Loading summary
Transcript91 lines
- [00:13]
A
Welcome to Ad Exchanger Talks, the podcast devoted to examining the issues and trends in advertising and marketing technology that matter most to you.
- [00:24]
B
This episode is sponsored by Verve, an ad solution that helps brands activate consumer intent in real time across platforms. As AI reshapes how people plan vacations, Verve helps travel brands show up when it counts.
- [00:47]
C
I'm Alison Schiff, Editor in Chief of Ad Exchanger and my guest this week is Ali Manning, COO and co founder of Chalice AI, an ad tech startup that helps brands build their own custom AI models for media buying. We'll talk about her journey from big jobs at Google and Snap to building technology that gives advertisers more leverage against the big ad platforms. We'll also get into how Chalice nearly ran out of money before turning a corner. Plus why Allie thinks the industry is way too obsessed over the wrong kinds of quote unquote outcomes. Oh, and if you're interested in containerization, then you're listening to the right podcast. But before we dive in, programmatic I.O. new York is coming up on September 28th and 29th at the New York Marriott Marquee in Times Square. There it's time to unleash growth across the open web. And to do that, we've got a seriously good agenda lined up for you, with excellent speakers including Chris Kane and folks from Horizon Media, pmg. Hey dude, Dentu X, Razorfish, Goodway Group, the Guardian, and more. Podcast listeners get 10% off with the promo code POD10. So what you waiting for? Get your ticket and see you there. Hey Ali, welcome to the podcast.
- [01:58]
A
Hi Alison, thanks so much for having me.
- [02:01]
C
So what is one thing about you that not a lot of other people already know? Like the fun or the weird, whatever the case may be, facts about you that someone couldn't find out just by looking at your LinkedIn where you're very active by the way, or googling you sure.
- [02:19]
A
I love ocean swimming. It's one of my favorite things and New York City is actually really great for it. Off of Coney island in Brighton beach, there's some really calm open water where you can swim. But the fun fact is I love it so much that one year I just said, you know what, I'm just going to keep swimming here every week until I can't. And I did it through October, November, December, January, and I got to, yeah, I got to mid February and I swam in 34 degree water and then I couldn't do it anymore. So for one season I was a real polar bear. But now I just swim June Through October.
- [03:08]
C
That is really impressive. I did something recently in the ocean that I'm both proud and not proud of. I took a surfing lesson and I was so bad at it. I was in Ireland and surfing is actually a big thing in the west of Ireland. They have good coastline. I took a surfing lesson in Sligo. I was terrible. It's so hard. I couldn't even stand up on the board. It was a two and a half hour lesson and I was doing it with little kids and they were all standing up on the board instantly because they're fearless and I sucked. It was extremely difficult and I was so achy afterwards.
- [03:43]
A
Yeah. I have also done surf lessons. I did get up on the board, but it was in Costa Rica, which I feel like has a much calmer coast than probably western Ireland. But yeah, also was like shocked how much strength it takes and coordination and other things I'm not great at. So.
- [04:01]
C
Yeah, well, same Z's. So let's set the scene a little bit to go back in time and then bring us up to today. So you're the COO and co founder of Chalice AI, formerly known as Chalice custom algorithms AI software company that builds customer ad buying. Sorry, custom ad buying algorithms for brands, for customers. You guys were named as One of Business Insider's hottest ad tech companies. You won best DSP from ad exchanger in 2022. So belated. Super belated. Congrats on that. But before Chalice, you had a career going at two very large, very recognizable companies. You led strategy and operations for Google's brand ads business. You helped establish the US go to market for YouTube and programmatic, and then you ran ops for Snap's global ads business. So you went from running operations at two of these enormous ad platforms, and then you're working at a startup now explicitly designed to give advertisers leverage against the big platforms. It's a really interesting transition, or I guess pivot, if you will. So what, what made you want to go and do that?
- [05:13]
A
Yeah, well, I will say when I was at Snap, which was like the heyday pre IPO Snap, where it was the hottest platform on the block, our sales teams would come back and be like, the customers are desperate for more options than, than Google and Meta and they really hope Snap could be like a third, third leg in the table. And so just that impression on me of that advertisers really want more freedom and independence. And then I saw at Google, like, we, you know, kind of told the CE CMO of P and G to like, you know, go take it. Like we're not going to, you know, we're not going to customize our products for you. And I remember Mark Pritchard, I think rightfully being like, what do you, what do you mean you're treating me like that?
- [06:08]
C
Don't you know who I am?
