Veritone, Inc. (VERI) Earnings Call Transcript & Summary

Hi, everybody. I'm John Hutchison, Executive Director from JPMorgan's Investment Bank. Here to introduce Ryan Steelberg, the Co-Founder, CEO and Chairman of Veritone. Veritone was started in 2014, about 11 years ago. Ryan started it with his brother. And Ryan has since grown the business to $100 million, IPO-ed the business and overseen the execution for the past 11 years. Ryan has 25 years of experience in executive management across media companies, tech companies, including divisional head, divisional leadership at Google. And Ryan is based out of Orange County and a native of Orange County, alumni of UCLA. So Ryan, first, why don't you tell us a little bit about your story, building Veritone and introduce the audience to yourself.

Well, thank you. And for everybody online, nice to have an opportunity to talk to you today. My journey started and it's kind of the journey of Veritone in the mid-'90s. So one of the sort of first pioneers to build one of the big first ad tech companies called AdForce right out of school, built that up to be, again, one of -- it was pretty much AdForce and DoubleClick that kind of led and pioneered the way for all kind of ad delivery on the web. Built that business up through its IPO in the late '90s. And we've had a lot of successes building and running, I'll say, tech-enabled advertising technology businesses over the course of my career. I think the one thing that we learned, and I think it's one of the reasons why in the disciplines of AI, you really see, I think, 2 big split and threads of where you saw so much innovation come from. One was e-commerce, we'll call the Peter Thiel and Elon Musk path. And then you'll see the ad tech side, right, primarily Google and that side. And there's a reason for that. And why that history is important is running AI at scale is very challenging, right? It's a massive data problem. And ad technology and ad serving is a major technology data problem. You need to be able to -- and I'll do the parallel real quick. But if you're going back and you go through the history of going from serving tons of ads on legacy data centers to now migration to the cloud and then the mobile phone comes out and then you go from websites to search, what you -- the common thread is tons and tons of data, ad requests, billions and billions per hour, and you have to figure out to serve the right ad to the right person at the right time. So fundamentally, the leveraging and invoking neural networks, rudimentary AI-based models was almost a prerequisite to continue to advance when the sheer volume of ad delivery exploded around the world. And then a whole another phenomenon happened, and this was really onset of the mobile device where video and audio exploded. So for many, many years, almost everything we were doing originally was a very structured. HTML is very structured, meaning think of it as structured data versus unstructured data. Common theme that we're going to be talking about a lot. I'm sure many, many people across this conference are going to be talking about it. But the fastest-growing data set was unstructured data. So think of it this way. I have somebody who's watching -- who's online and they're no longer just reading a written article looking at a few pictures, but they're spending 90% of the time full screen watching a video. Well, if you're a Google and others and us ad tech guys, that's both an opportunity, but very scary. Where am I going to serve ads, right? If it's -- so there became a strong interest, leveraging, again, rudimentary cognitive-based AI, and this goes back almost 20 years, is we need to start figuring out what's inside this content, and we need to do it at scale, number one. And number 2 is we need to start having a much better understanding of contextual-based targeting, right? So I'm not just serving an ad and trying to build an inference model about who you are when you're logging on. But also I need to figure out and do the parallel between who you are and what you're looking at, what your interests are, okay? Very -- it was very challenging. Legacy algorithmic processes and models weren't working. And so that's where you start to really see the emergence and investment heavy in AI. Obviously, Google -- when I sold my last successful business to Google in 2008, I was tasked primarily to go after the unstructured data market. How could they start to look at these huge corpuses of audio and video and how could we start to figure out how to monetize that at scale? So that was really the kernel of Veritone. So Veritone is an idea that, frankly, we were thinking about going back to right after we sold Demark broadcasting to Google back in the day. And our thesis was this and it holds true today, and it's been validated over and over is, while there's a lot of unstructured data being created, and this is the tip of the iceberg. And just to put it in perspective, 80% plus of all new data created every single day is unstructured, messy. Think