AIエージェントが訪問の5割を占める時代、データ基盤は備えているか｜Tealium CEO ジェフ・ランズフォード

We're still early days in agent adoption by enterprise, certainly. The future is going to be wild, but humans are not going to go away. Your traffic is going to grow. Probably 50% of it by mid-next year will be agentic, but that doesn't mean your human traffic got cut in half. In agentic commerce, there's a gap right now because agents need current state consented contacts on them. and a warehouse can only offer yesterday's picture. This agentic traffic is converting at about 38% more revenue per session than the human traffic is. Our customers today, about 20% of their inbound traffic is inbound agents, not humans. Even before AI, 40% of their traffic was bots. What used to be human traffic is now agentic traffic. When you go to a website with your browser, there's a hidden layer of data. The human eye never sees, but that's right where the agent goes to do its thing.

First of all, describe Telium to listeners, where it is in the stack, what it does, I mean, the software that Telium's making. Sure. So Telium is a neutral data collection platform that runs underneath all of a company's customer experience systems, So web, mobile, call center, point of sale, branch store, ATM kiosk, and above stack from the cloud data warehouses and the back end transactional systems. So we're basically a listening and collection and enrichment and distribution layer. And we position the company neutrally across the industry. So we integrate with about 1,300 different vendors across MarTech, customer experience, data warehousing, transaction processing, and help our customers just listen to all this real-time data, resolve it around identity, create enriched visitor profiles, and then federate those back out in real-time to all these systems that can benefit from that real-time view.

Is it essentially an orchestration platform or orchestration layer? Hi, the world is getting increasingly complicated. Every time your team deploys a new cloud workload or AI agent, another identity gets permanent access to your critical systems. Did you know 9 out of 10 organizations suffered an identity breach last year? Because adversaries are logging in with valid credentials, not breaking in. Non-human identities outnumber human identities 109 to 1, creating a massive, unmanaged attack surface in cloud environments. Legacy tools were built to manage human employees, which leaves modern machine and AI access largely unmanaged. That's where Edira by Palo Alto Networks comes in. Human, machine, AI, one identity platform for all. 96% of human users operate with excessive permissions they don't actually need.

Adira replaces permanent permissions with dynamic access so you can lock down every identity without slowing down your business. Adira provisions access on demand and revokes it immediately when the job is done, which removes high-risk permanent permissions. With EDIRA, you get unified control by replacing siloed IAM, Identity and Access Management, PAM, Modern Privilege Access Management, and governance tools with a single control plane across human, machine, and AI identities. Secure every identity at your organization with Edira by Palo Alto Networks. Learn more at paloaltonetworks.com slash I-D-I-R-A. It is an orchestration layer. It's data orchestration. And so, you know, and we can do customer journey orchestration through data orchestration.

And so as the visitor profile is enriched, you might fire off different experiences through different systems and use this as the data orchestration layer. And we also, though, are a very open architecture. And sometimes we have companies that just use us for data collection, filtering, and enrichment. Other times they use us for identity and visitor profiling and complete orchestration. And in all cases, we're sending data back out, you know, what we call activation of various customer experiences or various data endpoints that can benefit, which obviously the name of your podcast and the most obvious beneficiary of great data is AI. And so, you know, over the last three years, we've just seen a huge upswelling of customer use for AI use cases. Yeah. Most of the people I've spoken to about orchestration layers or orchestration software are talking about managing agents. But this is broader than that.

Is that right? Yes. But you also orchestrate agents within a workflow? So we do not grab a hold of an entire agent and orchestrate it. We just feed agents data and context and fresh real-time context. So trusted context where you know the consent status of the visitor. So what we are doing is collecting data, labeling data, filtering data, enriching data, and then feeding it into AI models. And sometimes that could be an agentic use case. Sometimes it's just a fraud algorithm. You know, sometimes it could be a next best action algorithm. The term data orchestration itself is morphed as AI is taken over. What does it actually mean in the modern stack as an architecture and not as a buzzword? Yeah, sure. Which I'm guilty of. So if you think back, I don't know, 10 years ago, we really started building this actually in 2012.

