Loading...
Loading...

This podcast is brought to you by The Build, a new podcast from the guys behind
Sincera, a Microsoft Sullivan and Ian Myers. They built their company by
figuring out clever solutions to a few important and tech problems in our
industry, and that's exactly what the show is about. Mike and Ian interview some
of the smartest tech lines in the biz to hear about how they identified
opportunities, solved their hardest challenges, and grew their businesses in the
process. Listen to The Build with Microsoft Sullivan, wherever you get your
podcasts.
Welcome to the architecture podcast. This is Eric Paparo. I'm here with Eric
Francie, and we're both tanned and relaxed and back from possible. Maybe
relaxed isn't a good word. How was your possible, Eric? It was a whirlwind. I
had, I think like average of like seven or eight meetings a day. Wow. So yeah,
I was, I was, you know, pretty, pretty busy. I think it was a fantastic event.
They say over 7,000 people were there, including, you know, we're
excluding all the, all the people that that just show up. I think, you know,
despite spilling into Eden Rock for another location, it still feels like they
need more space. And every surface is sponsored. I mean, my goodness, that
commercial just sells. It was incredible. Everything was sponsored. Yeah, it was a
great event. Christian told me 7,500 people. They announced they're going to do
Europe in 2027. So there's going to be a Lisbon show in October of 2027. I
think it's still a little tentative, but they did announce it. And yeah, they
expanded to both Eden Rock and the Soho brand on the other side, or Soho Beach,
just lots of happening. So house, yeah. And you know, it's a tradeoff because part
of the charm of possible is that it's not in a convention center. It's in a, you
know, interesting, elegant hotel. And the side effect is that the logistics
aren't great. You're walking on the beach to get to the meeting. And then you
have to go to the street. And then, and then you're in some weird, you have to get
ID to get into Soho House. And it's just kind of complicated. Yeah, agreed. I
have, I have two questions for you. Maybe we'll do the first one, which is you
had a lot of conversations, I'm sure. What were the key themes that you pulled
out in terms of what people were talking about? What are the priorities? Well, aside
for the obvious AI stuff, I think my sense was the industry is so very clearly
maturing, solidifying. There are fewer questions. Everyone knows who the vendors
are and what their positions are. And those vendors, the kind of common names
we've been hearing for years, I think they're solidifying their leads. You
don't see as much jockeying about, you know, who am I? What am I going to be
in the future? You know, people know who their trusted vendors are on the
data side, the transactions side, measurement side. So I think it's like a
consolidation of strength within the, within the confines of, you know, what
frankly is a leveling off or maturing of the overall, you know, open web ecosystem.
Yeah, interesting commentary. Funniest thing I saw a week was, for those of you
who've been to possible, the Soho, Soho House is in kind of a weird little corner
of the street where it bends in a certain way and has a very, very sharp, turned
driveway. It's like almost a circle, the driveway. And it is not a circle. It's
like a, it's a triangle. It's such an extreme. Yeah, it's really tough. Yeah,
yeah. It's really tough. Well, Miami is experimenting with self-driving
Waymos and Waymo got stuck in the driveway. Like it was like if a squirrel got
in your house, it was just freaking out. It was just like moving back a foot,
four to foot, and the Soho staff had no idea how to handle it. They were like
waving at it and trying to give it instructions. And that was making the Waymo
even more, you know, agitated. And I think they may have a no waymo policy
moving forward at the Soho because there's no way those things can get in and
out of that driveway. It just needs more data. That's hilarious. I'll say the
funniest thing that I saw and then what we can get to, you know, in some of my
takeaways, was there is a new ad format or a new ad surface. It is a helicopter
dragging a screen. Yeah, yeah, I didn't feel good about this. Yeah, okay, keep
going. I spent a fair amount of time at the innovation stage. We hosted this
brand challenge. It was awesome. It's a startup competition. And there was just
thunderous helicopters just going back and forth. One was was carrying a
screen that was lit up with some marketing message. And then one was I guess
kind of flanking it for security if something happened. So the combined noise of
two helicopters at once was like really tough, but I'll tell you what everybody
looked at this freaking screen because they're like, what is that? Yeah, not the
best use of carbon. I don't think we're going to see any carbon neutral ads when
you require two helicopters to show your ad. So Adtech God had a banner on the
helicopter. I don't know if you saw. There's something else. Yeah, yeah. Yeah, in
terms of my takeaway, it's an interesting observation what you said, you know,
specifically relative to our corner, the ecosystem right adtech and open web.
