicks and correl? Wait. Correlated conversions. You mean like view through? Click through.

A correlated conversion just means that it proceeded. Yeah. That there was a view that proceeded a sale and, and usually that you get credit if it's the last so often that has, not only does that have to. With a brand growth, but it could be opposite. So something I learned as the head of a trading desk at a big TV agency is that those algorithms will learn to, to send digital exposures to people who saw the TV commercial, because that's what it's trying to do. Like get an exposure and then a convergence. So they're predicting that these people will convert and they will, but it's 'cause they're influenced by tv. So the algorithm has actually ends up having. Driving less and less incrementality over time. Something I actually saw happen in the years 20 16, 17, 18. And one thing that my head,

yeah. So my immediate reaction, like if I was A outside of our industry, and you said you were building a custom algorithm on top of a platform like the Trade Desk or beeswax or whomever, my immediate reaction would be, well, that platform has more data than you do, so how could you possibly be better at this when they have so much more scale? You're only working on behalf of one customer, not thousands.

Yeah, that's, I'm glad you said that. And that's exactly the thinking. we need to turn around. To, to really succeed, become a, as successful as I think we we can be. So that thinking comes out of the mentality of a, a scaled software solution pre-cloud, which says if you scale a software solution, it has to do the same thing with the same data every time without human intervention. To get the best possible result. And that's why everybody optimizes to clicks and conversions, and all the data that's used by those algorithms is the data in the platform, and it creates an arms race around data. As you said, whoever has the most data ends up having the most effective predictions of clicks and last touch conversions. So, If you're Uber and Uber's not my client, but I read their case study. If you're Uber and your business depends on driving incremental rides on the Uber app all that scaled software is a disaster for you. 'cause it's gonna drive rides. But everyone has the Uber app, so it's, it's going to take credit for all the rides that would've happened anyway. And if you actually wanna drive incremental rides on the Uber app, you have to do what Uber did. They licensed beeswax and they used bid models to actually model to incremental rides on the Uber app. And they drove incredible results, right? They had 800. This case study said 800% Lyfts in driving incremental Lyft incremental rides on the app. And then they found out the conditions where they could get incremental rides as the result of an ad. And it was sort of made sense. It was like a rainy night. And someone who uses Uber two to two to three times a month, something like that. So an infrequent user rainy night, show them an ad, you can drive an incremental ride. Awesome. Finding awesome case

Right. It, it was a good case study. I, I forgot about it, but thanks for reminding me. I'm so far out of beeswax at this point, but like you but is it all about just different attribution? Like, is that the, the fulcrum on which a custom algorithm achieves its goal? Is that you have

No. Outside the platform. Yeah. Data that exists outside the platform, whether, whether a ride is incremental requires some input from Uber. Right. There's often l t V will be in that model too, and Uber isn't gonna send its whole c r m to beeswax the way they can. They can give it give access to it, to chalice, knowing that we won't share it or use it for anything but their project. And there's tons of data sets that live outside, right? In C P G, there's inventory, data, what lives on the shelf? Awareness. There's, there's data from the surveys. So having data outside the platform is something as part of your flywheel effect, part of your algorithm only became possible with cloud. And I, I think that cloud is, is bringing a new era of applications that are just much more flexible and configurable. And let. Engineers and data scientists, like, like the Chalice team, be much more creative about how they stack microservices and reuse code to do a whole bunch of different things for different customers and still scale.

So the second kind of skeptical view on the custom algorithm business would be, well, won't they eventually build it? So won't, it won't the d s P platform. So we're really talking about DSPs, right? We'll get better and better and better and over time, like it'll eclipse the need.

Yeah. Well, it's, I guess it's possible is I, I gotta ask Ari, is that why you didn't come in as one of our angels?

No, I didn't come in for the third reason. I haven't gun to yet.

I rem. I remember our emails on this. This is gonna be fun.

