Afternoon, and welcome to the Needham Technology, & Media Conference. This afternoon we're excited to have Veritone, with CEO Ryan Steelberg.
Hey, Josh. Great to be here.
Yeah. All right. Ryan, maybe we can start with an overview of Veritone for those less familiar and the transformation that you've accomplished in the business over the last couple years.
Well, Veritone's been around a little over 10 years. We were kind of one of the first AI-based companies to really go after the unstructured data market. I come from a media entertainment ad tech background, and we felt that there was a gonna be a huge problem of the creation of a lot of unstructured data. If you kind of think about what Google and other search engines of the past did with structured data and text and HTML, we had a vision of what if you can ingest and start to index and understand all this audio and video. I came from Google. That was my job working for Sheryl Sandberg and Tim Armstrong, was: How can we start to understand what's inside audio/video? Primarily, to frankly sell more advertising space. We spun off.
The vision of Veritone was imagine a world where any type of audio and video, whether it's streaming audio, it's television, it's security cameras, if we can index all this audio and video, we could really unlock a lot of value and efficiencies across the board. Started the business in 2014. We still work with many of our original customers in the media entertainment space, and we've expanded a lot into public sector, state and local law enforcement. We work for the Department of War. We've been doing that for a while. One problem when you start really early is when we were kind of waiting for, I'd say, the AI market to mature, we did get a little fragmented as a business. We started to go into many different lines of business, going back several years.
It really for us, over the last few years, was just getting back to our roots, our original thesis of can we just service, you know, our media entertainment customers, our police agencies, our Air Force customers, with, I would say, singular focus and that's what we've done. In addition to that, which you can kinda read by some of our filings, is, you know, we did have a lot of debt in the past, where, you know, we've paid off a lot of it, and we're down about $45 million of total debt, down from a peak of over $230 million of debt.
I think you're looking at Veritone as kind of a Although it's a lot of the same people, I'm still here, it's kind of a renewed focus, a renewed vision, and a much better balance sheet and, I'll say, fiscal responsibility.
Got it. Along with that, you spun off the advertising business.
Yeah
I guess, how long ago was that now? 18 months ago?
Well, Yeah, that was in Q4 2024. Part of it was, you know, again, there was just some peripheral businesses that I think were impeding some of our growth, which we've kind of shedded those, and now we're kind of back to our kinda core AI software business.
Got it. All right. You just reported Q1 this morning, and obviously now we're here doing a fireside chat. Maybe to catch people up on the quarter, what would you say were some of the key highlights from the Q1 results?
I think first of all was, you know, revenue was relatively flat year-over-year. I think the big highlights for us was a huge expansion in our pipeline for some of our new data lines of business called VDR, Veritone Data Refinery. We're gonna talk a little bit about that in the future. We did announce some very key signings of two mega new clients of ours, Google and NVIDIA, where we are now, I'd say, a tier 1 partner of theirs, helping prepare, package and license training data to them, somewhat akin to like a Scale AI, if you're familiar with that business. That's definitely another highlight. And I think a third is just the velocity and growth of our public safety business.
69% year-over-year on a quarterly basis in terms of growth. You know, new activation with the Air Force and the Defense Logistics Agency, which we'll go into. I think it was a combination of, again, just improvements of operational efficiency, expansion of the pipeline, and thankfully landing, which we're probably about a quarter overdue, some of them are more bellwether contract with Google and NVIDIA.
Got it. All right. VDR obviously is, kinda the key upswing factor in your business model right now. Maybe you could just start off by explaining what do you do with this VDR platform, and maybe why is it differentiated versus your competitors?
Veritone Data Refinery, for 10 years, we have been ingesting and indexing tens of millions of hours of audio and video for our customers. Those range from CBS News to the Masters Tournament to security camera footage. Organically, about 2 years ago, we really start to see how a lot of the large hyperscaler companies who are building foundational models, the OpenAIs of the world, the Geminis, they really started to need a tremendous amount of clean, IP-cleared audio and video for training data. Obviously, these models were originally trained on primarily text. They've advanced to imagery, now audio and video and other forms of unstructured data. Veritone organically has been processing and indexing this audio and video dataset for years.
We made the decision to introduce that new line of business, named the Veritone Data Refinery, then we've started to package and start to license those datasets to many of the largest hyperscalers out there. We're thrilled about it. We're very excited about it. What makes us really unique is, first, I would say, you know, our expertise in understanding the audio and video datasets. It's probably second to none. That's really where we cut our teeth.
