データに溺れていませんか？Inzata COO／SIGNLD Chief AI Officer のクリス・ラフターが語るナレッジグラフと意思決定インテリジェンス

Let's go. You're listening to Making Data Simple, where we make the world of data effortless, relevant, and yet even fun. Okay, podcast listeners, Al Martin here. Again, week after week, we do this every week, can you believe it? The title for today, this is how I'm going to kind of characterize it anyway, is the data revolution and why companies have more information than ever, yet they still struggle to make better decisions. So Chris, who I'll introduce here in a second, is an interesting guest because he sits at the intersection of business strategy, data analytics, AI, and revenue growth. His background at Inzeta will give us a chance to explore both technology and some of the practical business challenges therein. I'm here with Chris Rafter, who is the chief AI officer for Signaled. That is part of, and he'll explain this in just a second, of Inzeta, where he serves as the chief operating officer. Chris, thank you so much for being here. I appreciate you, man. Thank you, Al. I appreciate you having me on. Let's start with you. Tell us about yourself. How did you get in the technology business in the first place?

If you don't leave out any of the good stuff either, I mean, tell us where you started and what landed you here. Sure. I'm Chris Rafter. I'm COO and a co-founder at Zeta Analytics. And we recently launched a product called Signaled, and that may turn into its own brand, or it has turned into its own brand. So I'm also chief AI officer there. and it's a platform that's designed to help a company, turn a company scatter data into a knowledge graph that AI can actually reason over and use to provide context to help them get better value out of Signals AI or any other AI tools or agents that they might be using. Fantastic. But you're like a serial entrepreneur, aren't you? A couple, yeah. I started my career in the Fortune 500 in financial services, actually, and got pulled into technology. The last big bubble we had, which was in the Internet bubble, and got pulled in then and put in technology ever since then. Started getting involved in data about 25 years ago, building data markets, data warehouses, and things like that. Had a lot of experience across multiple industries, both big companies, smaller, medium-sized companies, as well as a lot of public sector work, working with state governments, state agencies, large and small K-12 school districts.

So I've seen what's different and what's common across different data environments, across multiple verticals and industries. Look, if I did my research correctly, you've been involved in just technology, operations, revenue, product, marketing, pretty broad collection of responsibilities. Which one have you enjoyed most? Definitely would be the customer-facing stuff. So when we get to go, I'm usually involved in the pre-sales side of things, but also in the post-sales and delivery. And my favorite moment is when a few weeks after the onboarding and the customer is getting results, and they're happy, and they come back and tell us what this meant for them, whatever it might have been. They're able to better predict graduation rates in a K-12 school environment and help students out or if they're able to save money in a large government organization or get people better services more accurately and on a more timely basis. Those are the moments that are most memorable for me. So then conversely, which one gives you the biggest headache? Definitely, there's a lot of education that goes with the job here. And a lot of times, you know, you're having to explain things.

And sometimes people aren't ready to hear the truth about, you know, things like data quality and master data management and how important that is to virtually any result, you know, any objective they're going to set. And so those are times, moments where you have to slow down, kind of meet them where they are, find out what they do believe and what they are behind, and then just kind of try to bridge that and move forward from there. So what's your brand? Are you a data guy? Is that what you say? Or are you an operations guy? I mean, it could be anything. I'd probably lean it more toward the data, product management, innovation type of thing. What I really love is talking to people, listening to them, and listening to them talk about their challenges and things that they find difficult and things they wish they could do differently, and then taking that back, turning it over in my head, talking about it with my team, and then coming up with solutions to those problems that they'll appreciate or get value out of. So I've got to ask you this before we move. No, that's perfect. That's perfect. I mean, it's your answer. As long as that's your brand, that's your brand. But let me ask you this. If you could go back and give your 25 year old self one piece of career advice, what would it be?

Oh, gosh, I probably it would have been it probably would have been to to jump into entrepreneurship sooner. I spent I spent a large, large, most of my career in corporate roles as an employee and moving into entrepreneurship, something I definitely regret not doing earlier. I probably I definitely would have gone and done maybe five or 10 years. But, you know, doing 14 was probably a little bit too long. I would have jumped over into I would have just taken the leave and gone into entrepreneurship much earlier. So why is that? What's the attraction? Money? Is it the is it more control, more broad experience? Well, I wish I wish it were money. The money's the money's no one's going hungry. It's just it's it's definitely the the autonomy and the just the stimulation that you get from doing lots of different things. You know, and one of the popular sayings in a startup is, is, you know, the thing is it's not my job. So whatever has to get done has to get done. But that's what I've always been that way. I've always been objective focused. I'd much rather be given a goal and be told here, go, go, go attack that goal. And here's some budget to do it. And I'll check back with you in six months than to be told. And I think, I think there's a lot of people like this as well that appreciate the autonomy,

like to think on their own and track their own path and then, and then be able to participate in the benefits when they achieve something. Very good. That's a good answer. Let's talk about Inzeta. Tell us the story behind Inzeta. What problem did you see in the marketplace that made you think, man, there's got to be a better way to do this? Well, sure. And it's still a problem that's out there today. and it may actually get worse. The problem we noticed and set out to solve a couple years ago was the fact that there was way, way more data being collected and generated than was being put into practical use to help support decision-making. It's been growing at an exponential rate, and unfortunately, companies' capacities to be able to process that data, get it into a usable state, get it into a data warehouse, a data lake, a data mart, dashboards, and ultimately in front of the people that need it to make decisions. This wasn't keeping pace, and that was the premise for the original Inzeta product, with Zeta Enterprise, which is an AI-powered data warehouse and data management module. So it automatically models data together, puts it into a usable state, and then allows users to very easily create dashboards and reporting off of that and basically make it available and turn it into something that's delivering value