A lot. So a lot of the ad tech exeters are, are angel investors and Charles, we're very proud of that. And I get tremendous mentorship and it really it's, it's a huge asset to be able to get advice and that all these people believe. So Ari's an exception and we'll, we'll talk about that

tables have turned like no big, a big read for not being an angel investor. Alright, anyway. No. So why won't the DSPs build

I'm curious. I'm curious. Well, it's very different. That's the first thing I'll say is it's very different. Like we see the DSPs, they have the scaled software mentality and they, even when they see us working, ask us to be more of the same every time. Right. they, they say like, this doesn't, this isn't our view of scalable. And we're like, we're plenty scalable. Like we, there isn't human intervention once it's up and running and the data is flowing, you know, we're refreshing every 15 minutes. It's different than the way you do things. And it's, maybe it's a little less predictable for you, but this is a new way of doing scalable software and they, I don't wanna say they don't get it, but it's clear that there's a big cultural difference between the scaled software engineers of A D S P and the custom algorithm software builders a chalice.

Okay, I'll, I'll take that answer. Okay. Now the third reason, third skeptical thing, this is the reason I didn't invest is I am skeptical, and I'll say it for the record, I'm still skeptical about whether you can create a big business. As a custom algorithm because ultimately the end client has to share their fees, has to, has to allocate fees to both the D S P platform and to your service, and there's sort of a hesitancy. If you're paying the D S P 10%, why would I also pay you guys 10%, 15%?

Ah, yeah, that's, that's a great one. So, and we, the way we address that in our investor deck, which you didn't even sit for by the way, but that, that's okay. The the, our c o o Ali Manny made it, made an amazing slide.

know, for companies I invest in, it's a fast yes,

Alright, maybe you'd seen this slide, but it answered, it tried to answer that question of where does the money come from? And in fact, we, we can, a custom algorithm can replace a lot of things. So what we showed is that people are gonna pay less for audiences. Like 15% is a big tax for a static audience that's probably overstuffed, right? We can create audiences that are custom and dynamic and we charge less than 15%. Then there's the monitoring. I don't, I don't wanna, I don't wanna disparage anyone, but they're taking a big cut for viewability monitoring and brand safety monitoring. That I believe will come down. And then there's, and, and that's just viewability and brand safety. Then there's carbon emissions and other kinds of, of quality monitoring. There's, right, there's a whole bunch of tools that all could be built into a custom algorithm. We'd call, we'd call it like a governance algorithm that let's clients embed their values as the rules for the algorithm's decisions. And we, it seems to us there's already a lot of money being spent on sort of point solutions that we could replace.

Adam, have you ever quantified the, the, the human or, or sort of like people resources that normally go towards like manual optimization and, and Measurement that, you know, you, you can then remove and put, put those people onto higher value things.

Yes. And, and people tend to not even believe us, but we have, we have a, a true case study of saving a trading desk. I think nine people.

Right.

Alright. That, that to me is the answer, right? Because you just, you know, you, you do the, the, the, the total comp of, of nine people, you know, you're, you're over a million dollars and I don't know what the A C V is for that customer, but I would think it's south of a

Well, the conversation I've had with VCs many, many times over the years, or, or just people, analysts, is that why, why should people be involved at all? Like basically it's ai, it's ml, rocket Fuel. Did it? We'll talk about them later, but Rocket Fuel did it so like, It should all be automatic. And I love what you said earlier about steering against the autopilot. 'cause that's really in my experience, what the best the best traffickers, the best clients at places like AppNexus and beeswax ended up having to do. No one wanted it to be auto.