of it as noise and harmonics of an energy grid. Think of it as audio and video, right, being produced. When I say unstructured is it's hard for rudimentary machines to understand that data without actually adding -- doing a lot of work to the data first, okay? So you have to actually -- it's like mining ore. But again, if you can figure out how to mine ore effectively, there's going to be tremendous value there. So the -- Veritone was the idea, and as a play on words means truth in the signal is we believe we could create something very meaningful and impactful that if we could build a system that could ingest huge amounts of unstructured data, primarily audio and video, that will be a common theme throughout this talk. And I could leverage both cognitive science and obviously, today, future, I'll say, more advanced AI models, large language models, et cetera. But back then when we started this, we were really having to deal with AI-based models we were creating ourselves that were, I would say, more rudimentary cognitive models. These are transcription, translation, object detection, face detection. In no way are we minimizing those. But when you are dealing with tens of thousands of hours a minute, and our volume today, we process well over 100,000 hours of audio and video every single day for our customers, the volume is astronomical. And so -- so fundamentally, you have to be able to, a, be able to figure out how to ingest and store and index all this huge amounts of unstructured data before we can even do anything with it. Number 2 is, do we have the right cost-effective AI models that can extract the enough of intelligent data from it, so we can act upon it. So here's the best way and I'll speed up. We don't need to go through the rest of the history. But this is now -- 2012, we have this idea. The original idea of Veritone was purely for media and entertainment, right? How can we help extend this technology base to the big media companies, the big advertising companies out there to extend. But right -- even before we sort of launched our first prototype, which our platform is called aiWARE, which is a key point in our technology stack, we knew that this was much bigger than just tracking advertising, right, and facilitating advertising. So I think what I'll do is I'll give you 2 examples, end-to-end of like a customer to sort of explain exactly what we do in the commercial sense. And then obviously, over the years, now we've expanded which we're very excited about into the public sector and other areas where there also is a huge need for understanding, leveraging unstructured data. So ESPN, kind of a prototypical customer of ours on the media and entertainment side. We have mostly large enterprise customers as our accounts. And they range again on the commercial side from large -- dozens and hundreds of media entertainment customers from movie studios to broadcasters, audio groups and video-based groups. Let's take ESPN and Disney. So for ESPN, who's been a customer now for about 7 years, they keep renewing and they keep expanding. We are their kind of primary AI back-end system. And what we do for them is we, every day, ingest every piece of content that they produce, obviously, starting with a lot of their historical archive, and that includes everything you hear over audio, podcasting, their video, the actual linear broadcast programming and all the components that go into that. So think of this huge amount of tonnage that comes into it. ESPN's job, and they do create some of their own singular shows, but for the most part, they are an aggregator. They're having to collect hundreds of thousands of different sports clips around the world, be able to ingest, understand this quickly. And ultimately, you and I, we see the Varian product, which is SportsCenter, right, that comes out every report. That takes a lot of work. Now obviously, with many different platforms for distribution: mobile, social, et cetera, their job, their speed and the time that they have to ingest index organized package and re-push out is incredibly high, and it's very expensive. When we start -- first started working with them, they had well over 500 interns, literally manually labeling content, trying to collect sources. Obviously, I think we've completely automated that whole process for them over the years. So now the new world is all that content is being ingested into an instance of aiWARE for ESPN, where as their system of record, we're ingesting all that content, we're indexing it in near real time. And think of it as just layers and so tonnage, so that -- visually think of it as creating a huge data lake first and then applying the right AI models to extract that value that has multiple layers. So you can -- and you don't want to trivialize part of the first result. You got a question?

Yes, yes. And so from a viewer experience standpoint, that's when they call up a clip from 10 years ago, that's because it got tagged by Veritone and they're able to pull it up quickly?