versus just sitting in a silo somewhere consuming cloud costs. So that was the original premise for Zeta. Signaled is sort of the next iteration of that thinking. Once we started to see GNI technologies come out, It kind of was the missing piece for us. But it turned out being so different that we decided not to just make it a feature set on top of Nzeta. I know that we see a lot of companies out there that are just throwing on chatbots and putting in a little chat assistant in their product and then saying, you know, it's AI, Nate, it's AI, but it's not really. One of the first things we looked at was security because that's something that is going to – it's a big topic for everybody. And one of the first things we noticed about the approach of going with and bolting on an AI chatbot or something like that is it all relied on AI APIs. You'd have to go get an API from OpenAI or Anthropic, and they work really well. The issue is that our customers' data cannot go out over APIs like that. It's got to be held securely. And so we architected a different approach. We are running all of our AI in a captive walled garden inside of AWS Bedrock. And so we are still able to run the most advanced frontier models there, like Sonnet and Opus and Fable.

However, it never goes out over an API wire to another provider's data center. And so that allows us to give our customers the assurance that they can still utilize all of the benefits and features of AI without having to worry about where their data is going to end up. Is it being sent to somebody that they don't know or worse, being used to train models, which we never do? And that's something that we're very – that's a decision we made back in 24. and we're very, very happy about that because it's come up numerous times and we've had that decision validated many times by both customers as well as other companies in this case. Trust but verify is probably the biggest lip of saying right now, but I couldn't be more important. Verify, I would weigh heavily on even more than the trust piece. Anyway, the point is sharing data, I think it was a key decision. So, yeah, I think you made the right decision early on. So I got so many questions in what you just said. I noticed like on LinkedIn, by example, you're the chief AI officer of Signaled, but you're still the chief operating officer of Inzeta. Like you kind of referenced it just a minute ago. You said a lot of people just add features onto whatever.

Is this a whole different offering? Do you treat it as a different company? I mean, how does it work? Yeah, we almost treat it like a different company. It has some shared resources between them, but it is so radically different that we wanted to and we didn't want to constrain ourselves in the design process. and we wanted to be able to free our architects and designers to be able to build what the market wants right now in 2026, 2027, not just a derivative off of something that we had before. So we're still improving both products, but they're sort of forked into two different paths. I'm going to say this wrong so that you can correct me intentionally, but Zeta is there for your data to make sure your data is in one place, it's trusted, et cetera. then Signal sits on top and it does the data intelligence on top of said platform using native language via an LLM so you can ask questions, do whatever you need to do on that knowledge graph of said data? That's one way of doing it, but you can also just buy Signal. You can also just use Signal completely standalone. You don't need Nzata to use Signal. So you can bring in data that was inside of Nzata into Signal,

but what we're finding now is people bringing data from all kinds of applications into Signal, from QuickBooks to NetSuite to Slack to their operational systems, HubSpot. Microsoft Teams, those are all some of the data sources that people are bringing into Signaled. And the great thing about Signals is by pushing that data into a series of tables that are accessible by a knowledge graph, we can eliminate months and months of traditional data modeling or setting up of data models and data structures that we used to have to do when we're building an analytics application for clients. Well, how long does it take then? I mean, I presume it depends on the size of the data, the clarity of said data, the cleanliness of said data. But if I said, hey, I'm going to go work with Signal today and I'm going to work with Chris, what can I expect? Signal is designed to be able to work with a progressive delivery of data. And what does that mean? That means that, you know, people think they have to have all their data ready and cleaned on day one, and then a starter's gun fires and it's go time. Signal is designed to meet you where you are and be able to work with whatever data you have. So if you want to bring in what we find, and we've seen this with other within Zeta as well, is giving people the ability to have a progressive approach where we have over 800 connectors and MCP connectors to various data sources.

Most people will come in with maybe one or two key systems that they want to connect to. It's probably usually it's like QuickBooks or an accounting system or maybe it's an operational system or a CRM data. And they'll start with that and then they'll start working with it in signal. And the knowledge graph starts to get built as they're uploading the data. that's one of the other cool things is that the knowledge graph is not something that you actually have to sit there and build. It builds it from its interactions with you. So while you're chatting with Signal, while you're asking questions of the data, it's also listening to what you're saying and reading it into the knowledge graph and then validating those parts of the knowledge graph back to the user. So it's going to say, so you have seven departments or eight departments? Because when I read QuickBooks, it said you had eight, but you just told me you have seven. So which is it? So it'll ask provocative questions like that of the user, and then it'll remember the answer. It'll also remember which user gave it which answer. because, believe it or not, people don't always give consistent answers to questions like that. So the Knowledge Graph will track that all the time, almost like a blockchain. So we'll track every single change immutably so that you have a complete record of everything it was ever told in history going back to day one. But your question was how fast can you get up and running with it?

You can be up and running within an hour and start getting questions answered about your data. And it allows you to kind of come in at your own pace and either add more data sources, connect up maybe some of your unstructured data sources to it. But everyone's journey is a little bit different depending on what their priorities are. It's not just a data lake per se. Again, don't let me put words in your mouth because I don't want to mislead the listeners. But I heard lineage in there. So you got lineage. I thought I heard governance in there as well. Yes. I presume security didn't hear that, but you can tell us about security. I mean, you already said that it's protected essentially in the cloud. I think people – well, can it be on-prem as well? I mean, I don't – I'll shut up now. Now, it's available in the cloud, but there are parts of it that can be deployed on-prem, which we'll get to. But one of the things we've also thought about a lot is what does the future of dashboards look like? We know up until now how people have built dashboards and how they like to consume dashboards. And so we started to – and a lot of people are asking the question about dashboards. What does the future of those look like?