- [06:09]
A
Yeah. And the sales team did, but the rest of the company didn't care. And so when my co founder Adam came to me with like, hey, we could use AI to build tech into the platforms that really reflects what the brand's true goal is, I thought like, wow, that's something that brands are going to want. It was six years ago. AI felt a little like nebulous to me. So it was exciting to figure out like that it was, is a real thing that we could deploy. Right. I was like, how are we going to deploy AI models that could compete with something like what you can get off the shelf with Google or Meta or the trade desk. And what was cool for me to learn is like, well, what matter? They have powerful models to like serve hundreds of thousands of advertisers and they're competing priorities all at once. Right. And like fit you into a good click through rate or like last touch cpa. But a model that's trained for a brand on like the data that matters for it and to their outcome is actually much more powerful for that brand. And so kind of learning that and seeing it like play out in the wild has been very cool.
- [07:31]
C
So bear with me for a little seemingly unrelated anecdote, but my fiance is kind of a dandy. He's like an aspiring dandy and he started a job where he has to wear a suit and he got a bunch of custom shirts made. But before, and jackets and pants and everything. And he's loving it. But before he did that, he bought a few off the rack. They were Calvin Klein shirts and they looked totally nice on him. He looked fine and then he was like, oh, I really want to do this thing. I want to buy these custom shirts. I'm like, well it's expensive, but okay. And he looks so much better. He looks really good in these shirts. So yeah, I mean the other stuff was perfectly serviceable, but it wasn't optimal.
- [08:18]
A
Yep. Yeah. And for brands who are like in often a knife fight for share. Right. Market share or any competitive. Competitive advantage. Right. Similar to if, if your boyfriend were like, you know, in a fashion contest or even maybe applying for like. Yeah, right. Or applying for a job. Right. Like, like the, that edge on maybe his like very Polished appearance could make or break him getting the job right. Or certainly it would matter in a fashion contest. And so the brands who are working with us right now are typically ones who are like, I need any competitive edge I can get. And models Built for Me feels like a really good at least thesis to try out to get that.
- [09:09]
C
So we are going to talk more about models Built for me and how that actually works, the technology. But before we do that, can we go back in time to when Chalice was first starting? Because I don't think you actually planned for this. Right. Like Covid hit the job at Snap, just, I guess, went away. You got pulled into Chalice gradually. You were helping draft the business model, you were workshopping pitches, but you hadn't fully, fully committed. But Adam, your CEO, your co founder, also, you're married, he was testifying about Google before the Senate. It was September 2020 in an antitrust hearing. And that was this like lightning bolt moment for you that made you realize that you belong on this side of the table. And we wrote about that hearing, actually. It was so fun. It was wild to see someone that I knew because I know Adam testifying about Petter bidding in front of the Senate. Amy Klobuchar was there and he's talking about header bidding. And this is years before the Google Ad tech antitrust trial, which was also wild. But what was it about that moment, like in time that made you finally want to go all in on this thing?
- [10:20]
A
Yeah, so I just, just an anecdote is when they broke in the Senate hearing for, for a break or whatever. I remember hearing Mike Lee say to Amy Klobuchar, where did they find? Like, they were really, they really impressed with him too, like that he. Yeah. Translate those complex concepts into like concepts that the senators could understand. So lightning bolt for that. I mean, I just, I remember sitting behind, I stood behind my. Sat behind my man and we had to listen to the Google testimony first. And I just remember sitting through the Google testimony and like, I was part of it. I was unknowingly, like encouraging on, what's it called? Uncooperative or like behavior that you would do as a monopoly. Because I didn't know, like, it, it was like it's a rational behavior to do in a, in a way, if you are a huge company who is very saturated, you have to get more market share. And so if you don't understand all the limits around antitrust, like, it's rational to do, you know, push for things that we are pushing for, like tying YouTube to DV360. We were celebrated the, you know, the hell out of that when that happened as a Google, like, go to market organization because we're like, oh yeah, that's going to make everyone need to buy DV360. Like, we were, um. And so just hearing the Google testimony, it just, it felt like a lot of bullshit for me. And it also kind of, you know, ran into that experience that I'd had personally and I just felt like it was disingenuous. And so by the time Adam got up, I was like, yeah, like, go for it. And yeah, and then I think at that point we were really then on the path. Like, I had been talking to Google about going back to Google because at the time I was like, someone's got to go get like a, you know, big paycheck and the benefits and everything. And so obviously, I don't think I fully understood the weight of it until like, obviously, like he was testifying and then I was like, oh, yeah, like, that route is like done for me now too. Yeah, it's actually kind of sad too. I like, lost some big friendships I had over it, which I think is lame. But some Google. Yeah, some. I mean, Google is a bit like a cult. You know, I talk to friends who leave it as like your deep programming from a cult a bit. And so, you know, like, it's sort of similar if you leave a cult and talk out about it and like, what's really going on in it. Some people who are in the cult are not going to want to be your friends anymore. And that was, that was really hard for me personally. But anyhow, all to say is that it sort of forced me into chalice in a way at least forced me not to work at Google in a way and then. But I really thought, like, yeah, we're on the path to something big and a little righteous. And that's like a really good combination for me to get excited about.
