B (2:17)

It was a new idea and it's called Lattice Brain.

D (2:20)

That's right.

B (2:21)

It's a bit of a banger. It's not bad.

C (2:24)

So that's why we did it.

B (2:25)

So, yeah, yeah, yeah. Terry Quadra is going to have to put a new. A new chart together for the musical inspirations of ad tech.

D (2:33)

Okay.

B (2:34)

So what we're going to talk today is about Lattice Brain, which is the name of your AI and then Outcomes, which is a new product on top of it. The what, what is it? Let's just start there. What is Lattice Brain?

C (2:47)

So we launched Vian AI about a year ago and kind of the North Star product vision for us has been autonomous advertising or Autonomy. And the fourth phase we rolled out planning. So we have AI planning, we have AI bidding, AI measurement, and then the last phase is really what we called AI decisioning. And really this product launch underneath Vyant AI is called Outcomes. But it is our first fully autonomous ad product. So if you think of a self service DSP like we are traders log into our platform, they self execute, design all their campaigns, you know, do it, they control every decision. We wanted to launch a product where you give us four pieces of information. The URL of the advertiser or product for flight dates, the budget and what your outcome is. You know, it's a CPA or roas, whatever it is. And then from there the Outcomes product runs fully autonomous and makes all decisions from planning all the way through measurement. It's different than traditional self service setup. We still have that if that's what you want to do. This is kind of aimed more at the Google's PMAX or Meta's Advantage plus offering and those are more automated. But we wanted to launch the first fully autonomous ad product for the open Internet.

B (4:03)

Right. So Viant Max wasn't available but vine.

C (4:06)

Plus, trust me, we argued for weeks.

D (4:12)

We decided no, those names are taken. We're going with Outcomes.

B (4:17)

Outcomes. All right. I understand what an outcome is, so does so the customer has to know what they want. They need a specific measurable outcome. How do you get around or how do you think about the problem of incrementality versus non incrementality? And one of the big problems is that online media biases towards, you know, cookie based outcomes. So is, is that part of the consideration or you just let the AI optimize what it optimizes to it will.

D (4:44)

Be, it's optimizing off of platform tracked conversions for everybody's knowledge. So it's using our platform track conversions and driving all the decision making off of that. But in the end the AI is determining frequency, caps, bid price, which publisher partners go on and it's going to hit that goal, either a ROAS goal or a CPA goal. We are measuring incrementality also with this we just didn't want to release an autonomous product that wasn't using a de facto ad standard. So we went with CPA and roas. But incrementality, it's available. We track and measure it as well. And we want to deliver an incrementality outcome as well. That will be the next phase. But rather than selling two things at once, we said, let's hit the lowest common denominator of CPA or roas, and then we'll bring incrementality once customers trust that the outcomes are actually happening and they're comfortable with the way the campaign's being executed.

B (5:49)

And so you did. You did. I want to call it a case study, but it seems more like it was a challenge. You know, I'm picturing tortoise in the hairstyle challenge here. Right. You had a human versus a machine challenge. Tell us, tell us about that.

C (6:02)

Yeah, and we knew that would be controversial, but I mean, it is what it is. Yeah, if. If you want to. Not if we're moving towards. Ever since we went public, if you look at our IPO roadshow video, it was all about autonomous advertising. That was again, Northstar product, you know, vision there. And we know that this is where the industry needs to move because the programmatic ecosystem, this is like flash trading 24, 7, 365 days a year. Humans go to sleep, they take vacation. In the end, we need autonomous systems. You still want humans in control and in the loop, of course. But really, we know that the market's going to need to move towards autonomy. And so it is what it is. And clients ask us all the time, if you're building that, is it going to be better than a human? We get asked that question, right?

B (6:50)

That's what we want to know.

C (6:51)

Yeah. So just do it.

D (6:53)

First off, head to head to head. Let's just.

B (6:55)

So what'd you do? Well, you give this. You got a human, you gave them some Gatorade and some instructions and a keyboard and went for it.

D (7:02)

Yeah.

C (7:02)

No. So it was.

D (7:03)

First the advertiser was Mackenzie Childs. The goal was selling e commerce transactions. It was a cpa.

B (7:12)

Okay.