We think they have a great future ahead of them, but I think that they're going to look different than they were before because of some of the benefits that AI provides. So we working with the concept of a living dashboard or a decision apps because dashboards at the end of the day they code They code that was created by humans usually using a tool like Power BI or Tableau to lay out different graphs on the screen and set filters and metrics and things like that But underlying all that is code. And we know that AI is very good at generating code, and that's exactly what we set out to do. So we've been working on this concept of living dashboards inside of Signaled. We've got a couple of them in beta right now. But basically what this allows you to do, going full life cycle, you start using Signaled, you take in a couple of your data sources, you start building a knowledge graph from it. Then you can, of course, start asking questions at that point based on the data that you have and the knowledge graph that's there. But we said, what if we could use the knowledge graph, which is really this wonderful repository of not just the facts and the details and the data, but also the interrelationships, how departments work with one another, how business processes are connected to one another,

and how they're all connected to the underlying data. What if we could use that knowledge graph to be able to produce these types of living dashboards that are almost like a customized web application for the employees of the company, the executives of the company, and it allows them to ask any question that they would want. It gives them daily updates. It even allows them to spawn work tasks and assign them to different employees or even to agents, AI agents that would spawn from those as well. So it's a very exciting app. We're really excited about bringing this out and what people are going to be able to do with it. So is this on the same level, a Tableau or a Cognos? Tell me how it compares. Well, the first thing is that Tableau and Cognos still require a lot of labor and time to set up. You have to design the dashboards. You have to set up the data mark, the data lake, and all that stuff. Signal allows you to sidestep a lot of that. You can bring in the data directly from the source system. Signal does all of the data modeling and the matching to assign and identify the entities and their relationship within your organization.

In the old days, we used to call that an ERD diagram, entity relationship diagram. It would say, you know, Bob has a baseball and he gives the baseball to Sally, you know, those kinds of pictures. Signal does all of that in the background without you having to take out a pencil and manually draw those lines. It determines those relationships by looking at the data, the histories, and also this unstructured data inside of the organization. I'll give you an example because I know this is probably a lot different, but we do a lot of RFPs at Inzata, given our public sector leanings. And I asked Signal to go back to our Slack and to give me a recap of all of the entire RFP procedure from when we started first doing it six years ago and how it evolved as well as how many people were involved in it, what their roles were, what they did. And I was very impressed by its ability to just by going through, you know, years and years of conversations and thousands of individual conversation threads, it was able to distill out a fairly accurate picture of what our process for responding to RFPs look like over a period of six years.

And it identified three different generations of that process, you know, an early maturity, a mid-level maturity, and a mature process. And it was able to give you a lot of details around it. So that's just one example of how there's a lot of very valuable information, I believe, contained in many of these unstructured data sources. They're not using RAG at all. It's all knowledge. It is. Well, RAG had its time. And RAG is great for – we do use RAG for, for example, processing or for parsing and processing documents on a one-time basis. I think RAG has some deficiencies when it comes to trying to do what we're doing, which is giving people consistent, accurate, and traceable answers. Because one of the issues with RAG is that it's like a librarian that you send into the stacks to go get an answer for you every single time. And depending on the path they take through the stacks, they might give you a different answer back each time. And that's something that we found our customers weren't big fans of. They want to be able to have consistency and traceability, especially when it comes to, you know, financial data. You know, what was revenue last quarter? What's my profit margin on this product line? They don't want their, you know, 98% accurate is not good enough.

It needs to be 100% all the time. So I know Signaled is distinct from Inzeta, but do you, does Signaled come with Inzeta if you buy it or no, you'd have to, if you wanted basically a data lake experience, you need that, you'd have to get Inzeta in combination with Signaled. They're available as a package as well, but like I said, you could use either one standalone. They're not dependent on one another. So we have people who are using just Signaled right now, and that works great for them. If there's parts of Inzeta that they want to use, then they're able to do that as well. What's Signal built off of? I mean, what's the underlying, if you could explain the technology underneath? Outside of, you already explained some of it, like knowledge graph and stuff, but I'm just curious. Well, I'll just shut up. I mean, you know what I'm asking. Well, it's custom coding, and it's proprietary code that we have for most of it. We do use DuckDB to store some of the data as it's coming in. We also have a number of connectors that we have built ourselves, as well as rely on connectors from people like Fivetran who publish connectors, so we can also use those.

So we have over 800 different data connectors to connect to enterprise data sources. Many of those come from the Fivetran relationship. The knowledge graph is really the center of all of it, and we do also have our LLM and chat features, and those are powered by the LLMs that are running inside of our walled garden at AWS Bedrock. Can you use any LLM you want, or is it that proprietary too? It's only the LLM that you're putting in the background. We do have a selection of LLMs that people can use, but they would only be the ones that are offered by AWS Bedrock because we have a no APIs policy. Why do you have a no APIs policy, can I ask? Sure, because it goes back to the security. We're a HIPAA-compliant entity. A lot of our customers are storing very sensitive data with us, and it's much more reliable security-wise to be able to say that the data was loaded into our AWS cloud environment, and then when a call goes to an AI, it's going to an AI that's running on the server, like literally right next to it, that can then return the response. Like if you have, you know, if you're running your LLM locally,

so you have your MacBook on your desk and then you have your Mac Mini that's running an LLM right next to it, and the entire closed loop there is your computer. Your question goes to that Mac Mini. The LLM is running on the Mac Mini. It gives the answer back. You don't even need the Internet in that case. I like how you position safety. You position the fact that you're not sharing data. That data resides within the client's. I presume single tenant environment, I don't know. But let me ask you this. How do you ensure end-to-end governance, meaning the guardrails are up, you're protecting the data, encryption, all that, but you're also protecting prompts and prompt injection, that kind of stuff? I mean, what are you using to make sure that the end-to-end security is paramount? We have a multi-factor authentication and an Auth0 on the login side, so you have to have a login into your signaled environment. And then everything in the environment is encrypted, especially the customer data assets that are there. Then in addition to that, the knowledge graph, which is expressed in a series of documents that describe the different rules and policies and metrics that make up the customer's knowledge graph, are also present in that same environment and also encrypted.