You kind of say if you think of AI and audio and video understanding and indexing, I think that's really where Veritone shines for a multitude of reasons, scale, efficiency, and frankly the end, I'd say the output of our processing to prepare the datasets for AI-based processing. The second is our supply. As Mike, our CFO, touched on in our earnings, we've been signing a tremendous amount of supply with partners to the tunes of, you know, 50 million hours of plus of data that we're able to now leverage and introduce those datasets to hyperscalers.
We've created, in effect, a kind of a multi-tier moat here, a huge repository of proprietary data assets to prepare and license. That we have now also signed, which we also announced, is really almost every single foundational model player, that are also under contract. We sit on both sides. We have a representation agreement with supply. Now we have signed contracts on the VDR side with all the buyers. We're in a fantastic situation finally to start, let's say, you know, sprinting and executing on the VDR business.
Maybe just to touch on, you know, what makes it difficult to index and make unstructured data searchable? Is there something inherent in the way that the data is, like, put together that makes it difficult? How have you kinda like, you know, overcome that, the difficulties with that?
I think, well, first is the scale. When, you know, understanding audio and video, obviously the models are getting more mature. Typically, when you see a lot of these use cases, you're dealing with minutes or a few hours of content, candidly. When you're talking about archives that have hundreds of thousands or even millions of hours, that is a tremendous amount of processing, and frankly, expense. When you're thinking about trying to prepare datasets for large language models, they don't need just a couple hundred hours. Often, they need thousands or tens of thousands or hundreds of thousands of hours.
This is as much, I'll call, an orchestration dealing with the details of bitrate and quality and performance of the video files themselves, as much as it is leveraging the cognitive science of turning that into structured data with temporal metadata. I think it's, for Veritone, it's the combination of both expertise on the underlying aspects of audio and video, going back a very long time of understanding audio and video in the media and entertainment space, coupled with our expertise in I'll call cognitive AI, that combination has proven to be very effective in servicing, frankly, the highest demanding customers out there, which are the hyperscalers. One other thing to note is when we prepare these datasets for the different, aforementioned customers, they all want it slightly different, right?
Don't think of it as just us transferring a video file over. There's a lot of work that we have to do and prepare it differently for company A, like a Meta, versus somebody else, like an AWS.
Got it. Okay. Super helpful. All right. If we look at the VDR opportunity, it seems like the near-term kinda major revenue opportunities remain with the largest hyperscalers. Is that the case, and how does that dynamic factor into the implied, second half, ramp in revenue based on your guidance?
I mean, you think we read in the press, right? The largest budgets that are funding this huge infrastructure build-out are these large hyperscalers, right? The data investment into data training sense is no different, right? What we did project is we do expect for this year at a minimum, the super majority of our revenues from the VDR are gonna come from these largest hyperscalers. That being said, because the cost of compute, the cost of storage, and I'll say the efficiencies of a lot of the AI models or the ability to train models, that bar is coming down. We are seeing an expansion of the type and number of customers.
Yeah, and don't wanna drop, you know, some names, but let's just say that you wouldn't readily think that they're in the business now of training AI models. To be clear, at least for 2026, it's gonna be dominated by the Metas of the world and the AWSs of the world.
Got it. All right. Then maybe just to wrap up that point, the implied second half ramp and the revenue, what gives you kinda confidence in terms of what you're seeing behind the scenes to make sense that there is gonna be some upward trajectory to that revenue? We may not know the- precise dynamic, but-
I think, well, two keys is one, some of the supply we've been talking about, was recently just signed, right? As, you know, let's say as late Q4. The amount of data that we have access to prepare, to package, to sell has greatly increased by stuff that was recently bought on board under contract in late Q4. Think of it as much more supply. Number two is, as we've just touched on, two of the biggest players in the entire space didn't even come under contract until late February.
Right.
If you kinda look at those two huge components in terms of fuel, right, or limiting factors, those have been sort of satisfied. The third element is, which we touched on, is our near-term pipeline is almost $70 million now for VDR, with a material amount of that already out for a quote, if you will, for booking. I think it's just, we were talking about such a compressed period of time here of semantics of a few weeks. If you kinda project that forward, we're very bullish on what that pipeline looks like and how we're gonna be able to mobilize that pipeline through 2026.