Every conversation that I had, you know, obviously, you know, AI being front and
center is kind of two things. Number one, everybody is focused on performance and
measurement right now. And you know, you know, sort of measurement of outcomes, I
think was, you know, kind of front and center, which was interesting. And then
too, it also feels like this is a continuation from last year. People feel like
there's opportunity. People feel like even though, you know, there are established
players, like in this age, things can shift pretty fast. And I point to a
variant who now with these acquisitions of TV, TV and Iris TV, like they're
telling a completely different story and they're able to measure things while
gardens that they weren't able to measure before. And that's, I don't want to
overroute it on vines to just one example. Just feels like there's a continued
sense of optimism that like, you know, the chess board might be laid, but there
can be a lot of reset in which it was great to see. Yeah, I wouldn't disagree
with that. The other thing was my article from a couple weeks ago, it's the data
stupid. You know, basically half the people who don't have data came up to me
and were angry about it. Because like, I do have data. You said I didn't have
data. I do have data. And the people who know you, yeah. And the people who do have
data all came up to me and like, I sent it to all my team. That's kind of fun.
It's total libous test on this stuff. So today, we're going to do things a
little bit differently since we, there's so much news and we're coming hot off
of the planes back from Miami, Eric and I are going to just dive in on the
news of the week. And then at the end, we have a interview idea that
possible with Andrew Casale to talk about his recent announcements of their
their system for hosting bidders and hosting data in their cloud. I went into a
bunch of depths on that technically about how it actually works because I had a
lot of questions and how it relates to the IB standards about ATRF. So that'll
come at the end of this interview. And for those on video, my conversation with
Andrew is audio only. So we'll, so please bear with us if you're watching on
YouTube. Otherwise, you should come in loud and clear for the podcast listeners.
So with all that said, let's dive into earnings AI and a bunch of other stuff.
Absolutely. Before we get into earnings, you had two things this week that were
very important and very newsworthy. That I would just like to quickly conquer.
The first was you asked if people could stop hugging and the second was
you asked if people would stop wearing blazers or come up with an alternative to the blazer.
You had a lot of feelings around some, I think, very relevant conference topics.
Okay, let me dive in. So I tweeted that we would need to normalize not hugging in business
context. And I feel as though I was being an ally to my female friends who are inundated with hugs
from random guys and blazers. They don't really know very well because somehow it became standard
that if people are of opposite sexes, they need to hug and save handshake. And I think it's
disrespectful to professional women. So I would like to normalize handshaking unless I
actually know you very well and have a warm relationship with you of which there's maybe
half a dozen people in the universe. So I mean, Eric, you and I are hug bases, but that's about
it. So that was my point. I don't disagree, by the way, with your disagree. But for some reason,
people hug me. I know they come in for the hug every time. They come in for the hug.
And the thing about my tweet is from the women in the audience, I got a lot of thank you.
That's a really nice of you to say. And from the men in the audience, they took it as a challenge.
Like they're coming in like they're coming in like Lawrence Taylor on a quarterback, right? Like
and that just says everything you need to know about the mentality of these these goons work at
ad tech. Oh my god. Now on the blazer thing, I did not say that. That was a comment to my that
was someone jumping on my train saying, oh, we should get rid of blazers too. I didn't say that.
And you know, my only point was like, if you get rid of the blazer, you got to replace it with
something else because a man's middle age body is just not suited for full disclosure.
You know, the blazer is a nice like a sheath. It hides the the lumpy bits. And you know,
you either got to go sweater or something else, right? And in the Miami heat, you know,
you're generally doing like kind of the button down short sleeve amorphous top, which I think
is totally respectful, respectful. I got some hazing for wearing a blazer. It was hot.
I wore a blazer. It was hot. I didn't do it every day. But I need I need pockets. If I'm not
if I don't have a blazer, I need a bag. I don't want to carry a bag. I have glasses, hand sanitizer,
gum, notebook, pets, AirPods. Like there's just a lot of stuff when you're out all day long
for practical purposes. So so I was he's trying to get me to go the merch route. I'm quite resistant.
We'll tune in for a can. We'll see how how this takes us up. So what's he's got the the
mercs and the man bun, which is, you know, the combo. He does kind of have a bad bud. Doesn't
he could? You're probably nice. I think I saw a man. All right, move on to earnings. All right,
let's go. So good transition there. In addition to possible, the four basically most important
companies in tech all dropped earnings last night. So that was the late one for your boy.