what data scientists were doing to get great data was they were running batch data pipelines. And they would just set up a series of batch backward-looking data pipelines. They'd pull data together. They'd run an ETL process, and then they might send that new pipeline somewhere else. And that is a cascading, backward-looking, expensive way and suboptimal in our view because what happens is you get Tuesday's data on Monday, right? And now roll the tape forward to an AI agent, and you don't want to give that AI agent context. It's 24 hours stale, right? And so not only is it suboptimal, it's much more expensive. So when we invented Telium, we said we looked at like TIBCO, which was message bus in the mainframe era, WebLogic, which was a neutral app server in the Internet era, And, you know, we said enterprises need a real-time data orchestration layer where instead of doing pipelines from each of these, you know, the average company uses 200 different software packages, and I'd say at least 30 contribute to the customer experience.

So instead of just running pipelines from all those 30 systems, let's just listen to the event-level traffic created from each of those systems in real-time. And then let's – it's basically real-time ETL. So we take the event, we resolve it around identity, enrich a profile, and then send it on its way to wherever our customers want to send it. And that may be to, you know, an open AI conversion API or Google or Meta or TikTok or Reddit for advertising. It might be to an email system or for audiences. It could go anywhere. And back to the data warehouse for analytics. and so that orchestration of those real-time events, I think the term orchestration really comes into play when you flip from batch to real-time because you're literally catching events in real-time, orchestrating where they go, what they look like, how they're labeled, and then sending them on their way. That requires coordination. It requires a hardened enterprise system because many of these software companies

and the Internet itself are not natively and truly real-time to the first bite. And so you have to harden it. So if we send an event and the other vendor doesn't catch it, then we need to know that so we can resend the event, right? So otherwise it just drops on the floor and you have data loss. And so that's a lot of the orchestration is think of it as trust, reliability, and observability, taking an event from point one and setting it to point two. And that's part of the orchestration. And the other part is, as I said earlier, like you can enrich that event, that payload, and it goes from raw event data to a robust visitor profile, whether they're a known visitor or unknown. Now you can create some crazy experiences that are going to convert better for that visitor. Can you give us a use case? Walk us through, you don't have to name a customer, but a typical customer and how they're using it.

Certainly. So let's take a large retailer, global retailer that has multiple brands. And that retailer wants to capture customer activity and prospect activity across their website, across their store traffic, across their mobile app, and across all four brands. And so they will configure Telium as the data collection layer. They will listen to event-level data from all those different properties, digital and physical. They will enrich the visitor profiles in real time, and then they can enhance an experience. So if someone's looking at something on their mobile phone in the mall, and then they walk into a store, and now these retailers have, you know, iPads for the store owners, I'm sorry, the store employees, and says, oh, yeah, Mr. Lunsford just walked in, and he was just looking at, you know, men's 34-inch jeans or whatever.

And so you can inform an in-store session with what just happened digitally. You can inform a call center session with what just happened digitally. You can put people in email, people that, let's say, abandoned things in a shopping cart. You can put them in a retargeting campaign, whether it's retargeting of display ads or whether it's a follow-up, you know, reminder with a coupon offer via email. And then you can take these enriched profiles and you can also send them back to your cloud data warehouse where you might run big workloads and break them into audiences. And then, you know, take an entire audience and you would use a tool to extract that audience and go activate that. That audience might be, again, for a digital media campaign. It might be for first page personalization. It might be part of a loyalty program. And so that particular customer uses us for basically everything I just walked through. I mean, this idea of real-time that a shop clerk could get an update while you're in the store,

that's from your online activity, right? Yeah. That's pretty powerful. Can you talk about why real-time? I mean, that's an obvious example, but why real-time is so important today? You know, real-time experiences convert at 30 to 200 percent higher rates than delayed experiences. Another example is an airline. You know, when you're traveling in the middle of a trip and your airline is telling you what gate, what baggage claim, you know, all of that is in the moment real-time customer engagement. And, you know, a sporting venue is an example. We were just at a sports league day in New York, one of my team members. And, you know, so all, if you think about it, almost all digital experiences will be higher, more engaging and higher converting if you're using fresh context, period, full stop.