Got it. That's super helpful. All right. You have an open platform approach where customers can bring in any third-party AI models. Can you explain why that's the right product strategy versus a closed-loop system, some of your competitors pursue? Maybe on top of that, are some of the competitors that did have closed-loop systems adjusting their approach?
Yeah
because they realized it was a mistake?
Let me reset for the audience in here. We're not talking about necessarily VDR anymore.
Yes
Let's go back to how I'm servicing ESPN, right? If I'm ingesting all of ESPN's content from all these different sources, we build bespoke recipes of different AI and cognition to turn it into structured metadata. Question would be, okay, great, you know, there's, you know, at the, you know, 3 years ago, we were using, let's say, a fine-tuned transcription model that's good for global English, hypothetically. If you, in this space, the one thing that's always constant is change, okay? It's a little cliché, but every single day, we're not seeing more innovation as it relates to models. Not just large language models, not just large multimodality models, but bespoke models like object detection, face detection, et cetera.
From day one, our thesis was, and this goes back all the way to 2014, was we made a bet that there was gonna be millions of models, right? I'm not gonna say we were clairvoyant in saying that we were gonna see the, necessarily the large language models, but what you're seeing today, right, is we're still seeing better efficiency and accuracy for hyper-trained, narrow cognitive models for certain things, like finding a Honda Accord 1983 with a blue dent in the back. That can be trained, right? I don't know why you're grinning. Maybe it was the car that you stole. Working in concert, well, I'll say with general understanding of video, that could be more appropriate with a large video-based model.
Yeah.
What we've seen time and time again is when you can blend these art forms, you get the both worlds. You get the best accuracy, and you can do it at phenomenal price. Remember, a lot of the data that we're talking about, the volume is so big, and a lot of it can't leave our environment. I can't take our video, for example, from ESPN or take our data from a police investigation and necessarily load it into a publicly hosted large model. Okay? First of all, I'm not even allowed to move it often, right? The datasets. Or two, the volume is so big that if you kinda look at the cost of the tokens and the processing for these, it'd be cost prohibitive. I'd be spending tens of millions of dollars a year just to create the metadata.
Our job is, without ruining the application or the workflow experience, if there's a better model on the market, our goal is to get it onboarded as a hosted, a containerized model onto the aiWARE stack, which is our proprietary stack, as quickly as possible. One thing to note is roughly about 30% of the models that we've invoked over the last, like for example, 24 months, are Veritone's proprietary trained models, right? Meaning we don't care, right? If there's a better model that has better yield, better productivity, we wanna have that onboarded onto our stack as quickly as possible without disrupting the customer's experience at all.
Got it. All right. In terms of the public safety, part of your business, how do you continue to build awareness with customers of your position in the marketplace, while at the same time work within customer budgets to kinda get a foot in the door?
We'll break this into two parts. We'll do SLED, state and local law enforcement, and kind of federal government. In state and local law enforcement, there's a few legacy titan players. Axon is a major player. Flock is a major player. Motorola is a major player. What we're seeing is a lot of these groups come in as almost a vertical stack. Vendor lock-in is something that's scary to everybody. This is a real issue. And also the price points that a lot of these legacy companies are coming to bear are very expensive. Police agencies, sheriff departments, municipalities are under major budget constraints. If you take the culmination of, frankly, us coming in later with, I'd say, an open platform framework that we can use with any data source, right, any type of player, it's fascinating.
The best way I can describe it is almost 40% of our current customer base is also an Axon customer. They're also a Motorola customer. What I mean by that is Veritone sits on top. Think of it as a data abstraction layer that sits on top of any deployed infrastructure. I don't care where, what body camera, platform they have. I don't care what CCTV or drone provider they have. The key is for us to come in very cost effectively and create that intelligence data layer on top. The goal here is I wanna solve crime faster, right? I wanna be able to ingest the data from any cell phone, from any CCTV camera, and if I can help that investigator move that case along and increase their case closure rate faster, everybody wins.
The other byproduct is they don't have to necessarily be over-reliant or basically take, at times, inferior solutions. I would say is our end applications, our redaction software, our investigation software is best of breed. We can compete against anybody, and so therefore, we are encouraging working with with the cities directly and with other competitors, other body camera companies, et cetera, as a, I'll say a partner horizontal complementor. Do we wanna provide the LAPD with the best of breed solution end to end? They shouldn't be forced to take a vertical lock-in that's overpriced and frankly has stuff that's an inferior product.