Let's walk through this stuff because they're super interesting. So first was, I met not first
though, dropped out for 30, but first was meta on our list. Big quarter. So 56.3 billion in revenue
of 33% year-over-year increase in ad impressions, increase in price per ad. But due to AI,
CapEx guidance raised from 125 billion to 145 billion. And I think that had
an effect on the stock. But meta continues to crush it and continues to, I think, be the leader
in showing how AI can drive performance in ads. Yeah, I guess the only, if you were evaluating
the stock, the one question you would have is like, if the consumer side was ever disruptive,
if consumers for some reason stopped going to Instagram or started not to, it would all fall apart,
right? And so that's kind of the implication of it not being 99% advertising.
Yeah, I didn't get through the whole transcript, but Zuckerberg was talking about consumer
interaction with AI and consumer interaction with other AI-driven surfaces, like glasses,
for a lot of it. So I think he's already thinking about that world.
Yeah, I think they're always, because they are intermediated on their platform by Apple,
and obviously they have some really tough, they've had some tough times based on that,
Zuckerberg is sort of obsessed with leading on whatever the next platform is. And he tried
virtual, now he's trying glasses, AI is a new platform in a sense. So he is trying to get ahead
of that, but the history is that they haven't been successful doing that. Yeah, but they're successful
ads. Two other things on Meta before we get into Google. So first was story and digital day that
they're considering pushing into CTV. This was something you predicted in our year and episode.
It seems like a no-brainer and, frankly, the way the industry is going and goodness, the return of
fan, but it might be controversially called man. Yeah, in my understanding,
this is sort of a no-brainer for them. They've, fan has been mixed results. They obviously
pulled out of web a long time ago. But the ability to bring small and medium-sized advertisers
globally to the streaming world is a huge opportunity. We see even Walmart just announced
they're in this business of bringing these advertisers to TV. And Meta is just in an amazing
position to do this. And you may say, well, they don't control the inventory, et cetera, et cetera.
Yeah, but it's just such a high CPM high margin business. It's crazy that they're not playing in it.
Yeah, or they can buy something. So one to watch, the other significant one on Meta was China
ordered Meta to reverse the Mattis acquisition, which was a $2 billion deal. It's been four months
Meta is already talking about how they've integrated Mattis and it's driving good results with AI.
And the decision has now been elevated to Xi's national security commission. So this is significant
from an AI perspective, but then maybe not significant in terms of what could happen if,
you know, they decided to pull out of China, which is for pointed out. Yeah, so a couple of things
here. So this deal already closed. People have the money in their bank accounts and their
investors, it out. So it's a big implications to the venture world and to the Chinese venture world,
very negative implications there. Manis has already been utilized for a lot of ad-related
use cases in Meta, but on the other hand, just yesterday, Meta announced their official MCP
server. So it's not, Manis is not the only route by any means towards automation. Advertisers now
can use cloud or whatever they want to access the Meta advertising system, which previously they
weren't allowed to. There were some cases of accounts getting banned. So I don't know that this
has a direct effect on the advertising business. I think Manis was a really cool investment for them
and they were excited about it. And they're certainly not happy about it being reversed, but I don't
see any immediate implication on their advertising business. Yeah. And Sufer pointed out, which I thought
was super interesting that because of the second price auction and the diversification in the
number of advertisers and the number of scammers in China, even though it's a $20 billion market for
Meta, they decide like, hey, it's dangerous for us to be operating here sustainably. It might not
impact the business too much. Yeah, but he does point out that virtually all of their hardware is
built in China. So what we said previously about Zuck investing in glasses, et cetera,
is in a lot of jeopardy if he gets into a fishing match with China. I doubt you will then.
Okay. Let's talk about Google. Google, 109.9 billion in revenue, 22% in year-over-year growth.
Cloud, I think I put 20 exclamation points in the document here of 63% to $20 billion,
printing money. Search is up 19%. So much about Google losing search. YouTube ads slight miss
at 9.88 billion. The only thing that was down, and it was like, oh, by the way, because this is just
so small, is the network revenue little old ad tech down from $73 billion to $7 billion. Their
CapEx fork forecast is monstrous as well at 190 billion. Cloud is mind-bidling up 63%.
Yeah, this is, it's got to be AI, right? I didn't listen to a transcript, but yeah. So you get
Gemini in Google Cloud services. GCS is considered at par with Amazon at this point, I think, for the most
part. And so, and Amazon is selling other people's AI, not its own AI. So I think there's a lot
of a lot going on here. And I don't think we have it in the notes, but Microsoft and OpenAI
just renegotiated their deal so that OpenAI is going to be available in other clouds. So I think
that potentially is kind of the break that Amazon has been waiting for. Yeah, Andy Jassy had a
post on X. And it was very intentionally. It says something around interesting news.
So we'll see what that does. Yeah, well, we'll take the next view. Again, continue
of the story. Microsoft 82.9 billion revenue of 18% year-over-year. Cloud up 29% to 54.5 billion.