It. It's interesting. I don't know. I guess what I'll say is we've been steering with the autopilot for some of our clients for. Two years, some almost three. So like J really thrives off same store sales, like the results get better and better. And when you're steering with the autopilot, there is a lot of work to do in interpreting the results, in tweaking the algorithm. Media is full of change setting up more tests getting creative insights out of it. Just trying to figure out what's going on in the market. I think a lot of VCs. Under, under how hard it is to influence people through advertising, right? This is what I'm fashion about. I'm an advertising guy, so I've seen a lot of engineers, data scientists, and money guys, bankers come into it, come into my industry with a lot of arrogant. Beliefs about how easy it is to influence people when you're actually doing it. We're actually driving incremental l t v for an insurer for two plus years now. It creates a ton of, of work of that to interpret results and, and set up tests and deal with the changes that come. And it's, it's, it's fascinating and, and worth human

Yeah. I mean, even a signal, like a click is not representative of what's really happening between people's ears when they see an ad. Advertising is literally working on the neuron level, and there's no way to measure that yet.

That's right, and we've been using these crude proxies and, and I'm in the world now. I'm, I'm, I'm in the fourth year of getting past those crude proxies and modeling to the real signals of influence and yeah, it's very different. Very interesting. And it's great work that we encourage people to get

So did we learn anything from the side Bids? Exit?

Learn anything.

Yeah. I mean, is it, is that encouraging that sign?

Oh yeah, it changed, it changed my world completely. 'cause now, now it's a real category, right? Like think we'll probably be on the Loom Escape. And you know, I'd raised a lot of eyebrows. I got reach outs from every ad tech banker and got to go run my mouth about how graveyard and how we're not raising or selling. So that's, that's been a lot of fun. Like, it's, it's, it, it turns attention to us saying, oh, custom algorithms are for real. Now Chalice is the independent leader, so, It's, it's been a huge tailwind.

Is, is there, should there be a bubble on the loom escape for this? Because, you know, from my understanding there aren't too many other companies doing this that are independent companies.

Yeah. I, I don't, you know, I don't see such a barrier between a generative AI company, like memorable and what we're doing. I think there's some technical difference, like the, whether the outcome is an ad or an

It's the application.

These are just different applications. So I think the box on the Lumi scape is, is AI applications that make advertising better and more operationally

I mean, shouldn't agencies do this eventually?

Yeah, that, I mean, I've come from agencies, so I had the experience of seeing amazing tools built on the outside and brought into agencies. I, so two of my mentors and investors are Matt Kreitzer and Art Muldoon, who have ran that path with accordant into Densu. So I, I saw that happen. Close up. And yeah, I I tell the team all the time when they deal with agencies, we, we might be working there.

All right. So let's talk platforms 'cause you have probably more insight into the platform. So first thing is really remarkable, which is that I believe you're the only company to ever get an investment from the trade desk. Is that accurate?

Yeah. At least publicly we were, yeah, the first and publicly,

Yeah, you're the first.

So why'd they invest in you?

The, you know, they were unlike you, they were swayed by myself, I was I didn't, I wasn't friends with them, but I did have a relationship with the founders from being a big, early customer. I built one of the first big agency trading desks around trade Desk. So I was at like their early summits when the only agency guys were me and Jay from Good way. Then we started seeing Omnicom and W P P show up and we had a less, less influence in what they built. But being there early, I had some rapport with Jeff and Dave and. I, I, you know, I like they built multidimensional bidding and, and the user scoring application. So I knew that they were in, they thought this way, right? Their whole pitch is around bid expressiveness. So Jeff had done a podcast where he talked, you know, he, he's also a competition advocate, which I am passionate about competition. One thing he said is that he, the competition should be around things like prediction, right? And not about who has the most logged in users, right? Or who has a massive private user data. It should be around things like using ai. So I brought that up in the pitch. But I don't know. I I wouldn't even have called him if Ally, my c o o, who incidentally is also my wife didn't nag me for months during our seed round saying, you should call Jeff. You should call Jeff. You should call Jeff. And I was like, I've lost so much confidence since Ari said no. I don't know

This is a great lesson To anyone listening who I've said no to, you could come on the show and just abuse me. That's like your. For if I invest in you, you have to be nice to me. But like if I said no, just come on up, come on the show. Alright, so

For

let's go, let's go around the horn about the different platforms and their, and their custom algorithm capabilities. 'cause you probably have more insight than anybody else. So, in my view, the four platforms that have custom algorithms in any real sense are Xandr, beeswax trade Desk, and DV 360.