Got it. All right. As you think about the future product roadmap for public sector, how are you thinking about selling just the straight platform to customers versus selling individual applications? Maybe explaining that might be helpful.
Right now, our core stack is called aiWARE. Every application we build for commercial, like the ESPNs or the applications or workflows we build for the federal government or state and local law enforcement, are built on top of aiWARE. The majority of our revenue is actually being derived from us licensing and them using those end applications. Think of it as Windows versus Microsoft Office, right? Everybody knows that Windows and other operating systems have SDKs. You can build your own proprietary applications on those. By the way, you can do the same thing with aiWARE. Far, we've had the most success of selling our applications.
What we are seeing, and I think that's going to be case for smaller agencies, right, who have limited technical resources or budgets, we don't envision them building necessarily custom applications or solutions today on aiWARE. The larger agencies, yes, because their use cases are more diversified, they're more complicated, and frankly, they have so many different departments that us reimagining what a, an improved workflow or a custom application for like, let's say, LAPD, makes a lot of sense, right? Versus you force-feeding them, right, I say a smaller or limited scope application. Very simply is I do see us still being more successful selling applications, AI-based applications to smaller firms. Larger firms, you'll start to see more custom builds and solutions built on aiWARE, which is what we're seeing in the Fed.
U.S. Senate, the Air Force, Defense Logistics Agency, by the way, they all started with buying single applications. Every one of those clients now are building or frankly at times engaging us to build custom workflows and applications. Not just taking the off the shelf applications, but doing something that's larger or more custom for their type of organization.
How does that compare to your kind of historical kind of dominance or strength in the media and entertainment vertical where you were just selling the platform? Why did it make sense for you, for those customers primarily to buy the platform and not build individual applications for those customers?
Well, actually, we have, we very rarely ever sold just the platform in the past to media and entertainment.
Okay.
It's almost always been the application layer, right? Of selling into media and entertainment, right? Very rarely were they buying, for example, Maybe if the first couple years, with like Disney and a couple others, we were simply, they were accessing our APIs and building custom solutions on top of that. That was not very successful in 2014 through 2018. It was too early in the space. It's kinda like, people didn't have a webmaster back in the day, right? Even though you could build very easy websites in HTML, people frankly just wanted turnkey applications back then.
Just to be clear, what gave us all the benefit was thankfully the way we designed the workflow and the SDK of aiWARE afforded us the ability to build these applications so cost effectively, right? That are very competitive or extensible. What you're gonna see here, it'll speed up, is 2026, the latter half of the years, you'll start to see the re-emergence of platform first for Veritone, right?
Okay.
Which really hasn't been the case for several years, right?
On media and entertainment.
No. I'm gonna say I think media and entertainment will still be dominated by applications.
Okay.
You'll see probably the biggest emergence in utilization of platform direct on, in the Fed space.
Okay.
More like a Palantir, right? If anybody's familiar with AIP and what they've been successful with. However, we're trying to do it without the requirement for four deployed engineers. We're trying to make it where it's more of a self-service type of platform and framework as compared to being, I'd say, overly SI or consultative based when you're doing an implementation at the platform level.
Got it. Okay.
Yeah.
Helpful clarification there. All right, the most interesting incremental data point to me today from the earnings call was around the data hub and expanding the TAM with more customer opportunities there. What are you doing exactly to expand the number of customer opportunities with that from a product and customer perspective?
What he's referring to is for the Masters, for example, in NCAA, the application built on aiWARE, the application they're using is called Digital Media Hub. That was primarily designed for, frankly, the largest entertainment shops, meaning the integration or the ingestion layer was done almost MAM to MAM, media asset management system to it. Think of as enterprise grade. It was never really designed for self-service, smaller creators, enterprises who have lots of audio and video. You know, Lowe's department stores that the theme here now is everybody in effect is a media company. Everybody is producing audio and video at scale. We were missing out on that business line. Meaning for those who wanted more of a self-service approach, they wanted a lower cost entry point.
That's something we've been working on now for a few months. Thankfully, it's not a huge lift, but you're gonna see us emerging over the next couple of months and reintroducing Digital Media Hub to, really, I'd say, reimagined in terms of onboarding, reimagined in terms of a price point, and reimagined in terms of an architecture, so we can go after those who have a few thousand dollars a year budget and not somebody who only has six figures or higher. We think that's gonna greatly open up, a large, you know, TAM for us in a big way, and I think we'll start to see the realization of some of those revenues as early as Q3.