One of many. The use cases I'll get to, but before you get to the plethora of different use cases, you have to go to the ingestion index organized. It's like creating the Google Search Corpus before anybody even starts to search for it, right? But yes. So now once you've done this full indexing, the opportunities, and I'll really focus on 3 main personas the end user who -- where is this value being derived from, really falls into research, programming, advertising and sponsorship. And then obviously, you're going all the way to B2B2C. The end consumer, obviously, is the final beneficiary of this. But now you can start doing really amazing things in near real time is pull me up all footage where Tiger Woods is on screen his face. There's a Nike logo in the background. In an aperture of 5 minutes, they're talking about redemption from a car accident to winning the U.S. open, okay. And by the way, what I just described up until the exciting inclusion of large language models into our stack is it was still efficient, but it was like bullion, your classic search, right? Now obviously, you can do it organically. So what I just described seems like, well, that's amazing, it's like truly a search engine for the unbelievable amount of corpus of audio and video. You're right. But that's just -- but what's come and been derived from that is all -- now it's been seamlessly integrated into their workflow. So they know what -- why -- which programming is working, what's not. They've now done the correlation to Nielsen ratings. They've now done all the integrations with working with us to first-party data for consumption. So they know what's working and what's not working. They know that, that host that they signed a $100 million contract with 5 years ago, his ratings are declining, right? And now you can actually have that layer of understanding in that data to, in effect, to prove and support your decisions when you're trying to figure out new programming ideas. The advertising side, just is probably the thing that helps fund this whole project is because so much advertising is now, as we all experienced, particularly in sports, is embedded inside the content, we now have complete visibility and resolution of the efficacy of those ads. Meaning, obviously, we're in Boston right now, right? The Celtics somehow won last night. I'm from L.A., so I can do a little pun there. But if you watch that game, you're seeing advertising embedded all throughout the broadcast, right? It's not a commercial break anymore. It's literally organic, right? When there's a time out. We all heard it, they're saying, this time out is sponsored by GEICO and there's a GEICO logo in the background. Because of AI now, that's indexed, we can now understand how valuable and how effective embedded organic advertising is native-based ads in contrast or in correlation with a commercial -- classic commercial break, okay? So that's just another big example. What we're finding so exciting is once you've created this huge corpus of intelligence from this unstructured data, every company is now taken in some different direction. So again, we work with CNBC, CNN, the Disneys, almost every audio broadcaster out there has been clients for years. And so I think we've created a great sustainable and growing business in the media and entertainment and commercial space. And I'll just give you a metric and then I'll open up for a few questions before we go over to public sector. But scale-wise, on behalf of our customers, we indexed and processed almost 60 million hours of audio and video last year, and that's like almost 11 petabytes of data. It's huge, massive scale. So much scale that we are platform-agnostic, right? So we run all of aiWARE and our payloads on AWS, on Azure, we're completely platform-agnostic. And that has multiple benefits, but also by the nature of how we've designed this platform, our customers, let's say, who are big Azure or AWS customers, they can actually knock off or use this platform against their commitments and their credits. So anyway, that's kind of the back story how we got here. It started with an ad tech vision, expanded. We are now, I'd say, if who is Veritone? We are the experts on large-scale audio and video indexing and understanding with AI, period. There -- I don't think there's any company that's better than audio and video understanding and leveraging the opportunities for that than Veritone is. So that's really who we are to our DNA.

Great. Thanks, Ryan. Bringing things up to speed in terms of a business update. Over the past few years, Veritone has gone through a transition, culminating in the sale of the legacy non-software business, Veritone One. So can you tell us about that evolution just in the past couple of years? Is Veritone through the transition to becoming the pure-play AI company?