You know, and I think AI has just shined a light on that. We've had data since 2012 that, you know, So if you send an email four minutes after they leave the website versus 24 hours after, the open rate is like 6x. You know, it's crazy. And so it's intuitively obvious to us. And it's interesting over the last decade, people would debate that, you know, you don't need real-time data. And, you know, I'm a pilot, so I'm always thinking I do not want to fly my airplane knowing what my fuel status was. I want to know my altitude right now and what's in front of me. So it's kind of obvious to me. In the case of someone doing a search for genes on their phone and then walking into a store, what's the data flow? I mean, is it the platform that you're looking at that then is captured, that data is captured

by I assume it the store is the customer that using Telium What are the different channels for them collecting data So the data flow is basically we're always listening to all of the systems if a customer has us fully deployed. And any interaction on any of those systems creates events. Those events stream into Telium in real time. And then our customers and we help them, you know, create use cases of, you know, I want if someone, an employee in the store pulls up the iPad, someone's profile, I want to populate it with the most up to the millisecond activity. And so that would be a pull from the store system out of this database, or it might have been a push because our customers can also set up rules where they say, anytime an event, anywhere happens, send it to the in-store because those in-store experiences are so valuable and high converting,

and we want them to have larger checkout volumes. This is something that everyone asks. You know, we've all had that experience of talking about something and then getting served an ad for it. Is that part of this real-time data flow, or do you have any insight into that? That definitely is, and so the listening capabilities would just be one of those customer experience systems that I talked about that we would be listening to real-time events, and it would be up to our customers to map from that software. If there's a vendor that sells that listening software, we would build a connector to that vendor and receive those events. A lot of times that's kind of happening in a walled garden where perhaps the device owner also who operates the OS here, you know, also might be interested in advertising to you. And so that knowledge might even just stay client side on the phone to be privacy compliant.

But there are all sorts of wild things going on out there in the digital space. But what that is is just classic intent targeting. So I can divine your intent, whether it's listening to you verbally, listening to the click stream of your website behavior, you know, listening to the clicks on your mobile phone. I'm still listening to that data. And then I can infer intent, and then I can go try to engage with you at the point in time when you have the highest propensity to convert. A question, you know, I've been back and forth with your team talking about why the modern stack needs to support real-time and event-triggered activization at once. And they said instead of forcing teams to pick one architecture, What does that mean instead of forcing teams to pick one architecture? Well, we definitely handle batch-delayed data because a lot of companies only have the data in batch format.

And we'll take data as fresh as we can. And so we don't go in and say, you have to be real-time. We go in and say, you know, let's get the best data you have and resolve identity and profiles around the best data that you have. and we also are, there's a big kind of debate in the industry right now of, you know, should your audiences live in your cloud data warehouse? Should they live in your CDP? Should they live in your marketing cloud? And our view is we can operate any of those three ways, right? And the highest and most valuable use of telium is not that we have to hold the audiences, is that we glue together on a trusted fabric all of these systems, you know, and we do it in a very trusted, privacy-compliant way, and we do it with high performance around the globe and with effectively zero data loss, you know, enterprise-grade reliability.

Yeah. And so how do you activate data that lives in Snowflake or Databricks or BigQuery, Redshift or one of these in real time without creating a shadow copy or a second source of truth? So we have a capability called CloudStream, and we'll go extract an audience out of the CDW and activate it in real time out to, you know, whatever activation point they want to go to. And, again, those rules of where does that audience get sent and for what campaigns are defined in our platform. But it's going to be triggered off of some inbound event. And if this inbound event happens, then go pull this audience and activate it out here. Yeah. Is this primarily for, I mean, I would imagine it's not, but the use case you gave was a retailer.

What are some of the other enterprise use cases beyond retail? There are many. So, you know, we have like 10 of the largest banks in the world, Eustelium, six of the 10 largest telcos, three of the five largest airlines. so massive companies across all industries, you know, in the banking space, when someone is going to apply for a home equity loan or the insurance space, when someone is going to purchase a new car and they need auto insurance, they're going to transact very quickly. It's a considered purchase, but there's only a window. And then auto dealerships. We have a lot of auto dealerships used to them. And, you know, when I'm searching for that new car for my daughter who's coming home from college or used car or whatever, you know, I'm going to probably get that done that Saturday because I've allocated the time to do it. And all industries benefit from real time.