- [13:45]
C
You were kind of radicalized in a way. You're like the Lisa Rinna of ad tech.
- [13:53]
A
Yeah, it went for me, like, just being like, I, you know, I stand behind Adam, like he, like, because he believes in this. To like, I left the senate room being like, I believe in this too. Like, I believe in his position as well.
- [14:06]
C
It must have been a surreal experience too, to fast forward a few years when all the testimony was happening in the Google Ad tech antitrust trial separate from the search one. It was so strange for me. I remember sitting in like on the benches and taking notes and Actually looking at the other reporters, the ones that weren't in the ad trade, so just antitrust reporters and some news reporters from bigger publications that were there because, hey, it's Google, and you couldn't bring a recorder into the courtroom, so you had to take physical notes. And they were just like, ah. I was, you know, learning about antitrust and what that really means. I didn't have any background in that, but I had the ad tech background. Watching people try to scribble notes about the intricacies of header bidding and all of the different Google Ad product names used interchangeably, I had such compassion.
- [15:07]
A
Oh, my gosh. Yeah. I would joke with new hires at Google that it's like learning another language. All the acronyms we had for every product and project around a product and yeah, it's funny.
- [15:20]
C
So I want to talk a little bit more about the origin story. Just I think it's really interesting and it's fun to do because you were successful in the end. Because, yeah, it's like, it's a bummer if you, you know, put your all into something and it doesn't happen, but it's, it's working. Um, it's a really inspirational startup story. So in the early days of Chalice, like, there you are, you're on like, cobra. Childcare is a challenge. Uh, I remember I, I think it was Peloton. I read this somewhere, like, your largest early client, they pulled back post pandemic. They were having a very well documented hard time of it themselves. And like, you nearly ran out of money and then things were like, really dire and the series A that you were collapse, it was just like, oh, how close did you actually come to like, pulling the plug and what, what saved you?
- [16:13]
A
Well, I like the very earliest days because, like, it's now two incomes on this one company and our shared savings account funding it. So I was like, well, I'm gonna come up with like the model that says, like, okay, if we're not making enough money to like pay ourselves, like, we, you know, like, it's done. And we would get to those near those milestones and every time it would be like, something though, is on the horizon, someone just verbally said they're going to work with us, or someone said they're excited about investing or just whatever it was. And it was like, okay. And I learned just how far your pennies can stretch because we stretched it a long time. But I, I'd never gotten to the point, like, I had to do this. So actually between Snap and Chalice, I worked at a property tech company and was like, I'm done with ad tech. And then I learned how hard businesses that require, like, physical goods and people are to operate, and I was happy to come back to ad tech. Remember that when Covid was hitting, that was really hard on that business because it was, like, about luxury real estate in urban areas, which was, you know, and so. Yeah, But I remember I had to write a plan there that was like, work with the founders to say, okay, like, in waves. Who are we going to let go as things get more and more dire? Like, I've never had to do that at Chalice. Like, think everybody. But I would say, yeah, like, there was a moment where it was the summer we had signed who was. Who would ultimately be our biggest customer and would essentially be our Series A by how much of a big customer they'd be. But we'd signed them, but it was, like, slowly starting, and we were like, we were raising a bridge round in a safe, and we kind of, like, gotten all of our existing investors on, some new investors on, but we were, like, still having a hard time with raising the rest. And one of our employees was, who is a gambler generally, was like, really wanted to gamble on Chalice. And he convinced, I think, at least eight family members to write checks, none of them huge on their own. But, like, the eight of them combined brought us through, like, a few months of the summer where, you know, we, like, had stopped paying the founders. Like, we cut anywhere we could. And that bridge chalice, then that customer started to scale.
- [18:49]
B
Scale.
- [18:49]
A
We were still raising our Series A. That eventually didn't work out, but as soon as that Series A didn't work out, that big customer invested enough. We, like, moved from test partner to, like, really on the plan all the time partner. And the budget's like 10x and we became profitable. And so, but it was like, yeah, really holding on is as much as we can. I like to say the only, Like, a company can grow a thousand different ways, and there's only one way the company can die, and that's by running out of money. So if you can keep some money in the bank stretched out, you still have a business and you can pivot.