How do you go about the go-to-market for that dynamic? 'Cause right, you're historically, other than the public sector side, maybe you chase larger-
Yeah
you know, media companies, right? How do you manage that kind of dynamic there?
Thankfully, there's a little bit of pent-up demand because, for example, the Big Ten and the Pac-12, I don't know what they are these days, but I think they're getting close to the Pac-12. They have teams. These properties have been wanting to do business with us for a while, but we really weren't built to go after smaller properties. First of all, I think, you know, is, I would say, is the affiliated organizations around these large conferences, if you talk about golf, there's the Players Direct. There's smaller tournaments that we've had to walk away from that we haven't really had the opportunity with. The Masters is such a large property. The amount of content they produce kind of crosses that bar where we could service them.
I think number one is there is an installed pent-up backlog demand for us, by introducing, I'd say, this, I'm not gonna say lower end. It's got the same quality, it's got the same sophistication, but it really, the packaging and the price points allow us to go after them. The second one would be, is gonna be more net new to us, is corporate enterprises. These are, you know, groups that have marketing departments with a tremendous amount of audio and video. This one will be more net new to us, right? A reimagining, new go-to markets. I'll do the third one is partner.
Partnership ecosystem, which I think we've done a pretty good job at, historically, is gonna be, we're really gonna rely on, which we're gonna touch on a little bit later here, the Oracles of the world, right? New partnerships that we've recently entered into who are trying to get very aggressive, trying to get more people to put, you know, both storage and compute payloads on OCI or Oracle Cloud Infrastructure. We are gonna be reselling with organizations like this, and we wanna make sure that we can go after everybody who's got audio and video payloads, not just media and entertainment, sports, and news enterprises.
Got it. All right. On the public sector side, maybe still hitting on that, you've done a really good job with the OSI in rolling out the Air Force contract there. What's the background there, and how should we think about other opportunities within U.S. Federal?
Patience is what you need when you go from commercial into the Fed. We're in a time of war, and so, and we're not selling tip-of-the-spear munitions, so things, let's just say, take a little bit longer. I think we signed the Air Force contract almost two years ago now. This is how long it kinda takes. The OSI, the Office of Special Investigations, is the division of the Air Force, which by the way, covers across other agencies as well. They're like the global policing force for everything. Whether it's security cameras going on bases, if they're doing active investigations, protecting their own or doing prosecutions, right, that are somehow affiliated with the global Department of War. Think of them frankly as a really big police agency or law enforcement agency.
Again, we sign that contract. For them, the security requirements are a whole different level. We had to take the entire aiWARE stack and all the applications and deploy it in their private cloud. This is not something that's not running in AWS or Azure gov or commercial. This is actually in FedRAMP and also in air-gap, network-isolated environments. That was a big task. It forced us from a maturity effort. We had to kind of reimagine some aspects of the aiWARE platform, so we could actually deploy the full stack into these network-isolated environments. It's taken us a while, but Air Force and OSI is fully active. We expect them independently to continue to grow.
We do think 2026 you'll start to see a ramp in direct revenues and expansion and utilization of services from the Air Force direct. Over time, the OSI and that body that's managing these services is been tasked to consolidate these efforts across the 17 Department of War agencies. Why this is so critical to Veritone is it's the land and double expand. It's the land and Air Force and growing that singular account, but it's also writing and helping support OSI, the Office of Special Investigations, as they deploy these type of solutions across, the goal is all other 17 Department of War agencies. That's a very important win. Just give you time perspective, you're talking about years before you start to see the revenue.
The good news is, I don't care what side of the aisle you're on or change in, I'll say, the White House administration, hopefully these, what I like to say is the middle-tier back office solution where really we're landing is something that I think this could be literally a 20-plus year relationship that's gonna grow and grow.
That's awesome. All right. Also on the public sector side, you're working with, is it L.A. County or the LAPD? I wasn't quite sure who the customer is there. What can you tell us about your work with this customer?
Yeah
The use cases involved?
We won't go into press release. We're working with both, LAPD and L.A. Sheriff's. It's a good setup. What's really interesting is we landed with them using completely different applications. One, in effect, launched with programmatic AI-based redaction, the other one launched with our investigative solution called Investigate. Built on the exact same aiWARE stack, with the budgets that they had, the entry point, we were able to land in, I'm not gonna give you specifics, one of the entities, they had the budgets for approval. We got into, we entered the relationship with a price point that we didn't have to wait for city council approval, right?