So we try to limit the number of actors in this sequence here. So when you're working with signals, you're logging in. Your data is housed in the same place as where the LLM that's answering your AI question and reasoning for it for your answer is running in the exact same environment. not going out to an API to a server someplace else that we may not know. Did you help build this personally? Yes, I prototyped it. I did some of the coding, but I'll have to give the credit to my dev team for making this a reality. Well, I can tell because you're very knowledgeable. You can tell those that are really in the code and those that are adjacent to it, if you will. Clearly, you know what you're talking about. Before I move on, I wanted to ask one other thing. It's related to a previous question I asked. I previously asked, hey, what problem did you see in the marketplace that brought you here? But I guess the follow-up question is, was there a particular customer experience or a light bulb moment that went off that convinced you that, well, you already had Nzeta that, oh, we need signals. I mean, we've got to move. We've got to take this next step. What was the light bulb? Yes, most definitely. There were a couple.

One that I remember very vividly was back in the early 2024, and we had reached the finalists within Nzeta for an organization that was looking for a data warehouse and dashboard solution and all of that. and we sounded like a good fit for them. And actually, they did end up going with us as well. But I remember in one of the sort of final presentation meetings, there was one of the folks in the room, we were demoing to her how you build dashboards within Zeta. And she said, well, why do I have to do that? I said, what do you mean? She said, why do I have to move the mouse around on the screen and drag and drop the widgets into different places of depth? Why can't I just use AI and tell it what I want? And I didn't have a good answer for her that day. The answer at the time was, well, this is the way it's built. But then when I went back and thought about it, I said, you know, she's right. I mean, why should I have to do that? And so that became a major motivator. And it also told me something about the market. Market expectations have shifted dramatically in the last 18 months. And, you know, we're wanting to get ahead of that and be where the customer wants us to be.

And that was why we started prioritizing the development signal even all the way back then, because we saw a time when people were just seeing this automation and this forced multiplier effect from these technologies. and they're just going to come to expect them from their everyday software tools. There's usually always something that serves as a trigger. That's a pretty good story. I want to key on the data here for a second. As you mentioned, everybody's talking about AI agents, large language models. What I do find interesting, and I have my own point of view on this, but I want to get yours. Good news is I'm the host and you're the guest. But most companies still struggle with getting their underlying data in order. Or is your experience that we're building the future of AI on top of a shaky foundation? A shaky foundation is essentially the data is a mess. Is that your thesis or what? You're about two years too late with that question. If you caught me 24 months ago, I would have said absolutely that data's got to be pristine and clean and people really need to worry about this. I would have really embarrassed myself.

However, I will say now that my mind has been completely changed 180 degrees in the opposite direction by evidence. What I've seen is that the AI is incredibly agile and adept at being able to navigate less than perfect data. And so one thing that we found with Signaled is that we're able to take in data, unstructured data, dirty data that had been there, that's been that way for the longest time. And the AI is able to still make sense of it despite all of those sins, if you will. and we've been very, very pleased with the results. So I had my mind completely blown and changed over the past 24 months of seeing what these AI technologies can do. And what we did is to take that a step further, is signaled when you're actually, when you get past the knowledge graph part of it and you start actually asking questions about your business, it's not making an AI call to go out to an LLM and use RAG or something like that to come up with an answer. The bulk of the actual answers it gives you, especially in the living dashboards, are deterministic, meaning that the data is fixed, the metrics are fixed, and that's what gives us the consistency of the experience

so that if you ask a question on a Tuesday and then you ask the same exact question on a Thursday, you're going to get the exact same answer as long as the data hasn't changed. So we've been very pleased with how the AI has performed with dirty or less than perfect data to be able to turn it into reliable answers. You know, the funny thing is, is my follow-up question, you took my follow-up question with your answer because I was going to say, now, wait a second. isn't AI advanced enough to be able to take less than perfect data and take a probabilistic approach by which you don't have to clean up the data like you used to? And I think you kind of answered that, but I guess I'm going to switch it then because I do agree with you. Now, it all depends on a lot of different factors. You've got to have enough data so that, you know, you'll get the same answer as you mentioned. There's a lot of different things. Like if you're just looking at a very, very small data set and it's incorrect, you might have trouble, et cetera. I'm just making an example. But where does the line of distinction for you reside then? I mean, it sounds like you're very experienced in data. I'm a data guy as well. When do you say, hey, we're going to have to clean this data?