- [19:33]
C
So a challenge, too is that it is a really good idea and that it's resonating clearly. And I do want to talk next about how it works, but it was a little early. So when you have an idea and it's just a tad early, you have to keep, like, running, even though you can see that there Might be a little precipice in front of the Runway, but like you still have to keep going full tilt until other people catch up with you. And then you could just like lay those bricks really fast like as you're running. And then eventually you have a, a nice road stretching off to the horizon. But yeah, so I want to give our listeners a really clear sense of exactly what it is that Chalice does. Although I think a lot of listeners are pretty familiar by this point because you guys helped create this custom algorithms category and a lot of people just refer to it now casually, although I know you're not that into custom algorithms as a category anymore, which we can talk about, but in plain language. Most advertisers are using a DSP. It's just running the DSPs built in optimization algorithm. It's essentially a black box. It's designed, nothing nefarious but just to optimize for generic metrics, things like clicks and conversions and viewability, et cetera. And what you guys do is build a custom bidding algorithm specifically for bad advertisers trained on their first party data. It's dialed into their specific business goals. And yeah, the core insight is that a platform's incentives and an advertiser's incentives aren't always the same thing. And platforms also optimize for what they can measure and for their own inventory. But give me an example, like a concrete example of what that misalignment looks like in practice. And like, do brands even realize what they're missing? Because what you're doing flips a lot of what Programmatic was built on like on its ped. Right. I mean it's an industry that's existed and worked in a certain way for, you know, a while.
- [21:36]
A
Yeah, yeah. I'll give you an example. It's our newest case study to come out which is with, with Bayer and their vitamin brand one a day. And so I think it's a great example. So Bayer is, is a Amazon DSP customer. Right. They have a broader relationship with Amazon that, that has, that has solidified them as an Amazon DSP customer. But over 80% of one a day sales still happen in stores. So like, so bears choices are like optimized to just the E comm part of the business, which leaves out 80% of where their ads can make a difference. Or in a traditional DSP you would then optimized to like, like you said, like video, video completion rates or view viewability, things that are pretty poor proxies for trying to drive offline Sales. So what we worked with on Behr is so they want to drive household penetration, right. That's like what their CEO would talk about on an earnings call as like how they grow their business. And so and the way they think about household penetration for one a day is getting more people who are new to brand shoppers. So people, people who haven't ever reached for it in the store like reaching for it. So Bear doesn't have a lot of first party data that would be useful in this regard. So they're like, they're kind of stuck. And so what we came up with is like hey, we have panel based partner who measures offline sales. We can use that to understand where ads are, where ad investment is impacting offline sales and essentially use AI models to model out who is most likely to be a new to brand customer. And so then what we'll do is essentially put more higher bids or more ad budget against those customers who are modeled as most likely to be that like new to brand customer for one a day. And so we did that, we pushed that into the Amazon DSP and so now they're able to like use Amazon DSP to drive offline sales and they've seen like a significant lift in new to brand shopping for the brand through that. And so it's a way to like the way I like short shorthanded is like getting using AM models to get most much closer to the metrics the CEO and CFO actually care about. Which means it's CMO cares about it and like just using the platform for something different than it's like. Right. It's off the shelf technology was purpose built for.
- [24:30]
C
Yeah, I mean you look better in a shirt that was made for you.
- [24:33]
A
Yeah.
- [24:35]
C
So we're going to take a quick break and when we're back we're going to talk a little bit about the category because it's changing and evolving but we're going to talk about containerization because we're a bunch of nerds and lots of other good stuff. So stick with us.
- [24:59]
B
Hello listeners. I'm Sarah Sleuths, editorial director of Ad Exchanger and I have with me today Anna Slater, VP of Global Brand Partnerships at Verve. Welcome Anna.
- [25:09]
D
Thanks so much, Sarah. It's great to be here.
- [25:12]
B
So we know that AI is changing how people plan trips. I've heard the anecdotes from my friends. I've done it myself. So can you give some insight into traveler behavior right now?
- [25:24]
D
Yeah, absolutely. I really don't want to date myself, but I've been in the travel industry now for almost 20 years and having worked so long in one particular industry, it's been amazing to see the overall progression. And I really feel now that the adoption of AI in this trip planning process is the most impactful transformation we've seen over that time. And looking at our own LLM data here at Verve, consumers are using it for general discovery. Around 28% of the prompts that we're seeing relate to these types of inquiries. And this is growing. And AI use for travel inspiration has doubled in the past year, especially amongst the 25 to 34 year old audience, which it's grown to around 18% adoption. And we're also noticing there's a shift away from keyword searches like cheap flights to Spain, and treating AI now more like a travel agent, which is adapting in real time. However, there's still an element of nervousness around utilizing AI to book. You know, since the prompts that we're seeing here are certainly dominated by that discovery, the timing, the itinerary building. But I'm sure if we're going to have this conversation this time next year, that will have all changed.