And it just it kind of makes no sense to engage in a debate that, oh, no, delayed data is better. It's slower. It's more expensive. It requires more storage space because you're storing it in a whole bunch of different places. You know, it just doesn't make sense in our view. Yeah. And so do you see these systems all becoming real-time? I mean, all data flows converting to real-time, or I would guess there are use cases where real-time is not important. You know, the irony is when I got out of the Navy in 94, I started working in bank technology. And as you know, banks, you know, process checks through these big reader sorters every night. And they run a batch process through their deposit system on the back end. And that's a classic batch process. Well, in I think 95 or 96, people started building real time core banking systems.

But the industry never upgraded to still today. You have batch processes with all these large banks. And so, yes, you can get a real time transaction done today, but there's still batch processes behind the scenes. And, you know, I think there's inertia there. There's just, you know, this is a system that works. And so that's why we found ourselves, well, obviously, except batch files, we do a lot of file import into Telium. But we also preach real-time. And everything we do is a complementary layer. We don't really replace anything, Craig, other than point-to-point integrations. So we work with all the big Martech folks, and we just complete the last mile of these real-time use cases, and we make their systems better because they get better data. And we'll go in and replace any of them. We'll just replace the point-to-point integrations, which is expensive to maintain that as well.

Yeah, I mean, there are vendor-imposed schemas, right? Why do those become bottlenecks as stocks modernize, and what does it take to activate data? A, let's say an email vendor, you know, they create a schema and it's optimized for email. And, you know, let's say a website A-B testing vendor has optimized their schema for that purpose. And we built an open schema so that you could map all of the different elements or attributes into this open and flexible schema to create a superset of all the attributes you want to track, you know, about your visitor's behavior. And, you know, Palantir calls this an ontology, and they're doing it kind of at the enterprise level,

not just customer data, but all of your data. We specialize in, I don't want to just call it customer data because it's really visitor data, prospects and customers, but you basically create a comprehensive ontology about your visitor behavior, and then you map the attributes from all these different systems into that so that you get this comprehensive view. and then all the systems can benefit from seeing what the others have seen, right? Yeah, and I would imagine in financial data this becomes critical because you want – otherwise you're going to end up with a gap between the data that's exposed to or displayed to a user and the actual data at that moment. Is that right? Yes. We're typically not doing, like, financial account balances. We're helping create an experience for, let's say, a loan application or an insurance application or something like that.

You know, we're more top of funnel stuff. And then the back-end transaction processing systems will populate balances. But basically anything behavioral is where Telium operates. and that behavioral data can then help you with all sorts of things. Like I said, pricing, next best action, fraud, personalized content, you name it, all of which leads to higher conversion. Yeah. I asked about agent orchestration at the beginning because the use of the term orchestration layer in most of my conversations is about orchestrating agents or switching between models and that sort of thing. But in agentic commerce, there's a gap right now because agents

need current state consented context on demand, and a warehouse can only offer yesterday's picture. So what does it look like when orchestration is done right? Well, why don't you answer that first? Yeah, sure. So our customers today, about 20% of their inbound traffic is inbound agents. not humans. Even before AI, 40% of their traffic was bots. And so the 20% is really 20% of their good traffic. What used to be human traffic is now agentic. And agents still come in through the website. So what we're explaining to customers is like, you know, agents aren't killing the web. If anything, you know, the website is more important because that's the front door through which the agents access you, but you want to be able to speak their language.

And, you know, whether it's A to A or MCP or A to C, what we help our customers do is prepare their data layer on the website for that inbound agentic traffic. So there's a hidden – when you go to a website with your browser, there's a hidden layer of data that the human eye never sees, but that's right where the agent goes. to do its thing. And, you know, we can pull data. We call this data layer enrichment. So we can pull data out of the back end where the visitor profiles are. And you have all these attributes about your visitors. And we can pull it out of the CDW. And we can put it in the data layer so that when the agent comes in, they have everything they need right there in the data layer. And then you'll get higher converting traffic. And this agentic traffic is converting at about 38% more revenue per session than the human traffic is.