So again, this is AI spent. Amazon 181.5 billion up 17% AWS up 28% to 37.6 billion of 15
quarter high adds up 24% to 17.24 billion. So we see where this stuff is going. It's going to
the cloud. It's going to AI spend. Yep. Yep. It's these companies are with the exception of meta
who is only ads. These other companies have all have two horsemen in the race, the cloud and the
ads. I mean, Amazon obviously has retail. We don't really care about Amazon retail, do we?
No, I mean, I try to think something. It's got the retail footprint to run ads.
I guess for this podcast, but it is just monstrous. These companies are getting bigger.
All right, let's talk about some announcements. Big week of announcements,
as possible, we won't nearly be able to get into all of them, but a few things that I think show
some signal on where the world is going. So in no particular order, Pinterest, who closed the
TV scientific former investment of ours at a period in February, officially launched Pinterest and
TV scientific. So they're integrating the Pinterest data, which is like super interesting data.
I think it's like very valuable. It's like mid funnel of 600 million users off platform
into CTV. So not unlike what you were saying with meta, which is just this obvious opportunity.
They're doing early tests. They point to a 27% lift and outcomes per $100 spent.
So it's like really exciting stuff. Yeah, I get briefed on this. They took me to breakfast,
which is nice. Sometimes people, some people times people pretend I'm a journalist and I
appreciate that. So Jason Fairchild gave me the briefing basically. So this is to be clear,
Pinterest data into TV scientific, not the opposite. No, no, I just want to be clear because I think
people can think this is like, oh, Pinterest advertisers can buy TV. And that's not really what's
going on. It's a TV scientific advertisers can use Pinterest data. First step, I think it's
telegraph that the opposite will be true. So Pinterest has an ad manager with hundreds of thousands
of advertisers. At some point, they could buy TV, but that's not what this announcement is. This is
kind of a first step. Yep. Yep. The other one that I thought was interesting, related, not
necessarily pure play on CTV was PayPal ads. So PayPal launched their ad ID. So it's a
deterministic ID built on 400 million PayPal and Venmo accounts, like 25 billion transactions
per year. And it's free to partners. And they launched on, you know, like usual suspects,
magnetic, pubmatic, taboo, and so on and so forth. I think this is a good one to watch as well as
you kind of think about the new forms of data, commerce being maybe one of the most valuable.
Yeah. It reminds me a little bit of when live ramp made their ID free to partners,
the business, business strategy being you use the ID. And then when a customer wants to,
you know, target or, or onboard or use attribution, that's when the fee comes in. In PayPal's case,
I don't know if they're going to charge a fee on a use case more so they're likely to
sell media to their existing huge base of customers.
Two similar announcements. One on the buy side, one I guess, you know, on the sell side. So
TTD launched basically integrations in commerce platform. So Packview where I'm on the board
and Sky, also a commerce media company. So TTD DSP is now manageable within these other
interfaces. These are I think going to go live next quarter. And it's it's cool. It connects
like upper funnel programmatic alongside retail and search and wall gardens. I love this.
Yeah. I think it's I kind of am a little skeptical. It's like incremental. It's like one UI and
you get the campaign side by side, but they aren't in any way de duplicated and they can't be
because the channels that Packview and Sky work with don't provide back any user level data or
impression level data even for that matter. I don't believe as a result, you're not going to have
you know, frequency cap adjustments. I don't think my guess is you don't have budget adjustments.
You're really just seeing inside by side. And I've seen these sort of partnerships before.
Nice. Not game changing. I'll push back on that in one way, which is these are being exposed to
primarily like performance and lower funnel marketers who are looking to grow into made an upper
funnel. So I think this is actually quite good for TTD because it gets an exposure to a customer
base that is just thinking and buying. Yeah. With a different strategy than their normal kind of
fortune 500, fortune 1000, upper front centric marketers. Okay. So you see this deal as a
TTD customer acquisition strategy. Yeah. Or growth strategy. Okay. That's interesting. That's
not the lens I was taking. I was thinking about it more like a workflow enhancement for common
customers. Yeah. And it's at its best. I think I think it's growth. And again, I'm kind of
looking at this from a pack view perspective where I'm on the on the board. I'll get some
answers to those questions. I'll come back on it. The other one similarly was Walmart. Walmart had
a couple of announcements around connect select, which is, you know, we'll talk about connect select
first. So it's basically a curated CTV marketplace in Walmart, DSP. It's got inventory from all of
the OEMs and streamers and as well as SSPs like Magnite index, Pubmatic and it's targeted at SMBs.
They're going after the universe leds, the TV scientific, the vibe, the mountain.