What tools do you use to look at it? I mean, how do you guide a customer in that? I mean, the last thing you want to do is say, hey, you're going to have to spend a crap load of money to clean all your data, and then we can do something with it. I mean, that usually doesn't sit well. I think customers actually expect it, but then they're not going to engage in that from a monetization perspective because it's going to cost them a fortune. What's your guidance? Well, and I 100% agree with traditional solutions that would prove to be disastrous to try and go and build a data warehouse or a data lake on a shaky foundation like that. But what we found with the AI ability to reason past those kinds of things but also remember those anomalies as well And so those also get embedded into the knowledge graph So the knowledge graph is not just you know how we calculate revenue and profit but it also keeps all those little secret sins like, you know, the records that begin with the letter Z in there that aren't real customers. They were test customers, right? And everybody has those. And the normal ETL would just read those in and they'll be calculated in your answers here. Whereas the AI can say, if any record starts with a lowercase Z and an underscore, that's

not a real customer. And that's exactly what the AI can do. I'll give you another example from a story, with a story. We had a customer who's in the safety equipment installation business. And they have employees, they have techs that go out and do installations of safety products to businesses like smoke alarms and things like that. And they had a certain brand of product that they had some employees who were certified to install it. And I believe they had seven of them. And every year they'd go through and say, which of our employees are ready for apprenticeships to get certified. They had seven of them in the data, according to their operational system where these employees, you know, the system of record for all those employees said that there were seven. This customer is using Signal, and their supervisor of this team was talking with and he said, tell me the names of the people who can go in to do this certification. And the system said, you have eight. And he said, no, I don't. I have seven. And the Signal said, I'm sorry, I'm sorry, but I'm reading that you have eight. And he says, why do I have eight? And here's why. because Signals was reading the notes fields in their transactional database of all their jobs,

and Signals determined that there was this one younger employee that the other more senior guys had been apprenticing and teaching how to become ready for certification. And while it was not officially in their data anywhere, the system was reading the notes fields, and it was putting together this fact or this projection based upon what was in the notes field. And needless to say, the user called me up and was blown away when he saw this. He's like, we never would have caught this. We never would have seen this. I was not aware that my guys were training this guy. But it turns out he checked and they were. And so he went and updated the system of record because of what Signal told him by reading the notes fields. It just hit me that the name of the, well, if we're calling it a company, well, the name of the company, Signaled, is purposeful, it sounds like. You got lucky, yes. I just talked to another client today that we're talking about signals in their data. You actually took the name. Say more. We lost a couple of vowels in there. It's S-I-G-N-O-G. I saw that, yeah. Yeah, well, that's, you know, the –

You've got to do what you've got to do. The domain name business, yes. But, yeah, that did fit really well. We were really happy how that came together. And we are working – we're riffing on that a little bit inside the product by having the ability – giving customers the ability to set up topics and signals. The topics are the things they care most about, and signals are the things that the AI can kind of triangulate around those topics. So, for example, if you're keen interested in expanding to another city or maybe it's about improving the profit margin of a certain product line, you can store that in Signal and say, I'm Chris, and this is important to me. I want you to remember this, and I want you to think about this 24-7. And when I come back and log in, I want to hear what else you've learned about it. So that's another feature where it can send off signals for the things that you care about the most. And that's really what makes it good at decision support. is this allowing the client just to ask any question it wants to allow about its data? Or do you like have different modules built in, like an assessment of the industry or assessment of what's going on in the industry, paralleled with the data, their own data?

I mean, do you provide anything else? Or is it Mr. Customer, your turnover on your own? Ask any questions you want, your data, I'll step back. Or do you add like, you know, different prompts or anything to help them make decisions on their data, to give them suggestions? Yes, absolutely. That actually describes a dream I've had for a really long time because I learned a long time ago that all the data inside your company is definitely not all the data that you might need or might benefit from. And so what we wanted to do was make it very, very fluid for Signal to be able to work with the data inside your company, obviously in a secure manner, but also give it the ability to kind of go out and perform research and spawn agents to go perform research on different topics that you may care about and bring that data back in. We had a concept in data and in Zeta called data enrichments, and those were third-party data sets that could be brought into a customer's data warehouse that would be applicable to the data that they had. It would be things like procedures codes for health care or pharmaceuticals codes and things like that, or CUS census data coming from the federal government. Those demographics data, those were all examples of enrichments.

But unfortunately, those were all fixed. The only way we could have an enrichment was if we went out and got it and put it in our platform. With Signal, we've opened that up dramatically. We now have agents with Signal that are able to do research independently and search the Internet as well as other data sources and data sets out there. They could be publicly available data sets. They could just be gathering data from other websites that are out there. And so it's opened up a whole new world of data sets that we can bring in and then make those part of the customer's Signal environment, going well beyond what they may just normally think of as data that they would ETL into their analytics environment. Are these pre-canned type of, what do I want to say, modules or something that's in the InSignals itself, or how does that work? Well, the enrichments, which we had in InZeta for a while, those were preloaded data sets that we would go out and get, curate, clean up, and then make available. But in the case of Signal, we have all of that still, but we also have Signal the ability to proactively go out and seek out data on behalf of our customers or the customer that's using it

that is pertinent to whatever it is they're studying inside of their data set. So, for example, we have a number of nonprofits that are using Signal right now, and they have agents in Signal that are able to go and look at publicly available information and demographic data on different neighborhoods and zip codes and localities and be able to use that information to bring it back into Signal so that the customers have to go out and gather this on their own. It's there waiting for them, and then it gets mapped to whatever data they have inside of Signal. Very, very good. Hey, look, I've got several other questions. I just want to get your expertise on. Sure. But before I do, anything with, because some of these are a little bit more general, not necessarily related to Nzeta or Signaled, just expertise that you have in the industry. What else that I didn't ask you that you wish I would have on Signaled or Data that you'd like the listeners to hear? I've got to say, you were pretty thorough. I can't think of anything that you left out there. I'm from this few times. We talked about how, yeah, we talked about how it works. We talked about RAG and where the knowledge graph fits in there. We see the vector rag as kind of the first foundational layer, then graph rag being on top of that,

and then the knowledge graph being kind of a derivative on that, each one being an improvement over the previous layer. And right now we're finding that the knowledge graph concept is, you know, that may evolve in the future again, and it is evolving very quickly. But right now we're finding that the knowledge graph combined with that deterministic layer is yielding the results that customers are looking for, which is to say that it allows them the easiest path of bringing their data into Signaled using the AI, building the knowledge graph, and the knowledge graph is then able to give them that deterministic experience where the dashboard always has the same value consistently. And it also saves tokens, too, because, you know, a lot of people who might be using, say, Cloud or OpenAI, uploading their spreadsheets in there and then asking you questions, that's consuming tokens left and right. What we've done with this is that the AI is used to basically interview and develop, interview the user, evaluate the data, structure the data, and build the knowledge graphs. But then the actual queries part is done deterministically. It doesn't consume many tokens at all. Or we could use like a very lightweight, low-cost model like JEV to do that.