So this is really important traffic. And we're just helping our customers optimize their data layer. Agents like JSON-LD, which is basically kind of a pre-configured definition of what data elements means, like pre-labeled data. And so you can put that on your website and then the agents are going to like you And you want to be one of those websites where agents can get accomplished what their users ask them to accomplish That means they rate you higher and rank you higher the next time around Yeah, and you're talking about websites that agents contact or systems that agents, external agents contact, But you're also talking about companies running agents and feeding them real-time data. Inside the enterprise, obviously, you're going to have AI models.

Sometimes they might be agents that are doing a specific task like fraud, you know, next best action. Whether you call that an agent or a model or a skill, that's up to you. And then, you know, that's part of the agentic enterprise. And then just the fact that your customers are going to be coming asking their agents. They're going to entitle their agents with their username, a certain budget to act, certain authorized activities to come do their bidding on your website. And you better be able to handle that. Otherwise, that and the customer will pick and or its agent will pick a different vendor where the agent can do their bidding. Yeah. Yeah, we're still early days in agent adoption by enterprise, certainly. And what did you say? Did you say 20% of traffic is agentic now going into?

How do you see that changing, and what does that mean for Telium? Well, we think it'll be 50% a year from now. and you just look at the slope and the trend and how much more usable these agents are becoming. People are going to be just talking to their agents on their phone and saying, hey, go look up this trip for me, go book dinner for me, and the Internet's going to disappear into the walls. So, you know, the idea of a screen and being required to access the Internet or information or transaction capability, it's always going to be present, but you're also going to be able to do it just through voice. And in that voice capability, it will probably access the same front door. Right. It just won't require HTML rendering. and yeah we just see you know the the future is going to be wild but humans are not going to go

away right your traffic is going to grow probably 50 percent of it by mid next year will be agentic but it's not that doesn't mean your your human traffic got cut in half it means there's just more activity because it's easier to do things now it's easier to send your agent to go off and do work and search for things. And so we think, you know, net-net overall sessions are going to go way up. Something like half of it will be agentic. You can't forget the human interface, though, because they still want personalization, still have to track consent, and all these things that you have to do today. So really, AI and agents are just another delivery channel on top of all of your other existing channels. Like, they're not replacing anything. And this is what's been going on, you know, since the beginning, when, you know, first you had brick and mortar, then you had phones to add to that, then you had IVR, and then all of a sudden the internet came

along, and all of a sudden mobile came along, you know, just as one layer, and none of these, brick and mortar is still there, right? And so these things are just layering on top of each other, and the best-in-class companies are going to engage their customers in each channel, right, and they're going to give them a consistent experience across all of those channels so that the channels speak with each other back at the data layer where Telium exists and understand what has happened across all those channels. Yeah, and you mentioned something interesting that that you don't need the HTML rendering because agents are talking directly to the systems. I had a fascinating conversation with a guy about clinical trials on new drugs, which is, you know, it takes a lot of work, and you have to produce all of these documents and the trial plan,

And you end up with, I don't know, how many pages and pages and pages of documentation that you send to the FDA. And then the FDA just feeds it all into an AI. So people are saying, well, pretty soon it's just going to be sending data to the FDA's, your agent's sending data to the FDA's agent. And a lot of that stuff will happen in real time. Do you see that happening in other areas, you know, where you don't need the papers? You don't need the PDFs. You don't need the HTML rendering. Well, yes. I think, you know, we've just seen increased digitization of everything over, you know, since I've been around here since, whatever, 32 years.

And, you know, I remember sitting at overhead projectors writing with acetate, right, and then came PowerPoint. But then you would print your PowerPoint slides up and you would print them out and put them on an overhead projector, right? then came you know projectors that could take the connect to your laptop and you never print it out like so look everything is getting digitized including forms hopefully for the fda so we can get more drugs faster and save lives which ai you know i think the coolest thing about ai is it's going to cure a lot of diseases and yeah create uh save a lot of lives and um in our world of digital marketing, you know, since Tullium's inception, I would say we're not replacing paper, but we are replacing batch processes. And we are replacing point-to-point integrations,

which have to be maintained over the years by people. And if you remember math, this is what's called an infactorial problem, right? So if you have, you know, two systems on the front end, like it used to be your call center and your internet site, and then two systems on your back end, like maybe your transaction processing system in your data warehouse, well, then you need four connectors, right? And so if you start going to where you have 10 across and eight across the back, and now you have 30 across the front and 12 across the back, you know, that simply the number of permutations, the number of point-to-point connections you need to have becomes unsustainable. But that's how corporations were growing. And then they would have teams of consultants who would build these connectors who would then go to a different company, and then they wouldn't have the corporate knowledge.