Yep. And medium one. And.

It will rise like a Phoenix. Okay, so you want to give us a hot take on each one? Like what's good, what's bad on Xandr first maybe.

Yeah, sure. I mean, they're not none of them are bad, but they're, they're different. And I would say yeah, they all have strengths and weaknesses and it's, it's very interesting, right? 'cause our software runs the same type of modeling most of the way, and then the last part of the pipeline, and it, it translates the model into bid instructions for the, for these different platforms. Yeah, so Apexus was the first built and is well built. It's Brian had a lot of foresight in how he built it or had it

no, it was pa it was past

Brian

time that they brought that out.

you weren't there. But anyway, it seems it's for something built a long time ago. It's, it's certainly aged well, and it's a, it's a bonsai tree model, which is, Not the latest th things in data science. So our data scientists say they have to, you know, they have to moderate some of the most advanced things they do to fit that. But it has a lot of foresight into things that are really important in making these bid lists, which are like really long lists of instructions. Like it has wild cards, and it has the ability to say anything that's not this bid that, right? Like that's, that

Yeah. Yeah. We used to sell against it at beeswax. Just, we just say Bonsai tree over and over again. And people would say, oh, I don't wanna do that. Okay. You wanna You wanna do beeswax next?

these wax is, is an maybe overall the best. It's the least limited, which everyone loves. We've done, we've done, we did a 50 million line bid list there and there. It might might not be true anymore. 'cause like that actually costs the d s dsp right. When you give 50 million lines to look through on every bid request.

I dunno, I don't know what they're up to. I mean, they're doing really well, so I dunno.

Oh yeah, I know they're doing well, but I, I was looking for confirmation that it costs the DS P,

everything.

it depends. Yeah. Okay. But anyway, it's, it's, it's unlimited. So you could, you could, you could upload a huge model and the combination of controlling the bids and the Q P S is ultra powerful, right? If there's a filter on the qps, no matter what model you put in its ability is limited. So in these vaccines you have, not only do you have QPS control, you could also put in a pacing model, have total control over the

That was awesome. I love building that. That was cool. Alright. Trade desk.

Trade desk is, is awesome, right? It's the most scaled software if it gives you the most service. There are clients in there using it, using the capabilities all the time, which means they're a little better maintained. And you can find people who can answer really d difficult questions about stuff that's not documented or. That's incorrectly documented as, as we found. So, so they, they have a, they have a big capability and it has something really close to bid models called multidimensional bid lists. And uniquely it has user scoring which is, which is a gold standard for finance. Like we tell, just tell finance clients, you know, the way you have every household scored for direct mail with their L T V. How would you like to do that for every household in America on C T V, that's what we could do with a model into

I think both BW X and AppNexus have user scoring, but it's just like a little awkward to use, whereas the trade desk made it really easy.

Yeah, that sounds fair. Yeah. You said, I know more about it than anyone, but it's the Chalice data science team who.

All right. Last one. DV 360.

DV 360 is not quite a custom algorithm capability. It's, it's, they call it custom bidding. And oddly enough, it's bidding that you don't control. Google still bids. The Google algorithm decides what to bid. You don't even get to enter a bid, you get to enter a max. But the Google algorithm bids, what you can do is customize on the value side. So you could give different values to every floodlight. On the page, a whole set of floodlights. You could, you could put value weights on geos and devices and times of day. It all goes in as a python script. They're the only one that will let you mo upload a model through the ui, which is nifty. And you could use the a p I and upload a bigger model. They give you some visualizations, the insights of how your model is doing. They're the only ones who do that, and it's the least buggy. Like it, it always works. The downside is you don't get control of the bid. So you don't get much cost savings, you don't get as much cost savings. It's most powerful to model the value side and the bid side. And the other thing is it's, it's very sensitive. Like girl Google is bidding with a, a very advanced neural net algorithm. I. And it looks to us like somebody's job, it was, was to just make the algorithm x more predictive by using a more advanced neural net. And what you lose when you do that is the ability to change anything without the algorithm needing two weeks to learn. That sensitivity is a problem. Like in the media world, people do not like waiting two weeks for their algorithm to relearn and get back to performing after