And so that's something else we're looking at. Because I do think that we're going to have – there is going to be some token reckoning coming up here with how much people are spending and people are going to want, just like anything, they're going to be looking for ways to reduce token spend and token usage. That brings up a question that I had that I didn't ask that I remembered. How is it monetized? In other words, what is essentially the purchasing plan that a client could expect? Let me pause there. How is it monetized? Well, I guess, let me back up. Let me ask it this way. How is it monetized, and what does an initial engagement look like? You can start with Signal for as little as $40 a month. We also have a free forever plan, so it's very economical. The pricing is based upon the number of data sources that you have, as well as the size of your knowledge graph and also how many different operations you perform in there. So if you have hundreds of different data sources and a very large knowledge graph and hundreds of users who are asking questions every day that are hitting the AI, then that would probably be on the higher side.

if you have, you know, fairly basic five or six data sources and they're loaded in their refresh daily, for example, and you have, you know, five or six users going in, then that would be toward the lower end of the spectrum. So it's based on the number of kind of a tokens budget that each account is given. You get a certain amount of tokens for the month. And then you can also build by overflow. You can set limits on that as well. And then the other thing that we've been working on, which I haven't seen anywhere else out there, is a value measurement per token type of score. because, you know, I don't know about you, but I use other AI tools all the time and I get my bill, but I have no idea which of my chats cost the most or which ones I spent the most on or which ones I got ROI on. And so that's something else we built into the platform is kind of an ROI tool for the CFO to use to be able to see not only how much did I spend on Signaled, but what did my employees get out of it. So we built a sentiment analysis monitor into the chat so that the chats themselves can be monitored for value delivered. So if the employee, for example, is getting answers to their questions, they're sharing those insights,

they're telling the system it did a good job, those kinds of things, that's all captured and stored and is put against the token's budget that they're spending. And so that was one of the requests from one of our early executive beta testers. They were like, I don't care how much I spend with you, but more importantly, I want to know how much value I got. That's ROI. Exactly. Makes sense. So it's like a real-time ROI dashboard they get. Yeah, I feel the same way. I mean, one way or another, all the customers I work with, you can't bill off activity or information gathering because that's prevalent right now, right? It's almost ubiquitous. It's all outcomes ROI in order to gain traction. Where can individuals reach you and reach in Zeta and Signals? Oh, sure. So for me, in order, you can reach me on LinkedIn. Just Google my name. There is a Chris Raptor who's a popular country music singer. You don't do that, too? That's not me, no. Oh, okay. If you get the country music singer, that's the wrong guy. Try Christopher Rafter and look on LinkedIn. Signal is at INZATA.com, and Signal is S-I-G-N-L-D.A-I.

And then we've got trials there. You can sign up. You can also, if you do do a trial and maybe you're not comfortable loading your data into it, that's fine. We have a data generator tool built into Signal. So when you sign up, you can tell it, hey, I run a marketing agency or I run a veterinarian office or something like that, And it will generate for you a small kind of playpen environment with data that would be applicable to your, you know, similar to the data that you might encounter in business, in your business. So that way you can kind of try it out in context without having to upload it or connect any of your data sources on day one. And it's fully functioning. You get to use, you get to see all the features. Well, Signal's been out for like, what, a couple months? Yeah, a few months. A few months. I mean, I'm sure you're going to tell me it's going great. But how has it been going in any trials and tribulations you can talk to that others can learn from? Oh, for sure. We've had people who are using it very intensely. We're really excited. People who are using it and staying up until 11 o'clock at night and exploring their business and getting answers in ways that they never had before. That was really exciting to hear. We also had some customers who were running out of context windows,

and we had some of those growing pains. Thankfully, we've taken care of that. Right now, we've got clients who are coming in through various channels. We've seen a lot of adoption in the finance professional space, CPAs, accounting firms. have been using it quite a bit. But we've also seen usage from nonprofits, from retailers, manufacturers, real estate developers as well. So we're still in the kind of figuring out which industries are going to get the most out of it. So we're watching that very, very closely and tracking each of them. Right now, every – What's your sense? And what's your sense on the industry that's going to get the most of it? So far, CPAs have been really benefiting a lot from this. We're hearing CPAs that are talking about saving upwards of 10, 15 manual hours per employee. per month, being able to eliminate software packages that are costing them tens of thousands of dollars a year and being able to just move the workloads into Signaled. We have another new flavor of the product called that coming out called Signaled for CPAs which is basically an agency model So let say you run a safe CPA agency You do taxes and books for hundreds of business customers in your locality

Signaled for CPAs would allow you to set up an account with Signaled and then also make the same Signaled experience available to your customers as well. So they could upload their financials and drop their receipts and PDFs and credit card statements and bank statements and things into Signaled. It flows into you. You could then use Signaled to do their monthly books and also give them insights and strategic advice about their business. They also get a signaled access as well to be able to go in and use it the same way to explore their business and ask questions. So we're really excited about that. These are things I'm interested in. I will finish with a quick rapid fire. So I guess I'll just ask one or two more and then hit in the rapid fire. But look, there's an enormous amount of money being poured into AI. And the way I look at it is NVIDIA is like selling picks and shovels, Microsoft, Google, building some platforms, AI, OpenAI and Anthropic are racing to build smarter models. Who do you think is going to make the most money out of this revolution? Where do you think the money is going to land? I know that's completely different from what we've been talking about, but I'm just curious. I ask these people the same questions, you know, other people that come on the podcast, but go ahead.