And so that's the most, you know, sort of the biggest transition from old to new, you know, is that we've said, gosh, let's just come in and replace all that with an elegant real-time data layer. And your life will become so much easier. Your conversions will get better. Your support costs will be lower. Your processing costs will be lower. Your storage costs will be lower. Real-time is a better way. And it's been fun. But that's really what we've been doing. Yeah. Yeah. Is there a domain or industrial vertical that you're most involved in? It's financial services, health care, automotive, retail, telecom, technology, really both B2C and B2B businesses. You know, I used to joke, everybody but oil field services needs a real-time data layer, you know,

and I think that pretty much holds true. Travel and entertainment, we've got a bunch of travel customers, so really across all industries, pharma, health care, insurance, you name it, we've got great customers across all of them. Yeah, and where does the telium layer sit? Is this a SaaS platform that people are hitting by API? Yeah, we're cloud-based, software as a service, and we do have customers who will run it. They will host the software in their domain, but largely we run and process the data in the cloud, and most of the data that we help collect originates from outside the firewalls to begin with, right? So we're like a stopping point on the way back into the enterprise, four walls. And with all the security and auditability and trust we have, it's the right way to do it

because then we can get the economies of scale processing, you know, 20 billion events a day in real time around the world on behalf of all of our customers and operated that way. Well, that's also an interesting question. You're not – your user base is global. It's not simply – Half of Telium's business comes from Europe and Asia and about half from North America. So very global. and, you know, like I said, these big telcos and airlines and banks and, you know, entertainment companies, retailers, they're all global that we work with. So they really like having a partner that can collect data. And there are a lot of data residency requirements around the world so that we can keep sovereign data local when required by law. Yeah, yeah. And you're agnostic as to what operating system or, you know, what the architecture is around.

They all run, every one of our customers has a different stack. None of them have exactly the same mix of vendors or geographies or platforms on which they run. And so we connect to all of it. Yeah. With Snowflake, when you say activate Snowflake data in real time without a shadow copy, what actually moves? I mean, something has to leave the warehouse. Yeah, so you take a profile out of Snowflake and send it to, let's say, Reddit, you know, for an audience activation or something like that. And you might take an entire audience or you might just take a profile, and depending on what the request was and what the rule is that you define in the platform. Yeah.

When an AI agent shops on a customer's behalf, who is the customer in your identity graph that you've built? Is it the person or the agent? Yeah, it's interesting. There was just a lawsuit today where that topic came up. Oh, is that right? Yeah, the court ruled that the agent really was the person. The person was responsible for the agent. So for telium, it would be a way to create a new attribute at the human visitor level, and then you would say, do they have an agent or not? And then that agent, this would be their agent. You might have an agentic identifier. And, you know, we're working on all this in real time with our customers. How would they like to see it, right? Because we also could give the agent their own identifier. And you can set rules up for how all that happens in the platform.

Yeah. Yeah, give the agent. That's something that would happen at the agentic, the vendor's side. It's not something a shopper is going to set up. Right. Correct. No, the vendor would decide, you know, there's a concept of householding, as an example, and, you know, or in B2B businesses of associating a person with a business. And so the same idea, I think, would apply where you would associate an agent with a human, potentially as part of a household, but you'd probably clearly know they were an agent. Yeah. You know OpenAI has this thing called instant checkout that compresses the funnel to almost nothing If agents are doing the browsing what happens to the personalization industry that built on browsing behavior

That's where you go back to optimizing your data layer for that agentic checkout, right, or that agentic experience. And, yes, if you have all the right data surfaced for that agent already through the API and the protocol that you'd like to support, that agent will be able to come in, transact, and leave faster than a human. And what is the point of a website other than to get conversions? And, you know, sometimes it's to educate. Typically, it's to get some type of conversion, like a loan application done or, you know, there's a lot of big pharma companies. They're trying to educate people about their new medications to different audiences. And, you know, so I think the instant checkout is just part of the future. The human traffic won't go away.

