00:15:26 · David Steinberg
[00:15:26] S0: No, it's great. But my point is that's what changed, right? And then all of a sudden our clients are saying, So we're now so good at marketing. We had retail clients. We would be perfect example of why Athena OS now exists, is we had clients that were saying, You've got to market skirts to females eighteen to thirty five in the Northeast Of The United States. You've got to market pants in the Midwest and dresses on the West Coast. So now the marketing engine is marketing the heck out of that, but the inventory's in the wrong place. Their skirts were in the Midwest, their dresses were in the Northeast, and pants were in the Southwest. So now all of a sudden, clients are saying, Geez, we're marketing, we don't have it. So we built our first real use case of the Zeta Business Engine, or ZBI, is Zeta Business Intelligence. Probably we have a lot of acronyms, it's early in the morning on the West Coast, so you'll forgive me. But the first one was inventory management, because it just made sense. And then clients were like, Wait, we're going to open 500 new hotels in The United States in the next five years. Where do you think we should put them? So we looked at every one of their loyal customers and every credit card transaction they were doing with their five biggest competitors in geographies where they had no location. We said, This is the concentric circle that you should be in, because 60% of your room nights will be consumed the day you open the hotel by existing clients. We can just move from your competitors. So people keep asking me, Why the move out of marketing? Why the move to becoming an intelligent AI infrastructure platform? It's not just a story. It's not just a narrative. It's a business case. Our clients pulled us in that direction. So then we build the Zeta Business Intelligence Engine, and it's massively opened up our town to new opportunities. So you look at it as one use case, right? So inventory. Another use case, hotel placement. Another use case, where do you put your next coffee shop or so on and so forth? And it just made sense to launch an operating system. The other statistic that I'm most proud of, Jake, and I'll date it because who knows how long people will watch this. But in the second quarter of twenty twenty six, we announced that 89.6% of all new code generated at Zeta was automated.
00:18:03 · David Steinberg, Guillaume Moubeche
Think about how innovative you can be when 90% of new code can be automated. [00:18:08] S1: Yeah, it's insane. [00:18:09] S0: It's insane. Right? So back to 2017, I needed 400 engineers in San Francisco to build that platform. Today, we'd need 40, and we can do it in a year instead of three or four. So when we looked at the endeavor of taking Athena, which was originally our super intelligent agent, that by the way is crushing it, and turning her into an operating system, the ability to automate that code base just made that an easier jump because the platform's already got AI and data's foundation. It's already there. You're building on top of it. But even if you look at Athena, and let's just talk about Athena as the voice enabled super intelligent agent for a second. You know this. You're a serial entrepreneur too. We all set out to build fighter jets, but our clients know how to fly a prop plane. [00:19:04] S1: I like this analogy.
00:19:06 · David Steinberg
[00:19:06] S0: But it's true. I have a test. I ask every investor wherever I go, what percentage of your Bloomberg terminal do you use? In The United States, the answer is five to 10%. In Europe, the answer is less than 1%. It's probably higher, then probably lower. Who knows? But the point is the average enterprises use about five to 10% of the enterprise software capabilities that they pay 100% for. It's no different at data. We provide them a... Sorry not to use a Dassault product, but we'll say an F-fifteen fighter jet. And they know how to fly a prop plane. But with the prop plane, they're getting to a 600 to 700% return on marketing spend and a two ninety seven percent return on technology spend. That's pretty good. My goal is to get our clients to an order of magnitude return on marketing spend, and then I'll raise price. I want to get to 1000% return on investment, and Athena is your co pilot to fly the fighter jet. So instead of having to navigate all these crazy platform tools, you go into Athena and say, Athena, I'd like to generate 2,000,000 incremental customers this quarter at an average cost of 15% lower than it cost me last quarter. And instead of being able to access 5% to 10% of capabilities, Athena can access 100%. She can look at every individual in the data cloud. She can look at every single activation point. Every channel, every use case can be looked at in real time. And she changes the UI in front of you to show you the answer. It's so freaking cool, G, because to me, and I'm obviously a nerd that this is what I find really cool, but I was building computers from RadioShack components just to show how old I am when I was a kid. To this day, if I'm in the country, meaning not a city, you'll drive by a hardware store that'll have a Radio Shack attached to it. They'll still have it. The brand's still out there. I'll scream to my wife, Stop. Stop. But the reality is that we said last quarter we had 197 super scale clients the end of Q2, two thousand and six. 40% of them had already adopted Athena. 83% of their usage is voice. And I'm telling you, we're seeing their revenue per enterprise growing at an even faster pace than the 44% we grew that quarter as a top line.