We can be flexible with what app to land, and that's the application that one of these entities is using for investigative purposes. This is collect all the data. Can I find or association of that person wearing the blue jacket may have been associated with the 1983 blue Honda Civic with a dent, right? Can I follow them? You know, ultimately, can I zero in on the suspect or exonerate them quickly to ultimately close the case faster? On the flip side, the other L.A. entity onboarded with Redact first. This is pure autonomous redaction, audio, video, and document redaction of all the tonnage of information they're doing. As we all know, by all this stuff is accessible through Freedom of Information Acts, right?
If we're recording it, right, whether it's if it's in effect citizen or taxpayer funded, those datasets eventually have to be released to the public. Some have to go through more formal process. Veritone Redact, which has won lots of awards, is definitely best of breed, makes that process incredibly more efficient. You're talking about something that's saving, you know, up to 90% efficiency in terms of speed of trying to prepare those files to get them out. That's a great example where you may run into a large agency that may not be open to throwing out their entire full stack, but they have the budgets to start working with you because if we can convince them that we have a best of breed of an investigation solution or redaction solution, that's how you land, right?
That's a good example.
All right. Let's make sure we hit on the OCI, stuff going on there. Maybe can you give us a little background on, you know, why did it make sense to partner with them?
Yeah
How you see the agreement going forward?
We've tried to be completely agnostic from an infrastructure deployment perspective from the beginning. Like many, when we first speed to our MVP, I think we were initially on AWS commercial only. As we continue to rearchitect aiWARE, containerize a lot of the processes, purge ourselves from a lot of the legacy managed services that a lot of the, that the hyperscalers provide, that then opened us to move into GovCloud. That opened up the opportunity to move over to Azure. Frankly, if a customer wants us, we want to deliver the application and the platform wherever they want it hosted, to be very clear. We don't want that to be an inhibitor. When we looked at, and really this pushed us to the next level when we looked at FedRAMP.
If you're not familiar with FedRAMP, that's again, the government's private cloud or just say another level of security. You can, by the way, be in FedRAMP. They have instances of Azure, Google Cloud, Oracle, AWS. Right now, we're also now in FedRAMP. First of all, us being technically ready to be completely platform-agnostic opened the door for us to start doing more stuff, more business with Oracle. Oracle, over the last few years, as you can imagine, I'd say the Oracle family in a greater sense has been pushing very hard in competing effectively with the AWS's and the Azures out there.
With certain family members, when they got really aggressive in terms of buying Paramount and now merging with Warner Bros., Oracle is pushing very aggressively to compete direct in terms of scale and full in scope with the AWS's of the world. I would say it's mutual. We were fascinated by the allure of running our storage and compute cycles for cheaper, right, on Oracle. We were well aware that Oracle, in terms of worldwide points of their infrastructure around the world was second to none. I mean, if you wanted to go from any class of security and you're thinking about going to the U.K. or to Saudi, Oracle's been there for a long time from an Oracle Cloud perspective.
Frankly, we just didn't feel, you know, we were quite ready to work with them on, frankly, our core businesses here in the United States. Again, they've improved and matured a ton, I think it was just kind of a mutual wooing phase that ultimately culminated in us getting this deal done. Yes, Veritone is going to be It's not exclusive, but we are gonna be looking at them as a premier, a preferred cloud provider. We are initially gonna be moving a lot of our commercial storage and payloads over to OCI over the next, you know, 1.5 years. The benefits of that is will be, A, definitively major cost savings on a run rate basis, to be clear.
They're making this, they're creating great incentives for Veritone to be very competitive. By the way, this is gonna transfer not just to help improve our margins, that's gonna flow through our customer base. Purely on a cost basis, this is a big win. Number two, their performance is second to none. In terms of our benchmarking of how we're deploying our software in these environments, we've seen stellar results that are competitive at the worst case and in other areas more competitive in terms of performance of how we're preparing a lot of these data sets to execute our business. That's number two.
Third, which is probably the most exciting for just growth, you're gonna start to see, you know, I'd say very acute, and short-term co-selling together with them. I will be speaking in their sales kickoff meeting in June, right, to thousands of salespeople, right? Talking about what we're doing and why. Big question: Why are you gonna choose to move your payloads onto OCI, right? One of the reasons is, well, now you can get Veritone, right, on OCI. You're gonna see really the three-prong. Highly incentivized to get us to start moving payloads, run rate performance and price advantage. Three, I'd say a very effective co-selling, relationship with them.