All right. This is if in case anyone ever says the architecture podcast is not delivering value, we just got a masterclass in custom algorithms. Alright let's take a quick break and come back with the news of the week. All right, we're back. So we're gonna start on a sad subject seismic. Let's have a moment of silence for everyone's favorite company, seismic. Alright, we're back. So let's do a quiz. Let's see if I, I did some research this morning. Made a list of every company that Seismic had acquired or been acquired. Every brand name they've gone under. Let's see if we can name 'em. Eric, you wanna give us a start? We'll go around. Good one.

I blast her.

Media mind.

I have the list, but, so I'll just throw one in that you guys probably wouldn't remember. I wonder.

I was gonna, my next one d d.

Adam.

Was, did they, were they ever called Amazon

I didn't even think about that one. Okay. Yeah, I don't know what

doesn't count. Yeah. And then the obvious, the obvious rocket fuel.

else? There's, there's another, there's another four.

Did they, did they take unicast or did that go

Uncast, the innovator

My God.

precast giant ads that took over your screen.

It inspired me, man.

did you ever get sued by them? They were like really litigious with their patents.

No

all right. Three more. There's one as a mobile. D s P.

Oh, did they buy

that was, that was that's violent.

no. It was via,

Strike add.

Strike ad.

pretty good at this. All right. There's two more. I'll, I'll just give you them. Pier 39.

Oh

Of course.

then this one I really didn't heard of was aerify. You guys hear of fy? I don't remember.

I didn't know that. Do you remember the time that seismic exposed every customer's data to everyone?

No, I don't. I don't remember that.

Oh.

Yeah, you do. That was incredible. What a day That was. Incredible. Day in agency ad tech world. You could just go through everyone's.

Yeah, it's Yeah, it's sort of an exercise to the reader to, or the listener to figure out how much equity value was destroyed in this process. It's easily in the bill, I mean, if Rocket fuel alone is in the billions. But basically this company got bought, sold. Dis Georg Disgorged DG is the survivor. So DG got spun out into Extreme Reach, so that still exists. And Pier 39 got spun out to Mario Diaz and his Merry crew. But otherwise all of that went to nothing and is now shut down.

Yeah. Wait, bit more trivia. You know, what was inside of Rocket Fuel which then

Oh, I forgot about x

was X

All right. That's a good one. Right? Which is total totally different from one plus x. Absolutely. Different company, which is owned by

Yeah. I'm, I'm, I'm a, I'm a ton of fun at at parties, as you could tell.

Yep. All right. So Adam, I wanna talk about your Twitter persona. 'cause you're a little bit of maybe a bomb thrower on Twitter, like you don't wanna get on Adam's wrong side, as we've seen in this conversation I So let's, lemme start with the easy question. Do you love Google or do you hate Google?

Neither. I mean, it's a company, right? There's a lot of people there. I like I, I think the company's too powerful and I get passionate about it 'cause I'm passionate about advertising and I think a lot of problems in advertising can be traced back to their conducts. I get frustrated that people don't see it. They say things, you know, people say things that are indicate that they don't know much about what's legal or or ethical in ad tech, or that they've, they've given up and don't care. Sometimes I, I get fired up. I don't, I don't wanna hurt anyone's feelings and I don't wanna be seen as, As unkind or cruel. So my, my intent I'm sharp. I grew up arguing, you know, in the seventies it was my, my first I learned to talk in a rough

All right, so I'm gonna, some greatest hits. Here's some Adam greatest hits. I'm gonna, I won't do a dramatic reading of this from Twitter. When ad experts promote the narrative of a corrupt middleman and ad tech, without mentioning header bidding or antitrust, they carry water for Google.