So many everybody going after this market. And meta too. Yeah. It's a it's a it's a it's a full rush.
God bless. It makes sense. The Walmart would do this. I'm rooting for Walmart a little bit. You
know, I think Amazon DSP and Walmart DSP should be peers. And you know, Walmart's got a long way to
go to get there. No faults of their own. It's a tough market. But I'm rooting for them. Yeah,
don't forget they have Visio, which I think over time, you know, if they if they execute could be
the real monster here. The other thing was Walmart also, you know, basically had the same announcement
as TTD in terms of integrating into a pack view in sky. These are, I think, a little bit closer to
the Walmart universe of shared customers. But I think it's just interesting how DSPs are
integrating into these commerce media platforms. Yeah. I think it's natural because you've got
the open web. They traditionally bid in is flat. So you got to go where the money's going.
Yeah, which is commerce media. So other one on the just like kind of feature and platform
rollout that thought was worth touching on was mag night. So they rolled out a whole suite of
agentic solutions to first was like a spring serve AI, mediation and buyer agent, buyer agent
rather. So they're doing tests with Kepler and MIQ using Disney inventory. And the other was,
I saw Matthew running around. He was super excited. A lot of the features that they picked up with
streamer around CTV, creative generation for SMBs are becoming built into the platform. I think
alongside the buyer agent. So mag night is running hard at AI. It's really impressive.
Or this is obviously still wearing that hat or as he switched from mag night hat yet.
No, he wasn't wearing a hat. He was not wearing a hat. Oh, usually there's a bright, bright pink
hat from, I think that's retired. It's retired. He's, he's SDP of product or sub or SDP of AI product
for, for mag night. Yeah. He's become respectful. Um, so yeah, the mag night,
mag night announcement was impressive in its brev. Um, you know, they didn't just do one AI.
They did a whole bunch of them. Uh, but I, I reiterated what my thesis on Twitter about this,
which is that it's again a single point AI similar to what Pubmatic has announced. So they say
it's buyer, but it's really buyer on their platform. We're still not seeing any, any investment in,
AI wear a buyer agent and a seller agent who are at arms length or negotiating.
It's, it's still single point customer to destination. Um, and I believe that's a much more
promising area to invest. Agreed. Agreed. Also quite complex. Um, and the final one was just
a monster fundraise. Yeah, I had a nowhere. So yeah, it's much, which is in CDP, which is a
category that was super hot. And then, you know, kind of went, um, dark for some time raised 150
million at a 2.75 billion dollar valuation. Um, there's a series D led by golden sacks and
Bane, TD seven, which is a trade desk. BCR invested as well. Um, and if you look at the story,
and I encourage everybody to look at the site, high touch AI, um, you know, they're telling the
story around what can happen when you activate customer data using AI from a creative standpoint,
from an activation standpoint. The story is very compelling. And it's a, it's a big raise.
Yeah, huge. Um, they were not on my radar at all. I, I, I'm trying to get their CEO on the pod
in the next couple weeks. So we'll hear from them, assuming I can book it. Um, yeah, their pitch was
not CDP. It was like, Hey, we're CDP, but really were agents of AI, which I think, you know,
sign at the times. Um, I'm in true. Um, I'd love to hear the backstory on how they got that
valuation. Yeah, exactly. Well, this year, according 100% growth, you know, year for year
for two years running. So the biz has to be, you know, executing at a high level. But, um, I,
this was a category that I paid a lot of attention to when it was emerging and hot.
Most of the competitive set has been acquired and you know, no longer independent names. So maybe
this is, you know, the, the one that's about. Yeah. I think the CDP world God, um, God,
disintermediated by snowflake and others where it was, you know, database plus a bunch of features
and people had the composable CDP, um, which really meant paying as little as possible for the
full package. Uh, and, uh, it seems to have worked. Um, so there's nothing wrong with collecting all
your customer data in one spot. You just don't necessarily need a single vendor to do it all.
Yes, very true. Um, you, you saw something on UCP you added to the socket. What's that?
Yeah. So UCP for those long time listeners is the retail AI protocol that Google announced right
at the beginning of the year. Um, and at the time, it was, uh, unclear if it was going to win or
what was happening because open AI had released a similar but different protocol, um, earlier in
2025. Um, and now it's unequivocally the winner because we had this week a list of people
including Amazon and Microsoft and Salesforce and meta all adopt UCP. Um, so UCP is the future.
What it is is a way for, um, effectively checkouts and discovery of products to work inside AI.