- [26:32]
B
So if I'm a brand and I know that AI is owning discovery, people are using it as a travel agent, but not for the booking part of the equation. Where does that leave brands who are trying to reach travelers?
- [26:44]
D
Good question. Travelers are using AI to navigate every stage of their holiday planning. But the biggest challenge with this new way of researching is having on brands is the pace at which it takes consumers from dreaming to consideration. You prompts get incredibly specific and in a very fast way. Someone can go from give me trip ideas straight to a shortlist without ever touching a brand's website. You know, and we're also seeing really heavy vendor comparison, you know, with travelers using AI to sort of audit those travel providers and then benchmark prices. So by the time the brand actually shows up in the conversation, the shortlist may already be set. So it is a very difficult moment for travel brands at this, at this time. And then furthermore, that intent is scattered across more and more devices and platforms. But these conversations, they're carrying some of the richest signals that we have ever seen on intent. And AI is forcing the conversation to accelerate the decision making. And if they can tap in to use these signals to target, they're going to get better performance overall and have a chance at winning that conversion of the traveler's itinerary.
- [27:52]
B
So you just described a very complicated, complex, scattered and rapidly changing set of consumer behavior. So what steps do you recommend to brands who are trying to capture this consumer intent at this, at this moment in time?
- [28:09]
D
Yeah, absolutely. And this is conversations that we're having with our clients daily at the moment. And the answer really is twofold. Firstly, travel brands need to see the intent and then also follow up and ensure that they show up in it. But there are a lot of brands that are still missing out on both right now. So on the seeing side, you know, that's LLM data, the conversational signals that we're pulling from AI chat tools. We layer them on top of actually search and zero party signals. And when you combine this, it provides a bigger, fuller picture of consumer intent. And importantly, you know, the LLM data shouldn't be seen as a replacement. It's a real complementary intent signal. So obviously we're encouraging brands to sort of see what's out there in market. And then on the showing up side, it means being present inside the AI conversation itself and not just hoping that the LLM happens to mention you in the brand.
- [29:02]
B
Well, thank you Anna for those concrete steps and recommendations and thank you to Verve for supporting Ad Exchanger podcasts.
- [29:10]
D
Thanks so much, Sarah.
- [29:14]
C
Foreign. Welcome back. And yes, let's talk about the category. So I remember seeing a LinkedIn post which I couldn't refind when I was preparing for our chat. I wanted to read it again, but it was something you guys posted where you basically walked away from the custom algorithms label. And it's a category you helped create, but you just like to shoot it. I do get it. Because you define a category and then everyone piles in, it gets a little watered down. Like buckets can be limiting. You got double verify buying side bids. Like most of the major DSPs have their own versions. Sort of like, like something like this now. So you change your name from Chalice Custom Algorithms to Chalice AI. Although I would argue that AI is hardly a less crowded or more defined category. Like, it's a mess in there. But yeah, like what, what was the thinking? And then what bucket can I toss you in now? Because I need buckets. I need buckets to understand this industry. Or I just, I get confused and lost.
- [30:22]
A
Okay, yeah. So initially the impetus for walking away in a more public, in a more public forum was the category to us had gotten like, sullied, especially inside holding companies. And there are two reasons. The first was there were at least one, if not a couple custom algorithm partners who were doing things that the agencies thought weren't on the up and up. And because of that, we would Hear stuff like, well, we're just, we don't, we're not interested in custom algorithm partners. And we thought like, oh my gosh, we shouldn't be in that associated with that. Also though, in the hold coast, a lot of them had custom algorithm products themselves. But what they really were were big data science projects creating one decisioning algorithm that got pushed once. And so there was also that confusion thinking that we were like that and it was getting, we wanted to really differentiate as like no, we're like software that runs continuously. Models are constantly refreshed. And then I guess and then so that's one that was the impetus. But then also like the, the, the term algorithm starts to get a little limiting. Like we now use generative AI in our models and we felt like continuing on the custom algorithm like as the category felt like it wasn't giving our tech all the credit it was due. Especially because we think we're on the cutting edge of using generative AI and new models to make decisions which you can talk about.
- [32:07]
C
Definitely want to talk about that. Naming can be so weird, right? Like Ad Exchanger. It's a name from another era, but we're sticking with it because it's recognizable but because early on enough you can make tweaks and you kept Chalice. I mean that's really what people would shorten you to anyway. So yeah, we feel like we have
- [32:28]
A
a really hot brand too. So like we don't necessarily need the custom algorithms in the name. Like we can stand on our own as Chalice come after building the category. So anyway, yeah, okay.
- [32:42]
C
And then what could we. Yeah, give me a bucket. Thank you for remembering. I actually forgot to press you on that. I need a bucket.