D (7:13)

The trader has. It was an internal trader for Viant. Over five years of experience, very well versed, knows the system. That campaign was live for about two months prior to the autonomous. So it was already a warm campaign. They got access to retargeting. We wanted to make it as hard as possible for the AI, so it didn't have access to a retargeting pixel, so it couldn't learn off of the past visitors. Nothing. Had no access to the segment and we let the AI find all the places in the open web that this should convert mathematically. So it generated the media plan, it generated the audience segments, it determined the frequency caps, the budget allocation by channel and publisher, by format, and executed the campaign completely. And one of the interesting things was normally when a human launches a campaign you're way off the goal and you quickly start tweaking what's, you know, get off what's not working. This thing from a cold start was already within the vicinity of the CPA target that the advertiser was looking for. So that just gives you an idea. It's cold start. It was a warm start right from the because the AI found the relevant locations across the open web. And by the time we got to the end it blew out the human trader. $15 CPA for the AI via an AI, $45 CPA for the human trader with retargeting. So who knows what could happen. And the cool thing of excluding retargeting is just now you can really believe. Holy smokes. These results are really good. And I just want to give one data point to your audience. That was the signal within the signal that this thing is gold. The AI campaign versus the human campaign. We're being generous because we're including view based conversion attribution. Almost 95% of the human managed campaign with retargeting was view based conversions that were in there very low percent click based conversions. When we looked at the AI campaign number one, the click through rate was substantial. I should have had that number for this interview, but I forgot sure. The click through rate was substantially more and 45% of the conversions were click based conversions. And anyone who's been in this as long as we have knows that's the signal within the signal that this thing is striking gold. So that what was gave us huge confidence that the product is ready to go. It's not just taking credit for what was going to happen already, what the walled gardens do best, it's actually finding the locations that are driving click to convert transactions. And that to me was the most impressive data point to share.

B (10:04)

Was, was the human sad when it was over? Did the human have to take get a drink or something?

C (10:09)

He knew, he knew, he knew he was up against it and he was like oh, so I'm going to be. And the best was we run, I don't know over. We, we have over 20 of these pilots running. You know, he knew that we were going to pick this one to talk about. And yeah, he was, he was a Little upset, but he's well, did you.

B (10:28)

Win all 20 or there are 20?

C (10:30)

No, we didn't win all 20, but.

D (10:32)

I think there was three that the AI lost on.

B (10:35)

Yeah, yeah. Do you learn anything when you look at the results? Can you look at the end result? Not just that it won or lost or whatever. Can you say, can you learn about what sites worked and what, what, what creatives worked and things like that?

D (10:49)

Yeah, you learn about what burns out fast. Audience segments burn out, banners burn out, videos burn out, frequency caps are way too much on humans. Things like that. The best learning that I can describe is autonomy. Once you implement an autonomous brain, you remove decision latency. And latency is where campaign performance leaks. That is the line of thought on why AI wins is there's no decision latency or thought or I wait till tomorrow. This thing just makes changes constantly. And when you look at the amount of changes an AI makes in a campaign versus the human, the volume of changes by the AI are so voluminous, it's like thousands of changes happening. And the AI version or the human version was maybe 10 changes a day.

B (11:42)

Between come in in the morning, have your coffee, make some changes, check in the afternoon. Yeah.

C (11:47)

And just typical human workflow in an organization. Right? Like just let's say that you have an insight. Human trader has an insight that we need to, you know, situational planning happens all the time. Where, hey, these sites, these apps, they're not performing. These audience segments, not performing, go back to the drawing board. That takes days. And then it takes even more days to get approvals. But that you might need from your boss or a client or whatever it is, frequency cap changes, I mean, these things are like, they're so politicized also, it's like, no, no, we're not changing the frequency. No. Well, the data shows that the frequency is either too low or too high. And so this is where it's unfair and it doesn't take into consideration, you know, cool motion. Yeah, right.

B (12:37)

PowerPoint. The machine doesn't have to create a new PowerPoint.

D (12:40)

Yes.

B (12:42)

Saves a lot of tokens.

C (12:43)

Set up a zoom meeting with anyone.

D (12:46)

And I just want to explain to your audience too, the difference of, you know, demand gen or pmax. Those are black box AI solutions. This is not a black box AI solution. It's full transparency but no control. And that's what's important for people to understand. You get to know ahead of time. And after every publisher, website, app, wherever, format, you get to know all the things you're used to in a Self service campaign but you don't get the control to change it. If you attempt to change it and override the AI now we're, for whatever reason you have, well now it's not an outcomes or autonomous campaign. It should be a self service campaign if that's your goal. So this provides the transparency. Advertisers want to know where my ads showing up, how often, where is my money being spent. But it does remove control that the self service traders are used to. So you get transparency but you don't get the control.

B (13:43)

Well, I say, I assume you can still set an include or an exclude.

C (13:46)

Yes, yes. When we say control, you don't get to override the system and say no. This strict frequency cap do not scenario plan for more apps and sites. Do not shut off this creative. Even if it doesn't work, keep running. Like you don't, you don't, you don't get that level of control. If you want that, like Tim said, you go back to executing a traditional self service campaign.

D (14:10)

Yeah.

C (14:10)

And the thing is our go to market kind of on this and how we're thinking about this. These are performance based ad budgets and those are typically, those types of ad budgets are what are hitting pmax. They are what are hitting Advantage Plus. And so those clients and those budgets are already used to not having to make every decision. They just want the outcomes.

B (14:34)

Yeah, I think this really stands sort of in contrast. Open web versus closed systems versus Hedge gardens, et cetera. There's generally a feeling like, and I've heard this from multiple people, the open web doesn't perform. Right. That you can't get performance on the open web. Do you, do you think that's true?