00:21:45 · Guillaume Moubeche, David Steinberg
[00:21:45] S1: Yeah, it's insane. What I love about Athena and everything you've built is like, obviously you're delivering huge amount of value to your customers, but at the same time, I feel like this is a product that people will keep on using more and more, because the reality is like, you're getting a lot of context also on the company and everything that they are doing, which means that you can reinforce Athena and everything you do to basically like suit them better. And any switch to a competitor will actually make the context basically go away and disappear. And eventually the results will be always worse than whatever they were getting with you. So they will always come back. So I think it's pretty magical. You should. [00:22:30] S0: You're tired of running your SaaS based startup, please come work with me, because you just said that better than anybody I've heard say it. Here's the deal. We take our clients' data, we take our data, we match it into a consumer data platform. The algorithms live exclusively in there. No data exhaust to the foundation or frontier models, but 100% of the knowledge base and the context lives in the consumer data platform. It does not go back to the enterprise. Now, we don't extract any of that to our data cloud. It lives in this ecosystem. If you fire us, it all goes away. So I think there's a reason that our net retention rate in 2025 was greater than 120% as a company. So if you think about a Jeep, 100% is good, 110% is like gold. We're [00:23:22] S1: both I wish at you these numbers.
00:23:22 · Guillaume Moubeche, David Steinberg
[00:23:22] S1: both I wish at you these numbers. [00:23:24] S0: No. And I'm not trying to show up, but my point is that it's just something that I think a lot of people don't understand about our business. If you fire us, you lose 100% of the context and inference model knowledge base. So if you think about it, and Chris Greiner, our CFO, does an incredible job on this. The longer a client stays with us, the more money they spend, because the greater the knowledge base, the greater the return on marketing and technology spend, the more of their dollars they move from other software and marketing partners or vendors to us. I mean, I'll tell you, I cannot remember the last time a client actually fired us. I mean, we have some clients that go in out of super scaled or scaled. I now have a new category that I lovingly call super duper scaled. It's been incredible. But because of the way the ecosystem works, their data plus our data, you then remove the personally identifiable information to protect the consumers. [00:24:38] S1: I was curious to get your view on models because, know, like, I mean, if we look at a year ago, I think the frontier models were way ahead. Now we're seeing like open weight and lightweight models, open source models to a stage where, you know, they're either at the same capacity or at least on specific tasks, they can either sometimes outperform like frontier models. So what's your view around this and how do you leverage it as a competitive advantage?
00:25:06 · David Steinberg
[00:25:06] S0: Yeah. I mean, so if you look at the Athena inference model, which we announced in partnership with Fireworks dot ai and NVIDIA, And you look at the product roadmap video or demo, they said something that I think is incredible. Our AIM models operate at a 65% more efficient level and 12% greater accuracy than using frontier models on our own data where we've tested it. They're just getting started. So you're going to have two types of models, right? You're going to have a foundation or frontier model, which is going to require billions and billions and billions of dollars of CapEx to train it. Then you're going to have inference and contextual models that are going to have to be changing every moment of every day. Because inference and context changes every moment of every day in most businesses. They're going to be hyper accurate, hyper scalable, and super small. They're going to be very targeted. And that's what we believe is the future of knowledge base, which is by the way, not to say that Claude and OpenAI are not worth trillions of dollars. I'm not the right guy on that. Work with everybody. And I believe as it relates to auto cogeneration, as it relates to operating business models, and I believe writing poems, they're going to be the guys for many, many, many years. As it relates to business tasks, we believe the future is inference models. And by the way, it appears as if NVIDIA agrees, right? Because they bought Hugging Face. They're buying these platforms that really, I think, are going to be less frontier based and more inference based. Sort of at Zeta, we're already seeing the efficiency and the efficacy of our proprietary inference models as substantially higher. Now, to be clear, I think that 6512% is like baby sticks. It's not even table stakes. I think ultimately we're going to be hundreds of percentages better using our inference models than the frontier models to create the tasks that are mission critical for enterprises to operate their businesses. So it's going to be very interesting to see how I think the world bifurcates into the frontier models, into the inference models, into contextual models, and then how the internet bifurcates into the HTTP layer, and then the agentic layer that lives below it.