You will get inbound agentic traffic that might want to use instant checkout. And if you have an optimized data layer for them, then you'll get higher conversions out of them than if you make it difficult for them. I would not fight the trend of agentic instant checkout if I was a company because you would then end up losing an entire generation. You know, my 20-year-old young adult kids, they just wouldn't use you, right? If their agent can't do it, then they're not going to use you. They'll just go somewhere else where their agent can do it for them. Yeah. The AI natives are going to demand that kind of service. Yeah, that's right. I have adult children, too, in their 20s. They're very impatient. Telium started as a tag management company for the browsers. Is that right? Can you describe what that is?

Yeah, so for sure. When we started the company, think of around 2011, and what had happened is AdTech had kind of really gone through this Cambrian explosion of different things like SSPs, DSPs, banner networks, so on and so forth, personalized content, A-B testing. And so companies were, you know, putting during Internet 1.0, you might have one or two pixels on your website that use JavaScript to read the data layer. And that's called a tag. And when we started Tilium, we wanted to solve the whole omni-channel data fragmentation problem, but we had to pick somewhere to start. And we decided to start on a website because we knew it well. We were a former web analytics team. and we built a platform. Our customers had gone from an average of two tags to 15 tags. And so then their AdTech and Martech teams were having to support all of these JavaScript tags on their website,

which became a big management problem and keeping a current problem. And so we just abstracted that to one common data layer. We manage all the tags. They could do that inside our platform. And that's how we started out on the website. And then we added, obviously, server-side for mobile. We added audience capabilities. You know, over the years, we've just added zero-party data collection. We've just added more and more to it. Yeah. Is there, you know, people hear a lot about warehouses, a lot about orchestration. Is there a clean line between them, or do they overlap? I mean, yeah. I would say the warehouses are building more and more capability. And, you know, when we think about a warehouse, that's for running analytics and running AI models and all that fun stuff. You know, we want to do the data collection part, stream data into that data warehouse.

And then we want to help them do the activation part. Like they don't want to go build 1300 connectors and maintain them the way we do. And so we can help them with the activation capabilities. And, you know, I think the orchestration layer is the most architecturally efficient place to put it is in between the back end and the front end. Well, you know, you go back to that in factorial problem I talked about because there is not just one back end. There never will be. Right. And there will never be one front end. You know, and so they're always going to be, you know, 20 to 30 on the front end and, you know, 5 to 10 on the back end. And so this orchestration capability should sit in the middle, in our view. You shipped an MCP server in June. What are customers doing with it? That's great. You know, our customers can use Claude or Codex or Grock, whatever, you know, to actually do anything in Telium that they used to have to log in and, you know, create rules, create attributes, create campaigns, create audiences.

They can literally just talk natural language now to their agent, and it can latch on to Telium via the MCP server and do it all for them. and it's the biggest unlock of the power of our platform I've ever, I could have ever dreamed of because, because we built a very powerful platform, you know, it took a lot of training to get people up to speak on how do I do this with Telium? Oh, wow. Once I, once I figured it out, it was great. Now they don't even have that, have to go through that ramp up in that training because the agent can just figure it out and you use MCP to latch on and make things happen. Yeah. I mean, I asked about use cases, but can you describe a deployment? You talked about the lift in conversion that this creates, the real time creates. Can you describe a deployment where this really clicked? Again, you don't have to name the customer, but what did the before and after look like?