- [32:52]
A
Yeah, I mean you would say like custom AI for outcomes is how we've been talking about it, but now we want to depart from Outcomes because we think that term has been completely co opted and bastardized. Like I was talking to someone recently who was like, we're driving so many, like so many amazing outcomes. Like you wouldn't believe our click through rates. And I was like, okay, you're like,
- [33:17]
C
no guys, this is. Oh my God, it's a flat circle.
- [33:22]
B
Yeah.
- [33:22]
A
Or like the agency's outcomes business are really just like arbitrage businesses guaranteeing things like CTR and cpa. So I need to come back to you with like a really pithy category because now we would say it's like it's too technical. Right. But it's like custom models deployed in cloud. Cloud environments, which is the container but that's, that doesn't mean a lot to someone who wants to know the benefit of it. So we say we're just like AI, that's yours to brands. Like
- [33:58]
C
I do like models for me because I'm such a sucker for alliteration.
- [34:02]
A
So yeah, yeah, I, yeah, I mean custom AI modeling. But sometimes when we pitch a brand, they like, they really don't know where to put us on the map there, you know? Cause it could be like custom AI modeling about what creative message I want to drive. So we have to very quickly try and understand for media buying. And we're not a dsp, we replace it and we can work inside your Meta and TikTok too.
- [34:33]
C
It's like, yes, cool. What are you. People probably just keep. They listen, they nod and they go, yes, but what is the Pippy name?
- [34:43]
A
Yeah, I can't hook them more on the story of like a bear story. I try to tell them a story of what we've done that I think is similar to a challenge they have because I feel like if I can explain in the most plain words, right? Like you have a CMO who's trying to report up the value of marketing to a CFO and CEO. CFO and CEO don't care about your click through rates and viewability. They care about your brand share. And so we could build something that uses AI to drive up brand share. You will get the AI win with those people. Then they're like, okay, they don't especially need a category. They need the answer to their problem.
- [35:28]
C
That actually sounds like a tagline in and of itself. You don't need a category, you need a solution.
- [35:33]
A
Yeah, right.
- [35:36]
C
So I'm going to switch us over to containerization. I want to get very nerdy on this topic and talk about the containerized RTB strategy that you guys recently built in partnership with Index Exchange. Zillow was the pilot brand. We wrote about it. James Hercher and our team had a story. So the idea, as I understand it, is that instead of the typical back and forth between a DSP and an ssp, you have a machine learning algorithm that runs inside of Index Exchanges cloud infrastructure directly. So there's no latency. The algorithm gets to see the full bid stream, not just a sample and. Okay, so why does that matter? Help me understand the significance of that for someone who's smart. I think I'm smart, but I'm not an engineer.
- [36:24]
A
Yeah, me either. And I really had to push my co founder Adam on this because, like, how do I explain, like, why Would a branch. Well, so the whole thing started because Adam was telling Andrew and the CEO of a DSP how powerful it would be if we could predict at the page level, not just the domain level. And so because like you know, a Forbes article about cloud computing matters more to like Azure than just like Forbes.com and he really wanted bid control over that. So he was saying that to the DSP CEO and Andrew Casale was like, I get that data. You don't get that data from the dsp. And Adam was like no. And then Andrew and Adam sidebar and was like, well could we build something direct so, so we could action on that data? So that's like the net of it. But so why, so why URL level is like just like I said, I'll even step back from that. The more data you could hoover up and predict on, the better your predictions will be. And the industry is kind of obsessed with IDs and bidding on auctions where you know an ID, you know the consumer and you can predict on that. But that leaves so many other predictive factors that are also have value like on the table that most don't use at all. And so our thesis is we can use ideas when they're available because they are very predictive. But there's also all these. At least half of the auctions don't have an idea available and most brands don't bid on those at all. And there's like still a lot of value. And so if we can get even better and more granular at predictions, like we'll just have a better predictive engine. So, but so what that, so what that means and sorry. And then there's what you said which is like full access to the bid stream. So instead of getting you know like a 50th of the bid calls, like the DSP kind of pre filters, we could see it all. What that essentially means is like a lot more opportunities to find inventory that's undervalued but has value for a brand and like grab that inventory at a cheaper price than the brand normally would. And so that's what we're seeing bear out is like we, we predict at the URL level. So it's like much more precise prediction of value for the brand and then we're able to pay less than, less than what like crowded auctions demand and still find value there. And so there's just like to really make it shorthand, it's like many more opportunities to get savings and value out of it. And so that's what all that technology enables.
- [39:27]
C
It actually sounds like another way to talk about the promise of Programmatic. Like what programmatic promised from the beginning, but then, I know, kind of lost its way.