Well, we went through -- I'd say we started out as pure AI pure play, and then we went through a journey and we're kind of coming back to our roots as the markets matured. I've done this. This is my sixth company, and timing is everything at times. We -- when we first launched this, we took this company public for some various reasons, very early when we were doing, frankly, like $10 million of revenue because we could, right? We were able to raise a tremendous amount of money. And my vision when we first -- and our vision when we first launched Veritone was we really didn't want to build end solutions or applications. We wanted to build this large infrastructure called aiWARE publisher APIs and just wait for the business to flow in and we would index all of Disney's content and that was going to be more like a Twilio, if you will, back in the day. But back in 2012 through -- early days of the founding of the company, the industry wasn't ready. They weren't ready with their data sets. By the way, if the companies or my clients don't really understand their data sets at all or frankly, a lot of their data sets are not even in proper digital form, what do I -- I need gas to run my engine. So we had to do a pivot for the first few years, I would say, from 2018 through the early '20s -- 2020, '22, where we had to go build the actual end applications. So meaning, instead of ESPN, giving them a bunch of unmined ore, we had to do a lot of work and so meaning what people are buying from us, that's running on aiWARE, but we had to build the end applications. So I'll make a rough analogy. We were building windows, and instead of people building applications on Windows, we initially -- because the industry was not quite there yet, we had to go build Office, Microsoft, and that's what we've been selling. So the majority of our revenue today, about $60 million of it is from the end applications built on our stack, which -- everything I told you, all the ingredients, all the cool AI, what turns into value and discernible ROI for end customers is the application layer, right? And obviously, I think, thankfully, we had enough subject better expertise to make that pivot. The other thing was in our history, just because of our background and other reasons is we were and we did raise a lot of money, right? So even with our experience in, I'd say, a few failures in the past and a lot of success, we kind of broke the cardinal sin of we started waiting for the market to mature with shareholders, we started spending money and we got distracted. We started -- we kind of applied our technology stack into a lot of different other areas. We were in the energy grid optimization business for a while, literally trying to optimize, right, solar and battery utilization with our stack and because we could, right? We were kind of waiting and their budgets were really not flowing. Bottom line is that we spent a lot of stuff got distracted. And so unfortunately, when the true bomb dropped in this industry, which was the release of ChatGPT and the whole world woke up and said, let's go. Thankfully, I had enough of a stable business, which I've just been describing on the commercial side and the emergence of our public safety side, but we were not unfortunately in a position to sprint and capitalize on it, right? That was a mistake we made. We spent too much money. We then raised debt. And so for -- unfortunately, with, in my mind, the perfect stack that was ready to go and get us into hypergrowth velocity, I was cleaning house. I was a private equity CEO, right, trying to manage downsizing when I should be investing in going for hyperbolic growth. So that process of me, I'll say, trying to clean up the business and get us back to, frankly, our roots has taken about 2 years, right? So I do believe we're through the Valley. One of those, which you just described is we did at one time own, a traditional media agency, like an ad agency. It was a big one. It's called Veritone One, and we finally offloaded it at the end of next year. It was powered. So what differentiated that advertising agency was our technology, but it was a service business. It wasn't -- we weren't selling -- the core product was not aiWARE our applications, it was this third-party services group. And that was -- I would say, it was one of the few major strategic moves that I needed to make to get us back to where we are today, where I'm pure-play AI. This is what I do. I ingest unstructured data, I turn into value, I sell applications, and I think we're there, which is exciting.

That's great. Jumping into what I think is the most exciting question today. Veritone Data Refinery was announced in November of 2024 and is an exciting new growth driver for the company, already driving about $10 million in pipeline. With premium data becoming more valuable for training AI models, what we think is going to be a $17 billion market in the next 5 to 7 years. VDR is positioned to deliver a huge amount of growth for the company to help monetize customers' content in a whole new way. So can you tell us a little bit about VDR.