And it is, it kind of balances the merchants desire to control the customer and control the
experience with the AI providers needs for, you know, access to check out, you know, APIs and things
like that. Um, so I think everyone is pretty bullish on this. The consumer behavior is still a
little bit up in the air as to how they'll shop using AI, but it's great to have the rails set
and for everyone to agree on how they should work. Yeah, seems big. Um, with that, should we get into
the interview? Yeah, absolutely. So this is a super interesting interview. It's short. It's me and
Andrew on a couch using using my phone voice memos. I hope the audio is good. Um, and I even
stump them on a question. So I ask them some pretty tough questions about how this all works is
containerization project. Uh, is it the future of the way DSPs work where they'll be, you know,
side cars to the SSPs or is it just the same old thing that other people have been doing for years?
I hope you enjoy that interview. Awesome. See you next week, everybody. This year.
All right. This is Ari. I'm here at possible with Andrew Kassali, the CEO of index exchange. We're
at the beautiful suite on the second floor of the Soho House. And we are here to talk about recent
announcements from index exchange. So Andrew, how are you doing here at possible? Thanks,
Ari. Uh, doing great. And it's hard to complain being in Miami and South Beach. Uh, whether it's
great and it's been a flurry of activity so far. So you've been pretty busy. There was a press release
that went out earlier about, um, sort of your containerization project. And the crux of it is
that folks can build inside the index exchange cloud that can deploy the bedrock platform was your
first partner and they're deploying effectively a DSP inside your cloud. So I want to hear first,
like, what is it? Why should we care? Uh, well, that's a great opener. Uh, I think the first
thing that I'd say is that this is not that new. So, uh, we're about two and a half years into
the vision now. You might recall, uh, the, the press and the announcement we did with Chalice at
that time, which was really the first time that we stood up code and a container on top of index.
We didn't call it the index cloud at the time because it was so new and novel. And we also weren't
certain that it was going to become a thing, but we were really excited by it. But since that point,
we've also deployed many other partners, like empowered, um, cybeds, all in the custom bidding
algorithm space, also a company called Nano. And then more recently in the fall, we deployed a
data provider, um, Grace note, uh, who brought their show and episode level data into a container as
well to provide that metadata at the edge. And then most recently last week, an additional
application we announced, which was the deployment of the first containerized bitter instance from
bedrock. And so now we have multiple iterations of ad tech running and we've also come a long way
in building out support across the cloud. So that's the announcement. I want to dive in a little
bit for our audience. Oh, what it means when we say containerization, uh, because having data in
the cloud next to a bitter next to an SSP is not new. App Nexus was doing it in the 2008,
nine timeframe, uh, at beeswax. We had a thing called augmentation where we had partners who
would put data near our cloud, but it wasn't really a package as well. So take us through why this is
kind of new. Yeah, it's, it's new in the sense that we are effectively running our partners code
on our compute. And I don't think in the examples you cited, it was that I know for certain that
wasn't that with beeswax. Uh, there's a bit of a difference between near and on. So we also have
a notion of real time data providers. We've had that for many, many, many years. And that's, um,
the opportunity that partners have to receive a real time request from the exchange,
I would vote five milliseconds to respond and add data and append data. The challenge with that
integration point and any other like it is the requests are still going outbound to the exchange
to either public cloud infrastructure or our partners infrastructure. Their compute is required to
look up and associate whatever they're doing with the request in a very, very short time frame.
It's expensive to do that at internet scale. The novelty here and the reason why we're pretty
excited by what's happening is when you reduce the amount of compute you need just to one
enrichment or look up, it's actually quite nominal compared to the amount of compute required
to adjust the entire fire hose and do it at scale. So we're radically reducing the cost of compute
to do decisioning data lookups and now bidding of the edge. And that's what's new. Right. So, uh,
I could speak from experience with beeswax. The way we did it was we told our partners,
you have to be in the same AWS region as us, but the rest was up to you. If you wanted to scale up
how many machines with their spot or reserved instances, how much CPU, that was your problem as a data
provider. So it sounds like what you're saying is because it's your compute and your container,
those problems kind of go away and the partner can just worry about the code. If you could respond
to that and also tell me about the economics, like if I was a new data provider and I wanted to be
in your cloud, what would my, my bill really look like? Yeah. So there's, there's time you've kind
of hit at the car, so there's two big points. The first one is the ingest and egress problem,
that ingress and egress, I should say problem is gone. So if you want to look at the entire
fire hose, we don't have to transport that data anywhere, which is a huge expense for a bitter
or any one of these applications. The second though is the speed is very real and in our world,
speed is a corollary to the cost of compute. We're not publicly revealing it, the speed of
this bitter, but it is blazing fast, which massively reduces the cost of compute as well.