Yeah, I real, I do think that's true. I think, I think a lot of people unwittingly talk un unwittingly support Google's position. The best example of that is third party. Like if you think third party is demonized, right? That it's bad to be third party. That comes from Google. That's Google's PR strategy. Like they demonized the third party cookie. And then they said they're a first party. When they're intermediating, they got that written into law into G D P R, and it's ridiculous and everyone should say that's ridiculous. When I'm reading the New York Times, Google is not a first party

right. Here's a good one. is a really good one. I like this one. I think the big misunderstanding is around what powers all the impressions. Algorithms. Impressive algorithms. It's like being amazed at engines and not knowing about oil. That that's gold. That's a good one.

Yeah, people talk about AI and algorithms without thinking about training data. Like every conversation should start first with the outcome. Like what outcome is this supposed to drive, and then what is it trained on, right? If you don't, if you're not grounded in those things, you can't have a conversation about

All right, one more. It's Google's domination of ad analytics, specifically their self-serving failure to distinguish correlations from causal ad effects that incentivizes publishers to create more instead of better ad placements.

Yeah, I don't remember any of these. But I.

you're in a fugue state when you get in front of Twitter.

It's a, I just, I fire them up really fast, so it's, yeah, clickbait. Everyone is aware of clickbait. Everyone hates clickbait, but very few people say, you know, why is it that publishers can only make money from driving more clicks? Why don't publishers make money from having a more. Higher quality publication with, with more reach to readers who buy stuff. Like, that's what advertisers care about. And the reason is Google's domination of measurement, right? They, they have an algorithm that doesn't care about quality, that doesn't care about things advertising advertisers care about, and it pushes publishers to make crap sites. That's the only way to

I, I don't know if you can blame Google for that. I mean, the last click, last impression have been around for 10 years before Google even entered the display market.

Well, it's a, then you must agree that it's a, a striking lack of innovation. That 20 year old technology is still dominant

All right, enough about your tweets. I'm just glad someone's not focusing on my tweets. So let's go through some news. So I thought this was kind of interesting. Camelot was acquired by P M G, so Camelot has long been a very innovative independent agency. I, the news reporting didn't really give a lot of rationale for it, but, but I thought it was kind of interesting. I not that familiar with P M G Adam, are you familiar with these companies? Does have anything to add?

Yeah, a little bit. P M G. Yeah. Their agencies, Camelot kind of had a custom algorithm narrative for a while. They were talking about doing more advanced measurement and P M G based in Dallas. I think I got a friend there. I. Yeah, small, small agency. I would just say there's a trend of small, independent agencies

They also tend to be the most advanced with the use of tech, I

Yeah,

Yeah. That's where the change will come.

Yeah. Camelot was like, invite media's first customer ever.

Yeah, yeah, yeah, for sure. I I, I used to like fly down to Dallas. That was always a, a, a really good, good market generally. And spent, spent a bunch of time with them and they, they were always forward-leaning, always like pushing vendors. My sense from reading a little bit about this is that they've been you know, really forward thinking and strong in C T V. So I think thesis around the deal here was was really around their, you know, their ability to to, to help catalyze C T V and have, they have a lot of relationships with the key platforms and so on and so forth. So I think this is a, it's sort of great you know, for, for the team at Camelot, and then b largely, you know, sort of like another, another bet on, you know, the future being being C T V.

I was just a speaker yesterday at programmatic mechanics summit. They're, they're a sponsor of this episode. So if I call them out, 'cause they have introduced a new C T V D S P that's interesting because it's very simple and that's kind of like what a lot of people want with C T V. All right, next. So the ongoing Google trial I think the interesting news this week was Microsoft, c e o, Satcha Nadella who went into court and claimed that little old Microsoft has no power. They're totally defenseless. Everyone should, should worry about their, they're being crowded out of the search. Business by Google. He, I don't have the exact quotes, but he said basically they have 3% market share in search. There's no way, he said there's no way for them to compete because of Google's virtuous cycle of getting more searches and therefore having better AI to make the searches better. He also said that if you think about it as software search is the largest software. Sector it much larger than office, larger than operating systems. It was kind of interesting.