D (14:52)

Well, I think not. With AI now historically it was very hard. It was a true statement.

C (14:59)

Yeah, I mean I think it's, man, this is a big, this is a big question. There's a lot of. We're going to piss a lot of people off. Okay. It is true. But it's, there's, there's obvious reasons why, you know, meta. Do you think meta knows who's a bot in their, you know, on their app and who's a human? Of course they do. Yeah, they do you think that they waste their time showing, you know, shark ninja ads to bots? No, they don't, but the open Internet does. So that's, that's one thing right there. I mean if you go back to the, you know, I'll go into the lattice brain. One of the things that we're doing is bringing in data Proprietary inputs or data that we have, that's our household id, IRIS ID on what the content is. Let's you know, video content is that you're actually sponsoring or figure out in as well as our supply quality model which is very important. And those inputs are need, you need to make a real time decision right now. Is this a human, is this a fake device? Is this MFA site, is it a real site? Which household is this? What content is on the screen right now? It needs to be able to make decisions in real time. Not post, not in a database, in a report, then make a decision two days later. It needs to be able to make a decision in real time. So in terms of performance for the open Internet, this is where we have to get to. We have to be able to, you know, use first principles in advertising. This must be shown to a real person. They must actually see the ad. The ad must be relevant to the concept that's on the screen. Basic things like that that I don't think that the open Internet and a lot of the players are incentivized to provide. And it's why you've seen a lot of the performance based ad dollars go to Google and go to Meta. And we actually think that the open Internet has a chance to go after that. But I think that these autonomous, you know, an autonomous system is what's going to be required in order to compete with those Wall garden offerings.

D (17:00)

And we wanted to do CPA or ROAS because we want everyone to see with e commerce transactions. So it here's the open web and here's how you perform across the open Internet. Leveraging AI to find all those little nooks and crannies of the web that are highly relevant for your brain in.

B (17:20)

Terms of where you see your customers migrating or using this product. Earlier you were talking about how you already released media planning AI like it was a year ago and now this is everything. Do you think that the hybrid approach is going to be the most appealing for a while where the customers still do say the media planning separate, separately, then hand it off? Or do you think a bunch are going to move whole hog and just jump in the pool and do everything?

C (17:45)

I, I think you're going to see there's going to be a hybrid approach. I mean like I said, I think a lot of the money in the open Internet is that kind of mid funnel, upper funnel. Right. Especially thinking ctv. CTV today is not really a click to buy ad product.

B (17:58)

Yep.

C (18:00)

I think that those same customers though, they have upper funnel, mid funnel and Lower funnel and they execute different tactics for those. So I think that we will see a hybrid approach and for good reason, too. You know, a lot of our clients, you know, they have upfronts, you know, with Disney and with Paramount and like, they want to execute those and of course they want the control of those and that makes a lot of sense. But they also have the budgets for, you know, driving E commerce transactions. And we just want to have an offering that goes after that.

B (18:31)

Yep, absolutely. So I always ask the same question. We're not gonna do a full lightning round, but just have to ask the question. So if the outcomes. If outcomes was an animal, what animal would it be?

D (18:45)

A cheetah. It runs a million miles an hour. Wherever the fastest animal on earth is. I'm going with the cheetah.

B (18:51)

All right, that's a good one.

D (18:53)

Removing that latency in decision making, performance goes up quite a bit. And that's really a huge learning that we've seen in running all these tests.

B (19:04)

All right, well, with that, I appreciate your taking the time out of your CES schedule. We are going to play the song and insert the video. If you're watching on YouTube, what's it called again, the song?

D (19:15)

Lattice Brain.

B (19:17)

Lattice Brain. Tim and Chris, thank you so much for joining us.

C (19:20)

Thanks.

D (19:21)

Thank you.

E (19:23)

The Vanderhook just brought the lattice brain online Fully autonomous every decision no human required. Watch the whole market get smoked household know the who iris know the screens new let it bring the sign everything in between big channel caps publishers all on his own Jeff Green on the side let's now f matching this cologne person only platform where the machine do the thing URL hit and let it brain already think 260 million average never blinking frequency kept perfect they part on demonic every get right left in the climate big breathing live moving money like water trade this still waiting on the trader in the quarter lying AI say it slow let it ring new brain online and it just took the whole ring Let it break, let it break, Let it break, let it break no co pilot, no chat just a clean rose recording. You want more than that. Trillions of signals cooked into one decision engine Proprietary dietary sauce they can't copy the seasoning Google gets an Amazon text and TTD still ask let it brain already cashed out and relax and tell the street we don't tweak, we don't fall we don't wait I just turn the future to the day first full autonomous now the open web is home Let it break, let it break Joe Green turning lift fully autonomous lighter brain making every decision right in AI. It's not coming.

B (21:20)

It's here.