Yes. So in the workflow and ecosystem of AI, training data is the lifeblood, right? Before you can run these models against new data sets, like running transcription and, you have to actually build the models. Building these models takes training data and lots of it. And it's in that -- it's exponentially growing the demand for training data. It took a whole another level when you introduce large language models and now multimodal models, meaning models that can understand and discern audio and video other things in a common model. There's been a few companies over the years that have emerged to assist in the preparation of training data. And it's something that we've been looking at for a few years now. One is a private entity called Scale AI, which you may have heard of. Just to put in perspective how important preparing data is, the CEO of Scale was actually with the President in the Middle East, just perspective. You're seeing Coke and you're seeing Jensen and NVIDIA and Elon Musk, but then there's somebody there who prepares the data literally. And it's not just pure technology, it's a Mechanical Turk Human Labeling company, right, which is a big portion of the revenue. That's how important data is, and it's an insatiable amount. You mentioned $17 billion, it couldn't even surpass that. So what we looked at was how can Veritone because we've been now for over a decade, ingesting and amassing the largest corpus of clean audio and video, this might be a big opportunity for us. What made us made the decision to open up this organic new line of business was Shutterstock. Shutterstock, a public company, recently been, I think, taking private and consolidating with Getty. But Shutterstock, who is one of the largest stock photo businesses out there, they entered that space meaning they started to mobilize and license and work with the hyperscalers. These are the Google Geminis. These are the OpenAIs. They started to license their imagery to these groups to help them train their models. We are watching this very closely since obviously, we're sitting on a tremendous amount of audio and video. And then when we saw their growth that really went from like $0 to over $130 million in just a few years in their data services business, it was when we started to craft our idea really in the -- like second quarter of last year was when we said, "Hey, let's really look at what the efficacy of this business line expansion for Veritone." So obviously, fast forward a little bit. We designed and took to market VDR, Veritone Data Refinery. And we kind of came to market in mid-Q4. So it's only been a few months, and the reception has been incredible. So we are -- we think that this is going to be one of the largest lines of business for Veritone for the next several years, where we are actively booking business. We're generating material revenue. Contracts are 6 figures and higher that we're working with. And specifically, again, is we are working with -- on both sides of the equation, where on one side, if you will, representing and working with our customers who -- we've had customers for 10 years in some instances, not just large media companies, but I'll just say other entities that have large corpuses of unstructured audio and video, including surveillance video, which we'll cross. We'll talk about public sector in a second. But we're now -- we're ingesting, we're preparing that data, ironically using AI to prepare the data to train other AI, right? It's a really interesting virtuous cycle. But now, we're engaged with all the main buyers, and we do think that -- we thought that this was going to be a few million dollars of contribution. I think it's going to be significantly more than that in calendar year 2025 and beyond. And I think we're -- we've got some interesting moats around us. I mean, Scale is a big one. And I'll give you just a couple of other examples is we're not just limited to facilitating the data sets that we already have under representation. It gives us a huge competitive advantage that we do with the groups like the NCAA and others that have these huge libraries. However, we are also being tasked and working with the model development companies, the hyperscalers who are giving us task on their side. They're telling us, under contract with them to go find new data sets, help us prepare it. So if you want to think about it, is Veritone is the audio and video version of Scale AI, right? And that's a very, very exciting opportunity for us. And I've seen lightning in a bottle a few times and this could be one of them. Even if it's not lightning, it's going to be producing a lot of revenue ore for us, which is exciting.

That's terrific. Jumping into public sector, you've alluded to several times now. Veritone has made great strides in the public sector with some large wins with the DoD, DOJ. What are some of the problems that you're solving for the public sector? And how are you competing with -- how are you taking on competitors who are also trying to dig into the space?