- [39:40]
A
Yeah, yeah, for sure. What we. Adam just published some papers on this this week. But like it was really interesting. So Meta was forced out of ID based prediction because of changes with Apple. And what they decided to do was like predict on everything else they could put into a model. And ultimately losing the ID actually made them a better predictive engine. I've read the exact number, but it's something like eventually they become 10% more capable of predicting than when they were only doing the id. Programmatic has gone the opposite. We just keep double downing on IDs and different ways to get them and different ways to match them. And that focus has meant that a lot of other predictive variables have been considered. Exhaust. Like we're literally talking about some partners in the ecosystem who want to give us data and they call it their exhaust data.
- [40:50]
C
But this is valuable stuff.
- [40:53]
A
Yeah, give us your exhaust for free. And they're like, yeah, you can have that for free.
- [40:57]
C
And we're like, great, give us all man's treasure. Yeah, yeah, right.
- [41:02]
A
It's actually, yeah, it has value and when you, it's none of it is as powerful as an ID on its own, but like if you build enough of an engine that's taking all of it into account, you can get, you could get potentially more predictive than IDs, but you don't even need to because there's so many. Half the auctions have everyone bidding on them because like someone's considered in market or valuable. And then you have half the auctions where like, you know, with essentially like nobody or there's very few people participating in the auction brands participating in the auctions and they still have value. Like you know, probably I would argue a lot of the consumers who have like blocked themselves from being tracked on the Internet and their phones are more savvy and might have more, you know, value overall to brands.
- [41:53]
C
So yeah, I mean that's often said of people that use ad blockers. They're actually, you know, they, they like, they'd like to be respected, please. And if you can do that, they won't turn the ad blocker on, but their tendency is there and if you don't show them respect, they will turn it on. They would be really good customers for you though because they're often ignored. But they, they buy products for God's sakes.
- [42:19]
A
Quality context, like for a Cheap price.
- [42:22]
C
So yeah, sticking with containerization for a second, you guys are also a founding partner of the agentic real time framework that the IB Tech lab put out there. The ARTF or the art, I guess Art. So the premise of agentic bidding is that AI models can travel across platforms without being locked into a single environment if it's in a containerized way. Do you think that this sort of architecture is really like the future of Programmatic? Is this how it's gonna be? Because it does make sense and it feels less just messy.
- [43:01]
A
Yeah, definitely. I mean, well, bears out our thesis that like every brand is or every brand is going to want an agent who's acting in the brand's interests. And so right. What this enables is a brand to bring their decisioning like directly to supply and that can essentially is through an agent who's like making decisions on their brand's behalf. So we think so we're even seeing like some DSPs for flirt with the idea of being, of essentially being able to offer what's equivalent to the RTIF spec inside of their platforms. Like it was started kind of as an SSP thing and now I think yes, I think ultimately it will be the technology to enable agentic buying. And it's just also like it's like a way to take all of the frameworks and pieces that the IAB has put together over 20 years and like make them available to agents. And so you know, our bet is instead of like reinventing all the programmatic pipes like with the Gentic is kind of the like tagline to, to do that we already agents just need really good instructions. The IB has made really good instructions. Probably not all of the instructions but like really good instructions or like you know, standards and like that bringing it all together in a real time framework is like what will enable Agentic fast
- [44:56]
C
and yeah, well, definitely tbd but like very quickly we'll have answers to a lot of the questions that are still questions now things are moving so super quick, including this episode. We are nearing the end and I have a penultimate question before if you're down to do it, I have a little lightning round in mind to close us out. Like real super short answers. But I wanted to talk before we do the lightning round about firing clients because you guys have a fairly hard line. Like you actively screen out clients who don't have the right data foundations. They don't have genuine strategic intent, they aren't open to change or experimentation. Like if someone just walks in the door wanting a better click through rate or whatever, you literally turn them away. And it's the sort of thing that sounds great in principle and great on paper, but then pretty hard in practice especially even though you guys are good now, but you know, you probably have some PTSD from like earlier on. Right. So, I mean, clients are clients. So did you have a particular experience that made you realize that you have to be tough like this and is it actually good for the business? Which I can see it would be because I imagine if someone isn't ready to like roll up their sleeves and really do this thing with you, they're not going to see good results. And then if they have a bad experience, they'll be like, oh, custom algorithms. I'm just going to use that. Forgive me. Like they, they don't work. Right.
- [46:30]
A
Yeah, yeah. Not just one experience, but accumulation of experiences where, yeah, we would say yes to something, but there were like little red flags that like, you know, we couldn't get really specific success metric. Right. Like, okay, like, you know, what, what is ultimately the goal? And if they're like, well, we really want like viewability above a threshold and we want sales, but like they can't define what even sales means. They're usually like a client who like, if they're not even going to commit to exactly what the success metric is, they like then don't commit the budget to actually like, you know, bearing it out or the time. And so how do you help them
- [47:19]
C
reach a goal when they're not sure what their goal is and then they're going to be like irrationally annoyed when you haven't helped them reach an amorphous goal that they haven't defined.