Yeah. And what about Looc? I mean, almost no one reported on it. I have no idea why, but this should be huge news in Ad Tech that the Joshua Lowcock, the Chief Media officer of I P G, went and testified and talked about how you can't really multi-home with SA 360, right. Which is, has huge implications for the ad tech case, right? To see how the court reacts to this news that you have to use Google's tool and Google's tool doesn't do a very good job of bidding into Microsoft search. I think Joshua. Yeah. Right. No one's covered this.

Is SA is SA 360

Joshua should be held as

but is SA 360 a focus of the case? I thought the case was all about distribution.

It's a no. There's, it's, it's a piece of it. Yeah. That the, that you can't, that the, there's barriers to multihoming in the, in a tool that you must use.

So.

Joshua should be held as a hero. Like I, I testified to the Senate, it was very easy. You know, the questioning was very friendly. Joshua went down there and stood up to cross examination for real. Like, it's, it's a lot of work and takes a lot of courage. I, you know, I think he was mostly alone down there, as far as other buy side agency leaders and I just wanna commend him and call it out. It's

Interesting. So we'll continue watching that case as it evolves. So Jeremy Gordon is out at Netflix. So Jeremy was the spearheading the advertising. So adage reported that she is out. Jeremy is a female name. I, I don't know. Her personally, Mike Shields, who's a contributor to architecture in his newsletter. I think some of it up, I'll read what he said. She's being replaced by Amy Reinhardt, who had previously been Netflix's VP of Studio operations, and that was after helping to oversee content acquisition. So clearly this is not an ad person, meaning the person's replacing Jeremy. It's interesting. I think that it's not really, in my opinion, it's not really surprising that Netflix building a whole new ad business from scratch might be difficult and might not immediately pay off some of the dividends that might have been expected by Wall Street or the c e o. This seems like it's some news about, this seems like it's meaningful news, I guess I would say.

Yeah, well, we were not gonna find out what happened for, for a while. If, if ever it's a very, it's somewhat secretive company. They were, they were among the first to use custom algorithms along with Uber. There was no case study, but I was able to find some stuff out like there. They've set up some very productive learning loops between, you know, what shows help people acquire and, and using that for making certain shows and choosing what to promote in certain regions. So I, I think it's just an amazing company and I don't know why they'd have, would have trouble doing the sell side thing, but, you know,

Yeah, they certainly have the data science.

Yeah. Yeah, they've got a giant data science team. I think that your instinct that this, you know, feels significant is, is right. But also, again, like to Adam's point, it's such a, you know, like se secretive company and, you know you know, insiders that have, that have sort of like grown up there and, and, and know how to, how to navigate could be better suited to, you know, helping to, you know, drive, drive a, a new biz drive, drive change. But they also set the expectation in one of their last earning calls that, you know, the expectation for this year, Is that revenue is di minimis and they're experimenting with different pricing tiers, and they're gonna be rolling this stuff out slowly. So I think they're giving whoever's a leader, the, the headroom. I, gosh, I, I'd love to just, you know, like talk to somebody there. And, and understand like why they don't go for a buy strategy. Like, you know, they, they just, you know, could, they could buy any company on the, on the loom escape, right? And, you know, rocket themselves into like a really interesting position from a, you know, from a a buy side perspective or even a sell side perspective, right. You know, depending on how, how you think about that. And you know, is that even a possibility? 'cause Netflix is not very in acquisitive as a, as a business.

that the piece they would want would be a publisher side, ad server, yield manager, and there aren't any available for sale. Like publisher side yield managers have all been acquired. You know, our, our, our sponsor Publica Springer. FreeWheel's too big and part of Comcast a magi is too big, is big Indian server side company. There's really nobody out there to buy. So yeah, but they, I, I think they obviously are looking at, at how to accelerate. Alright. So Eric, you wanted to talk about Facebook's generative AI tools.