I'm going to go back. I used this analogy with some of my employees the other day because I was around, you know, do you remember the days in the early dot-com era where there was like hundreds of search engines, like hundreds? Yeah. Or Lycos. Yeah, yeah, yeah. That's a lot. Right, hundreds. How many are there now? Three, right? So, you know, Google Bing, maybe Yahoo. There's really one that could do one. Yes, I got you. So I think the same lesson, so I'm going to use that lesson to kind of inform my answer to your question. I think it's going to be the company or companies that figure out how to deliver the most value to users. I mean, you remember the story of Google was that they kind of stumbled across a special way of doing searches that was based on how many sites linked to that other site, and that was unique. And I think it's going to be the companies that figure out how to do this in a way that delivers the most value for the lowest outlay. So I don't know who that is, so I'm not asking you. Well, I'm not asking you to be, of course, you know, you can give me a stock pick if you want. No, I'm with you 100% on that. I mean, look, I think, like, you could use so many different examples. You used a great example. You could use fiber optics or whatever. When they got the fiber optics, there were three companies, I think, that were investing like crazy.

They overinvested, kind of like what we're doing with data centers. And then those don't exist anymore. And it's the person that comes along afterwards that gets at a fire sale. You know, at IBM, that's one reason I stayed at IBM. We're building the platform. We're not investing in data centers. I mean, I think the data center craze is just out of control. And NVIDIA is, like, pushing a lot of that. So it's going to be interesting to see how this all plays out. That's why I wanted to get your view. Here's a provocative one, and then I'll go to the rapid fire. It's related to your business, and, you know, hopefully you'll slam this one. And it's, look, if I open like ChatGPT, you know, upload a spreadsheet, ask it to explain my business, why do I need an analytics platform now? I mean, does AI, like you said earlier, AI eliminates the need for perfectly clean data. Can AI not or eventually eliminate the need for dashboard, BI tools, and AI data analysts? Well, I don't know that AI alone will do that. I do think that decision intelligence tools that use AI will definitely put a big dent in that and eliminate that or largely eliminate that.

Where we're betting on now is the concept because coding and building websites and building apps has gotten cheaper than it ever was before in history. It's kind of like when, you know, we used to print out a piece of paper and put it in an envelope and stick a stamp on it to mail a letter, right? 20 years ago, we had email come out, and that just went away, and sending a message became free, basically. The same thing is going to be happening with creating a website or creating a dashboard. It's all code that can be produced by AI in milliseconds. So the price of that's going down, which means that the quantity is going to go way up and the quality is going to drop. So I think we've got to go through that sort of zone first to see that. And back to your original question about dropping a spreadsheet into OpenAI, a couple of things to note about that. Number one, when you open that new ChatGPT window, it has hardly any context, unless you're one of those people that goes and edits your MD files and has all the personal contacts, which you probably are. But most people aren't doing that. So when you drop that file in there, it doesn't know anything about your business. It only knows what's in that file and what you told it. So that's one of the things that, you know, there's a lot of talk about context being the most important mode or the most important thing here. We agree with that, and that's one of the reasons why we're trying to help.

What our knowledge graph does is it allows you to very quickly and easily extract the context that your business is going to need to get the most out of AI, put it in a place where it's safe and secure and can be maintained and is also auditable, and then be able to use it with any of those AI models that you want. So ChatGPT plus a spreadsheet, that's going to get you to one destination. ChatGPT plus a signal knowledge graph that's connected via NCP where ChatGPT can go and check its work with your knowledge graph that you developed and signaled. We think that's going to get people a lot, lot further. Makes sense. Makes sense. All right. No, I appreciate it. Nice answers. That was like you were ready for it. But I didn't see these questions, folks, I promise. All right. So here is, I'm going to ask you 10. This is rapid fire. Okay. And, you know, there's no right or wrong answers because sometimes I'll ask you, you know, who's going to be the AI winner or something like this. We know this can change. We're not asking you to be a fortune teller or anything else like that. But I do want to get your view. You're an expert in the industry, so let's leverage it. All right, so I'll go with just quick answers. And I got 10. I should have 10. We'll see if I have 10.

But anyway, the AI winner is going to be OpenAI, Anthropic, IBM, Google, Microsoft. Who has the biggest long-term advantage in your mind? It changes weekly. I don't know. I don't know. So I do think that there's going to be a kind of a backlash on token consumption, but I also don't know what these companies' plans are or what their future economics are going to look like. So I don't know. I'm going to go back to our second answer. All right. The biggest threat is AI most likely to disrupt Microsoft, Salesforce, Google, or traditional software startups. Who's it going to disrupt most? It's definitely disrupting software startups, companies that I think it's going to be – I think there's going to be a lot of very – the companies at the very, very top are going to be relatively safe, and the ones at the bottom that are developing the niche apps are going to be relatively safe, you know, the lawn mowing business CRM apps. But the ones in the middle are the ones I think that are going to be impacted the most, you know, the companies that are, you know, in the $10 million to $10 billion range. Those are the ones I think that are going to be impacted by this. Here's a slam dunk, the data problem. What's worse for most companies, bad data, too much data, or data that nobody can access?