Yes. So reflecting on the commercial side with media and entertainment customers, it's interesting. Their core business is leveraging the audio and video they produce and make the money. The core asset that we act upon is their core product offering. So meaning media entertainment customers have triple PhDs, if you will, on the format of audio and video, even before cloud and going back cold storage and tape in, et cetera. So meaning we've proven there was a huge problem. We've generated a nice business from helping them mobilize and advance, right, for media entertainment customers, meaning -- now imagine you look at almost every other company whose core expertise is not managing their data, right? It's not their core product offering. If you look at State and Local Law Enforcement, I look at 2 groups, the Department of Defense and certain agencies like the Air Force or State and Local Law enforcement, these are entities now that are being bombarded with having to, again, understand and leverage tons and tons of data to run their business. If you are Beverly Hills Police Department, a client of ours, and you are involved in a case or an investigation, this is no longer cops going out and talking to people and taking manual notes and stuff. This is now a massive data collection. We know -- the one we've all celebrated now and it took years to really understand is DNA. Imagine doing cases now without DNA, right? By the way, DNA was probably one of the most unstructured things that you could figure out or try to understand. It took years to have any credibility, right, before even it was acceptable, right, in case is now, in fact, it's considered as ground truth, right, ironically, even more than I witnesses at times, as you know. So if you are -- let's go very practical, there is a homicide or there's a domestic violence incident. They -- every investigation has to start collecting evidence. The fastest new form of evidence is audio and video, body cameras, dash cam, security cameras. It's -- and obviously, we like to say, crime travels. Something bad happens, the person fees, right? We call it like Jason Bourne, they're running across it. You have to collect this information. You have to build a case. The Boston bombing, we're here. That mean -- it took literally thousands and thousands of people. They rented out warehouses and sifted through the stuff manually, okay? But imagine -- and that's unlimited budget with the FBI. Imagine now, right, you have to do that in the case where it's citizen upload video. So we have applied Veritone and aiWARE technology very similar to what we do for audio and video for ESPN, but now we're ingesting and integrating and harvesting all those disparate data sets to help accelerate investigations, and there's many different layers of this and it's so exciting. But again, ultimately, is we want to close cases faster, which is time savings because I don't care where you are and what your position is, but all these budgets are being constrained for the law enforcement. And then there is all these residual applications or use cases that popped up like FOIA request, Freedom of Information Act. So when you see footage show up and it's being redacted. So it's not just using our technology to find the bad guy, it's also protecting, I'll say, the identity of citizens. They have to release that footage to the public. It's called FOIA, Freedom in Information Act, where that information has to get released to the public, we need to protect people. So we use our AI to obfuscate their faces and change their voices, right? And that's what you see when those video clips pop up on the news and stuff like that. So public sector is a very, very exciting business for us. What I just described in kind of that micro use case for state and local is the same applications at a much bigger scale, which we're doing the DOJ and the Department of Defense. So we're actively working with the Air Force. We're actively working with the Department of Defense Logistics Agency. It's taken us a very long time to get in the space, getting into the Fed, in FedRAMP and having your authorization operate takes years. But I think successfully, we're there. And you're going to see -- 2025 is going to be not just the year of VDR. 2025 is going to be the year where Veritone of breaks into a new electron level for public sector growth as well.

Any questions from the audience in our last few minutes here? All right. A couple more from me. You mentioned these great really exciting growth vectors in public sector and the VDR product. What are a couple of things you wish the market better understood about Veritone in the story?

We've been around for a while. We've -- I think it's probably -- here's the core pillars. One is Veritone probably has the largest number of enterprise core ad customers outside of hyperscalers. We have thousands of customers that use our core AI platform and applications every single day. We have tens of thousands of end users that use our applications. So our solutions are mission-critical to the missions for these commercial businesses as well as the public sector. So I would say that when you dig in, the more interesting and exciting it is, so there's real meat there. The testament and the proof point there is our retention rate is very high, high at the 90 percentile. And many of the companies I mentioned before. The NCAA is just -- I think, just signed a 5-plus 3-year deal, right? I mean these huge extensions. So we are here to stay, and we're experts at this. So thankfully, we've done a good job of being, I'd say, strategic and fiscal stewards to get the business back in shape, right, back to our roots. But here's -- the second thing I want people to know is through that journey of cleaning up the business, we did not get rid of any of our core assets, not one. Our core lines of business, we kept all of them. So I think people are -- this is really an entry point. I think, frankly, we are probably the most undervalued stock on all NASDAQ, [indiscernible], right, relative to it. And I think that will change very, very quickly. Hopefully, it shouldn't take one big announcement, right, with a new big DoD customer, it shouldn't. But again, we have to prove ourselves. At the end of the day, it's a numbers game. We're not -- we shouldn't be a story stock. But again, I just think once we get back into the light, for over 1,000 days, our stock was over $20 right, going back years. So we should be at a different point. I think some were strategic moves we made, right, poorly in the past. But we're through that. And now it's just, "Hey, take a look at Veritone." Because -- if you're looking for AI "exposure" this has been -- this hasn't been a diamond in the rough, it's been a diamond that's kind of been under a [indiscernible] here for a while, and it's still there and people are going to see some amazing things this year from us.

Wonderful. Thank you, Ryan, for joining us today here in Boston at the JPMorgan TMC Conference.

Thank you. Thank you, everybody.