Sure. So, I mean, we have just so many of those out on our website. You know, we've got ROI use cases galore. And we have customer conferences three or four a year around the globe where our customers literally stand on stage and talk about, you know, their ROI. And, you know, I think what I would say, rather than giving a specific example, so there's a couple areas where there's just obviously low-hanging fruit. One is the conversion APIs that exist in ad tech, so like the Google, Meta, LinkedIn, Reddit, Pinterest, Snapchat, TikTok. They have conversion APIs. And when our customers send data, they connect to those conversion APIs via Telium, their conversion rates and their return on ad spend go up on average 25%. And that sounds like a small number, but then you realize how much they're spending on digital ad spend,

which is typically at least 10x what they're spending on their tech stack, easily 20x. These are big dollars. When you can optimize your campaign spend by 25%, you get a great lift. We talked about the conversions. And, you know, we see in the travel industry, you know, 30%, 40% conversion improvement when we go from day-old data to real-time data. We just see it everywhere. And the other thing is the value of pulling together multiple channels of data. So one auto dealership, you know, an auto dealer has leads that come in through digital and marketing, and the average auto dealer across the U.S. converts about 10% of those leads to a completed sale. Well, you have one dealership that was kind of the thought leader that embraced Helium early, probably three years ago. And they optimized their data, cleaned up their data sources, created this clean data layer, and their conversion rate went to 30%.

So they tripled from 10% to 30%. And now the automotive industry is a zero sum in that, let's say, in that county, only so many cars are going to sell. So what that means is this forward thinking, digitally savvy dealership is taking share from the other dealers in their community. Yeah, not. And 10 to 30. Just think about that. This is a considered purchase, you know, 30,000 on average, 40,000 dollar purchase on average. And so that emboldened this dealership to go out and acquire other dealers because they knew they could underwrite the M&A once they could convert, you know, their 10 percent conversion rate to 30 or let's say 20 on average that, you know, they would get a great value out of that acquisition. Yeah. How do you guys charge? Is this, you know, everyone's going to sort of metered. Yeah, we charge by volume, and, you know, with our customers, we'll assess what their volumes are going to be.

We'll create an annual budget, and they'll commit to that budget, and they're, like, usage tiers if they go over that. But typically, our customers come in within that budget. We don't like for there to be surprises. That allows them to budget and get the ROI they thought they were going to get. Yeah, and on ROI, are you talking about revenue specifically, or I would imagine the benefits are also in speed and engineering time and things beyond? Yeah, it's all of it. It's definitely efficiency and speed and agility, the ability to try new things. Once you have that neutral data layer in there, you can try new vendors, swap them in and out and use who you like an open architecture stack. And then there is, you know, higher conversion rate is higher incremental revenue, greater ad efficiency, greater targeting efficiency is you can, you know, either plow your savings into more growth or you can plow them to the bottom line for more earnings.

Either way. Yeah. And you've been doing this, what did you say, since 2013? Is that what you said? Telium, we gave Telium, we funded it in 2011. In 2009, they started as a two-person consulting company, and then we really started building this software company in 2011. Okay. What's the biggest thing that's turned out differently than you expected? I mean, certainly, as you said, you started out in tag management and moved up to the data layer. But in the – I mean, obviously, a lot has changed. But is there something that turned out differently than you expected? I think the biggest surprise was the 150 other companies that wanted to copy us once we showed, you know, 12 years ago what a big need there was here.

and there were literally 150 other companies coming out of the woodwork saying, oh, I do what Telium does. And so it's like, wow. Man, the ZERP era was crazy with overfunding for VCs. But that is now pretty much subsided, and we're sitting here as the largest independent provider of what we do, delivering that trusted neutral data layer to the enterprise. and so with AI coming over the horizon and the great need for, you know, clear consented data, you know, I think our future is bright and we'd love to talk to anybody that's interested. Yeah, and what are you working on now? I mean, what is, you know, I would guess that your market is your addressable market. You know, you can spend the next 10 years addressing that market. But where is this going? I mean, what do you see on the horizon for – are you guys going to eventually merge with the warehouse so that you're doing the warehouse?

We're just building AI into our platform and allowing our customers to be able to talk to their data and listen to their data and do everything agentically. And that is being very well received right now. And then we're just building all the connectors for the AI ecosystem that our customers want, you know, like the OpenAI connectors, Google, Amazon Bedrock, Google Vertex, all those, so that our customers can get active and really embrace the wave here. Yeah. And in two years, if we talk again, and I hope we do, what do you think will have changed about how enterprises handle customer data? It's going to be wildly more agentic traffic and very much omni-channel, a lot of voice interfaces, and the smart companies that embrace it will have higher converting, thriving businesses. So it's going to be fun.