Yeah. So they launched like a whole bunch of tools this week for for generative ai. So I'll, I'll rattle 'em off. It's just, I think this is super interesting in that, you know, meta, Facebook, meta, you know, whatever you wanna wanna call them, they're they're, they're, they've got their foot on the gas when it, when it comes to this stuff. So background creation. So create multiple backgrounds to compliment the advertiser's product image. This, again, this is all using generative ai. Image expansion. So like, have something expand to fit into newsfeed or reel or Instagram post. So, you know, taking the, the sort of like fit and creation point part out of, out of the, the editing text variation productivity boosts, like, you know, people are saying that this can save advertisers and agencies up to, up to, up to 50% of time. Like this is big stuff. You know, just given the amount of spend that goes to, to, to, to meta across the various properties. So I think it's a neat use case of all that is being done with generative ai. And I think it's interesting to Adam's point earlier that, you know, I. Applications of ai. But it's starting with creative. Which again, I think is like the, the lowest hanging fruit with respect to efficiencies to be gained from a a, a people perspective. And you know, ultimately, like if you move the needle on performance, on creative, I. You can have some like, really big impact because it's such low hanging fruit and everybody forgets it to the expense of things like targeting algorithm algorithms and, and optimization algorithms. I e you know, 95% of the time we spend here. So neat, neat stuff worth checking out.

It's also the, the best way for a network to compete with a open web is, is creative. Like they, that they can innovate much faster, find interesting ad units, exciting ad units, new ways to get to grab attention and, and to speed up the productivity of their customers. It's amazing, you know, in, in open web, programmatic, we have to wait for the I A B to approve it on a new format, and then everyone has it. So it's really an, a natural advantage on the network side, and it shows Facebook's competitive prowess, that

a closed environment with standardized ad units, it's kind of the perfect environment for using ai. And it reminds me of Mar Pipe, which is a company that I did invest in, and so does Eric that does, does this sort of thing. The last news article I saw that I, I thought was interesting was a Financial Times article about Twitter X And the crux of this article is that one of the factors that's really hurting their revenue, and we've heard so much about the revenue decline, is that Twitter's ad business had a very large investment in what they called Amplify, which is where Twitter sold ads. As pre-roll on content from TV networks. So this some, sometimes you'll see on Twitter, it won't be an ad. It'll be like in the search tab. There'll be like a clip from a, from, you know, the previous night's football game or basketball game and you click on it and there's a pre-roll ad on that. And and that's a very valuable spot, needless to say apparently it was generating more than a billion dollars in advertising revenue. And the problem is, That Musk when he came in, fired the partnerships team. Punchline

Yeah.

Hard to do partnerships without a partnerships team. So and, but meanwhile, Musk is saying, and I've heard from various sources that they're gonna re double down on sports which kind of raises a couple of eyebrows here and there. I don't know. I don't know if either of you guys have a thought about this. This just kind of sounds like a a dumpster fire.

I went to the search page to see it right now on Twitter, and I'm getting this ad for the Exorcist

Yeah, that looks like a scary movie.

very scary.

Yeah, I'm not, I'm not down for that. I'm gonna see Taylor Swift and imax. I'm not, I'm not doing that Exorcist stuff. Alright. I think we've, I think we've covered all the bases. I think we're gonna wrap this up. Adam, you're, this was a great conversation. The thank you for being here.

Thanks. I appreciate. Thanks for having me, and come on in. We got it safe for you.

alright, so please stay tuned for justify your existence with Bon Bond is another company I actually did invest in. And it should be interesting. So thanks Ann. Thanks Eric.

Thanks guys.