Probably the too much data one. It's too much data for them to be able to process, yeah. But I'll say the biggest thing that is – the biggest risk is not thinking broadly enough about what AI can do for them. That's probably the biggest risk or mistake I see people making now. They kind of think about AI. They say, I'm going to implement this, but they get to a point and then they stop. And I always ask them, well, why'd you stop there? Like, why can't you push it further? And then they'll say, well, I didn't think so. And then we'll say, well, here's what you can do. And they'll say, yeah, you're right. So we're constantly encouraging people to expand their thinking, expand their expectations. Just like that woman that did to me two years ago when she said, I don't want to make dashboards using a mouse. I got to thank her. She did well. She should get some residuals. I'm sure you're not going to do that. No, skill. All right, let's talk about skill. If you were talking your career today, what would you start with it? What would you learn? Data science, software engineering, business strategy, AI? I think business strategy and also customer-facing skills are extremely important. I love the concept of the forward-deployed engineer.

We've been developing that role at our company for quite a few years. We also just became Anthropic certified. We got all our FTEs, Anthropic certified, CLAW certified. So that type of a skill to be able to work with someone, quickly understand, have enough business strategy training to quickly understand their business, but then be able to go in and help them with their problems, I think is never going to go away. It's always going to be a needed skill. So that's where I would start today. What technology or industry, buzzword or otherwise, you think is getting way too much attention right now? That's a good one. I'm going to have to think about that one. There's so many out there. I mean, anything that come to mind that you say, I'm just so sick of this. Well, I'll say, well, I can remember a time when data centers were at least innocuous. I think there's been a lot of backlash against data centers, ironically, on social media, which wouldn't exist without data centers. But I think that there's a lot of understanding still that could be distributed out there for people around how these actually work. So it's a misunderstanding of data centers is your case.

Yeah, I think people, I think it's an over, I think that the risk of every small town getting a data center in its backyard is probably not realistic. I mean, I'm working in data centers for years, and, you know, most of them are where Fiber is. And so I say to people, well, is there Fiber on your street? No. Okay, well, guess what? Probably not going to be a data center there. So it makes sense. Underrated technology. What's not getting enough attention? I would say GitHub and some of the amazing development that's going on there, the number of developers in GitHub has just exploded, and there's some great, great tools on there. So that's a place where I didn't used to spend two years ago, three years ago. We used it, but I wouldn't say it was like, you know, today it's like you're going on GitHub. It's like going on Facebook, it used to be. You know, there's something new there every day. There's some great stuff to see. So that's been interesting, the way that people are using it. You know, I just saw somebody posted a way to update, somebody live coded a way to update the firmware on your smart TV so you don't have to look at your heads every time you turn it on. I mean, I like that. Who wouldn't, right? Yeah, I'm going to have to investigate. This is why I do these questions. All right, I put that note down.

What's the first white-collar job or business function you think is AI is going to fundamentally change? I would say if it's people who aren't communicating and who are really just performing the same process over and over and over without looking for continuous improvement or trying to make it better or deliver more value. One of my favorite sayings is that it's not that AI is going to replace your job. You're more of a threat is someone who knows AI is going to replace you or knows AI knows how to use it well. So that's what we're seeing. We're seeing AI as being an extremely valuable partner for nearly every role in our company. And that's what we're setting people up with is the knowledge and skills and the agents and the token plans to be able to make the most of it. I got to say, I've been extremely impressed with just my own team and how they've developed and just begun using it. We've automated stuff that hasn't resulted in any job cuts, but it's resulted in major elimination of manual tasks, moving things like proposal production and product management into AI tools. We have been able to eliminate a couple systems as well that we were spending money on that we just don't need any longer. So that's all providing positive results without impacting the single job.

All right. If I give you a million dollars, only a couple more, then we'll call it quits. You're doing great, by the way. Very good. If I give you a million dollars, I force you to invest it in one technology over the next five years. What are you going to invest in here? Well, I really like SpaceX's concept about putting data centers in space, but I think we may be a ways off. But I really think they have, you know, between the – I'm not going to say anything about Tesla, but on Starlink, I think that's a great technology, and it does have the promise to change the world positively. And I think, you know, being able to – if you look at that big vision for SpaceX around putting data centers in space is – But that's probably the one thing I've heard that I think could provide major benefits for humankind. So it's also probably the least likely. Ten years from now, I know that's like light years today. Yep. Forever today. It's not even fair to ask this. But ten years from now, which is going to be commonplace in business that most people find hard to believe right now? I think we're going to continue to see the removal of friction from lots of, especially friction that we're used to right now, that we just put up with.

just like there are things we used to do 20 years ago that we don't do anymore and they just quietly, calmly fell by the wayside, I think we're going to see a lot of that type of friction removed as well. And you've got the coding going to almost zero. I think that decision-making is also going to become much, much cheaper in the future as well. It's not going to require as much to get higher quality decisions, so that's going to benefit many fields and many industries. And so I think just the continuous pursuit of the removal of friction in every process, every transaction, we're going to continue to see that accelerate. Makes sense. All right, man. I think we're just short of 10, but I'm going to stop there. I thought you answered those great, man. It's like I did prep you, and I promise listeners, no prep whatsoever. First time I've met Chris. Chris, I'll give you the last word. Anything you want to say to the listener before we sign off? No, I just want to say thanks. Thanks to you and your listeners. I really, really enjoyed this, and I appreciate the opportunity. You're a good man. I wish you the best. It sounds like you've got a great strategy around both in Zeta as well as Signal. So you guys know where to reach them? Signal.ai. Signal.ai. There you go. He's pushing it as he

should be. Anyway, wish you the best, my friend. You've had great answers. I can tell you're in the code, which is best what you'd want a founder to be. So thank you for being here. Listeners, thank you for listening. Hit us on almartintalksdata at gmail.com. Love to hear from you. I'll see you on the podcast. Bye-bye.