Awesome. Great summary there. All right, the Veritone Hire business, it's, you know, been performing well in what's been a pretty difficult operating environment. What have you been doing on the product front there to distinguish the offering from others in the market?
Broadbean's been around for, geez, almost 15+ years or more. Ubiquitous. Thousands of customers, if you think of, you know, talent acquisition and wanting to place, you know, job advertising across thousands of job boards, Broadbean has been there for a long time. It's an old stack. What we've been doing is reimagining the good and bad is they have a tremendous amount of data. I mean, you're talking about nearly 50,000, somewhere on average between 30,000 and 50,000 unique recruiters every month use their software to manage and place job ads at over 7,000 boards. Their data is incredible. It's also been very consistent. If you're trying to think about what an agentic reimagining of their solution is, there's nothing better than starting with that level of data set, right?
The actions they're taking. What we've been doing is in effect taking, and it will be introduced by a project called Tao and then the new job management, set of modules, that is really bringing an agentic, layer across that that's gonna make those tasks, even though they've been relatively efficient, it's gonna take those tasks that those recruiters have to do on a day in and day out basis almost completely automating it. That coupled with our scale, is really I think gonna be very competitive against anybody in the market.
To prove the point, and kind of the on the go-to-market side is, because of not just what we have been doing, but part of that roadmap is, we've resigned and expanded the type of relationships with some of the largest HCM providers out there, Workday, Oracle, SAP. Our fastest sales growth channel for our hiring business today is Workday referrals. We actually are, quote, unquote or whatever you call it, a gold standard, whatever, the gold level partner with Workday, which means they are selling our solutions, and we have a direct integration with Workday. I think that's, it's kind of the combination of us improving the product, having a AI agentic, you know, first approach and a roadmap, and then leveraging very effectively the largest players in the space who are co-selling with us.
Got it. All right. In terms of the cost cuts implied in today's announcement, it's about $25 million-$30 million on an annualized basis, going forward. Really more starting next year on a full run rate basis. Why is now the right time to make these cuts and refocus the company a bit more?
Veritone today has, I'd say, a certain percentage of our revenues is relatively very stable. It's like our baseline run rate with, I'll say, you know, kind of almost locked in, secure, you know, growth and, and X margin. And I'm not gonna give you that exact breakout. VDR, which is an exciting new lines of business, and even the Fed, they're hard to forecast. VDR, Veritone Data Refinery, unlike our SaaS subscription-based businesses, is all consumption-based. When we do a deal with Meta, there is no subscription monthly or quarterly or annual. It's literally they have a demand, we have to prepare datasets. These the size of these deals are, at the lower end, 6 figures, some of them, on average, 7 figures. Relative to our revenue base, right, it's a huge swing.
What we've decided to do, and I'll tell you why it's the right time and why we can do it now, is why are we chasing our losses so much? We have a great operating business. We have killer accounts, right? We have great entry points in public safety. We are going to reduce our baseline operating expense by up to 30% to get us that close to profitability even when revenues are moderate in our baseline business. What I mean by that is, when we do deliver the VDR and I'd say some of the more challenging to time when Air Force and some of these government contracts will ramp, we don't really care.
We've corrected our right size of our cost structure. Therefore, we can reap the benefits on the outsized returns when the variable revenue happens. Why we can do it now is, first of all, we've been talking a lot about the technology platform. Pretty much all of our applications now are, I would say, on K8. It's not to get too technical. We finally have transitioned everything to Kubernetes across the board. We're a lot more efficient in terms of the platform and the application layer. We can be much more efficient on how we're managing the applications and the entire infrastructure. We're finally at a point of maturity where you're gonna see more consolidation in sales and marketing and G&A.
To be clear, we are not doing anything that in any way is gonna impede or impair our ability for growth or in any way impair our ability to drive these new exciting lines of businesses. This is, I would say, almost obligatory based upon what we've been able to do. I'm not gonna say it's way overdue, but I think it's time for us to introduce these things to bring that cost structure down.
Awesome. All right. Well, with that, I think we can call it a day. I wanna thank Ryan and Veritone.
Yeah
for the time today.
Thank you.
Thank you.
Appreciate it.