To answer your specific question though, we don't charge for ingress or egress as you
traditionally have in a public cloud environment. We do charge for CPU cores, but what we're able to
get out of each core per like unit of QPS is insanely high compared to the benchmarks that we've
compared this to with partners, we're pushing the same QPS into either their own on-prem instance,
that's co-located in the same data center or to the own public cloud instance for their doing
this in the public cloud. And that's really the additional bit. We process the request so fast,
we can get more out of each CPU core. That's pretty interesting. So I want to step back a little bit
because the IV released a standard around this called the agentic real-time framework. I think
and then people like Adam Heinlick started adding me about why I was an idiot because I didn't
take it seriously and why wasn't I going to write about it. So this is officially my writing about
it. So two part question. Part one is this related to that, is this built on that? And the second
question is, can we really get rid of the A, the agentic in this acronym? What does this have to do
with the agentic? Happy to. So I'll take that in two parts. The agentic real-time framework
is what governs the standard behind the custom bidding algorithms or the optimization vendors that
really started this containerization party and then also more recently data vendors with the
specific announcement last week with the bedrock bidder. That's using OpenRGB. So the OpenRGB
protocol is what governs a bidder. You can almost like a containerized bidder is the same thing as a
standard bidder in that we send a bid request with a standard bidder to an endpoint on the public
cloud or their own infrastructure. In this instance, we send the same bid request but to their
container, which is cryptographically signed that runs their code. Otherwise, it's effectively the
same request and request response out as you would typically see with the DSP. With respect to
the ARTF and the word agentic, my comment there would be, we're not done yet with where this
vision goes. And so I'll give you a window into where I think a lot of us think things go next.
If you take a typical bid request bid response of the DSP, which you know around 200 milliseconds
in and out and you shrink that down dramatically into a world now where we have models at the edge
that respond in five milliseconds, you have bidders at the edge which respond very quickly. We'll
announce when soon. You end up with a lot more time. If you then have an ecosystem of more data
providers and more models, you get to a place where there's an opportunity for orchestration
in that very, very, very brief moment in time to make really cool new decisions. Instead of calling
one model, maybe you call three. Instead of calling one data provider, maybe you call 200. Instead of
calling one bit or maybe you call them all. And maybe they can all be orchestrated together. This is
where we think things will go next. And so I would say that the agentic part of ARTF is yet to come.
But I also argue the agentic side of what everyone is preaching in programmatic is also yet to come.
But I'm pretty sure that the agentic version of the market that is coming will be in press
and level, will be super fast, and will be born in this protocol. That's my hot tick.
That's pretty hot tick. I'm still objecting to the A in the acronym. So for those of you who don't
aren't super familiar, Bedrock is a sort of a next gen DSP founded by Shane Shevelyn in the UK.
I think their inclusion in your announcement will kind of raise my eyebrows a little bit,
because I think this sort of edge computing around DSPs and SSPs has been very commonly used for
data. The idea of augmenting data so that you could do curation or you could do filtering,
et cetera, is pretty common. But bidding in a SSPs cloud is a new thing. And so I want to talk
about that for a little bit. So obviously a bidder inside your cloud has some advantages around
latency and around sort of inventory access. But if you fast forward and you said, well,
what if say a Bedrock or another DSP did this with multiple exchanges, because there are other
exchanges out there? Last I checked. The problem, you start creating new problems, like how do you
coordinate budget between them? How do you create a frequency cap between them? And I wonder what
your opinion is of that trade-off for buyers. It's a great question. That was one of the key
pieces of the original architecture before this bidder could even go live, which is Bedrock
does bid into other exchanges. They do have their bidder on the public cloud. They were previously
integrated into index that way. So part of this design was, how do we maintain that construct
while also enabling the bidder to live in the cloud and gain all the benefits of it? And so there
is an orchestration layer that they designed that allows the bidder to effectively do what bidders
do today across the region. So if you think of, you know, you know this well, any DSP who has
infrastructure on say the West Coast and the East Coast, they actually have the same problem,
which is campaign information, frequency capping, and pacing need to be seen across coast.
Technically, the problem is a little simpler because we're a lot closer to their infrastructure
than the West Coast. The West Coast is 70 milliseconds away. We're 10 milliseconds away. So we're
effectively being synced in the exact same way. And we could see this scaling quite well
to a handful of exchanges. Will it be every exchange? Probably not. But for every exchange,
their base bidder will still be able to participate. And so effectively, the same orchestration
that allows the DSP to sync all this information across their endpoints is being reused to enable
this index cloud instance. Yeah. If you want to give a add tech executive, especially in the tech
side, PTSD, just say, how do you think East Coast, West Coast? Because I could go off. I could
have a whole podcast on this question because, you know, Europe, US is easy because you don't have
any of the same users, right? The same person is not in the US and the Europe at the same time.