- [47:29]
A
Yeah. So, and like now we do take some business that's like click through rate or cpm, not so much clicks, but CPM based on. Because we're like finding all these savings through the container. But like I, and I would rather have a client who is like, I just want cheap, cheap CPMs. Then it's like, oh, we balance CPM with like blah, blah, blah, blah, blah. Because like if they can't tell me at least what's the number one thing they care about, like, it just like your models need to train to something. So. Yeah. And I just feel like could constraints drive clarity? So, you know, we don't have a, we have a 50 person team, which feels huge to me. But it's not that huge in the broader sense of things. And so, you know, we've, we've had issues where we burned too much team time on something that We've now learned, like, you know, it feels like real business or, you know, it's an even, maybe a bigger brand name. But the sign, warning signs are there that we're going to end up seeing spinning wheels. And like you said, they're gonna have either. They're gonna leave with the best impression and we, we should, you know, either wait until our tech is ready to do what they want to do or you and usually really it's like, wait for them to see if they really want to do what our tech can do. So, yeah.
- [48:55]
C
Not leave with the best impression. Ha. Impression. Anyway. Okay, so lightning round to close us out. Answers need to be 5 seconds or less. And single or two word answers are not only acceptable, they're preferred. So do what you can do. And I will fully acknowledge at the outset that these are not totally fair questions in that they're not necessarily single or two word answer questions. But I'm going to ask you anyway. So does the Google antitrust ruling actually change anything for advertisers in a meaningful way, or was it too little, too late? Five seconds. Only one?
- [49:41]
A
A little. Mostly for those who already were leaning away from Google.
- [49:48]
C
Nice. Oh, my God. Thank you. No one ever takes me seriously when I do a lightning round. They just, they give me like four minute answers. I'm like, guys, the most overused word in adtech right now is blank outcomes. I agree with you. It's like tossed around completely, like, oh, performance outcomes. Like, you don't know what you're referring to. Is curation something new or is it just another fancy way to say pmp?
- [50:18]
A
It's a new fancy way to say ad network. But I think there is a lot of really cool innovation that can happen in curation. But curation is like the delivery mechanism, right? Like, we use curation to just deliver the container stuff and that's
- [50:36]
C
totally fine. Is the programmatic supply chain actually getting cleaner or is everyone just pretending?
- [50:44]
A
I think it's getting cleaner.
- [50:47]
C
I agree with you. I didn't mean to have such a cynical bent on that one. Would you, would you personally let an AI agent shop for you autonomously?
- [50:58]
A
Only if it's like skincare and makeup and I've uploaded photos of myself to it.
- [51:06]
C
What's the best thing about working with your spouse?
- [51:18]
A
There's so many good things. It's like when things are going really well and like, it's like chemistry and lightning in a bottle. And it's like so cool to know my spouse on a different intellectual and adult level than like the Raising kids, and even, like, life we had prior to kids.
- [51:39]
C
And then what's the worst thing? Lol.
- [51:43]
A
We bicker like cats and dogs about marketing materials. We both think. He's like, this is pandering. It's. It's not expert enough. And I'm like, this is so expert. 98% of people don't understand. Like, and we usually actually come to a great middle ground. We get to the lightning in a bottle, but it's. It's a lot of bickering.
- [52:07]
C
Say containerization. Five times fast.
- [52:10]
A
Containerization, containerization, containerization, containerization, containerization.
- [52:14]
C
That is very impressive because I tried to do it before the recording and I messed up on the fifth one. That's true. You got to know how to say containerization. Last one. If chalice were a custom cocktail, what would be in it? I was so proud of myself for that one. I'm like, that's funny. Gotta ask her that.
- [52:37]
A
Good one. Well, let's see. Custom cocktail. So we would put in some, like, amaro, some rum, and, like, a cherry. I don't know. Those are things we like to drink. And yeah, it's got, like, a lot of texture and flavor to it.
- [53:02]
C
So the next time I see you, I'll raise a custom algo to you tomorrow. And rum and a cherry on top. Thanks for the time.
- [53:12]
B
Cheers.
- [53:13]
A
Cheers.
- [53:25]
B
This episode was sponsored by Verve, an ad solution that helps brands activate consumer intent in real time across platforms. As AI reshapes how people plan vacations. Verve helps travel brands show up when it counts. Find out more at www.verve.com. that's V-E-R-V E.com.
- [53:49]
A
Sam.