But if you're in Kansas, you might be in East Coast, West Coast at the same time. And it's really
unpleasant. I'll get off that hobby horse for a minute. So let's talk about other aspects of
this around privacy and whether this is better. I'm kind of giving you a layup here. But like,
it seems as though moving data around less is good. And I wonder if you have more deep thoughts
than that about this, especially as it relates to Europe and their laws about moving data into the US.
Yeah. This was a huge part of the original vision, which is the bid request never leaves the
building. It doesn't go on the internet and it doesn't even leave the exchange, which is
inherently more secure. Nothing is on the wire. But the other hope was that over time this would
also allow for an autonomous bidder to be able to leverage more data that historically publishers
tend to hold back from programmatic. So we have a lot of publishers with great first-party data
that they'll never put in a bid request because they don't want to grow someone's graph. In this
particular integration, data can be made available for a bidder that can understand it and use it
and the data never leaves. So I also think not only will it be more secure, but it will allow for
more over time, liberal use of publisher first-party data in a secure construct. It'll ultimately
take time to get there. But I think as publishers start to understand this design,
they'll see more upside-to-make their data available than to hold it back compared to the
original concerns of enriching a third-party's graph. I want to follow up on that. Can your current
architecture specifically restrict a member of your cloud from logging certain data, be it
user data or publisher data? This is sort of inherent to the way curation works. So in curation,
you're not directly sharing data with a participant or a bidder. And so we see that same model
applying here, too, where if you take a publisher with a great dataset who has packaged it
into effectively a deal idea or something like that, they're not directly sharing the audience
data or information that created that package, but they are making it available. What we hope,
though, is in this scenario, we'll be able to not just do that, but we'll also be able to potentially
represent the data in a non, let's say, human readable form, which will lend itself to also models,
because the modeling companies also have the potential opportunity to use a lot more points of
signal to make really smart decisions. And we think this integration will also lend itself to
that. So just think of it as slightly non-traditional approaches to solving the problem,
become a lot more possible in this environment. But to be perfectly frank, like this is not invented
yet, we're still walking this path, but this is where I think it can go. It seems like this would
also enable a movement towards Tee trusted execution environments. We had Jim Payne from CloudX
on the show a couple months ago, and that's one of the innovations that he's trying to bring to
the in-app market, where it's very opaque for those who don't know Tee, sort of a cryptographically
signed way of proving what happened inside a black box around the auction mechanics. Have you
thought about that at all? This is one of the few topics that I ultimately have to point to our
CTO. Ray has been thinking and talking about this a lot, but I would mature it if I even tried.
That's totally fair. Unfair question. I went for double jeopardy on that one.
Okay. So in some, like, Payne has a picture for where you think this is going kind of in the long
term. If you were to prognosticate three to five years out, assuming programatics still exists,
what do you see this cloud looking like? I think programmatic won't just exist. I think it'll be
bigger and better in three to five years than it is today. And I think it's innovations like
this. It'll make that possible. I'll give you a few versions of the future that I see.
The one obvious one being, I think, very few bid requests will ever leave the exchange in five
years. There certainly might be obvious cases, big giants that will have their own infrastructure.
But I think the vast majority of bidders will containerize because there's a huge business
benefit. If you're number two line in them and your company is the public cloud and we can radically
reduce that cost, that's a lot of capital that you can realize back into your business.
You know, it's been one day at possible so far and I have had a lot of meetings with DSPs about
what it would take to containerize their bidder. And there's a lot of interest in it because the
costs are real and they're rising. The second one is what kicked off the portability of models of
the custom bidding optimization party was about five milliseconds in the hot path. Imagine a
world in the not too distant future in a few years where instead of giving a custom bidding
algorithm five milliseconds, if the bidder is so fast, we can give the algorithm a lot more time.
Instead of the algorithm meeting, you know, 12 features in a tight memory store to be able to
hit that response time, it can have hundreds of features in its store. It can use a ton of memory,
it can be slow, but it can make really smart decisions. We then get to a place where I think we
can have a whole lot of companies building things that start to feel a lot more like PMAX.
And it can get really, really fun in our ability to try new techniques at massive scale
with a huge internet firehouse with really low costs to start to find new novel ways to drive
outcomes at scale. That's the hope that I have for the market and that's what I think this is all
about. This is about finding a way to make this overall market perform better and drive better outcomes.
Well said, really interesting conversation and an interesting vision for the future of
programmatic. Thanks for joining us, Andrew. Thanks so much for having me, Erie.
Marketecture: Get Smart. Fast.
