All right. Welcome back to another episode of The Mostly Unstructured Podcast. bald guys back in the house. Yeah. I have not grown any hair since we got back together. I'm really trying hard all the effort. Listen, we don't have to spend money on hair and makeup. It's just makeup. Haven't had a haircut in thirty years. There you go. And we haven't had makeup either. So as far as, you know, saving a lot of money. Well, let's not go into that. Um, so good to do this again. Always a pleasure. Love getting into this, this stuff and helping people understand it. And, you know, I want to start out by saying it's really interesting. We say this a lot. Like if anybody tells you they know everything or they know what's going on and they've got the definitive answer, they're probably lying. I mean, everything is, you know, it's moving so quickly. We talk about it all the time. I mean, the pace of play right now is is breakneck. And you know, you're you're trying to stay about one minute ahead of everybody else. And we're going to talk some talk about something that's a little bit new today. And in past episodes we talked about orchestration layer and orchestration layer is the layer that, that, um, you know, it controls how it connects to, to your content and your organization. Uh, and, and that's super, super important. But I think we want to talk about this layer that's underneath the orchestration layer. Super interesting. Um, and the title of this episode is what does your business actually mean to AI? Not, not what does AI mean to your business? But when AI looks at your organization, what, what does it see? What does it understand about you? So I think by and large, we've made the point that AI has the ability to reach in and understand your documents and, and your content, unstructured, structured, etc. and there's a lot of solutions out there. Some do it well, some don't do it well. Um, but it's really this question about does AI understand what those documents mean in the context of your business. And that's where we want to start. So Ed, you're the CEO here at Keymark. Thinking like a CEO. Um, why does a CEO or maybe better yet, why should a CEO care about a meaning layer if we don't call it that meaning layer? Why should they care about that? Well, I mean, I think you're talking about sort of the critical crux of, you know, what is going to make AI work in your organization or not in a way, you know, I think that to the point of, you know, what does AI know about my organization? You know, the day I quote, turn it on, the answer is nothing, really, or nothing that hasn't been publicly disseminated. So they might know about you by what's out on the web. It may know about you based on what's on your website, also might know about you, about what somebody wrote about you on Reddit. Um. Mhm. Of which, you know, you could say there's no accuracy to some accuracy, but it certainly isn't the version of the story that you would want AI to know internally. Right. And so, you know, as you talk about the meaning layer, you know, one of the things that becomes critically important to your point about AI's ability to read documents and those kinds of things, that really becomes the first iteration of sort of teaching AI about, about me, right? And I think the important part of that is every company, you know, every CEO thinks their company has some secret sauce that makes them different. I mean, if you don't think that what are you doing? Right. Right. So assuming that you believe that there is something about the way you do things or what you do or how you do them is differentiated, and you want AI to sort of accelerate your differentiation. You sure better expose it to what you think that is. Mhm. So before we go any further on this conversation, and I think it may may help to throttle an analogy, get kind of human here and and set a set a stage for for further discussion. So let's just say that you and I every morning stop by the ubiquitous coffee shop that is the mermaid in the sky. Yep. And we buy a mocha Frappuccino latte, whatever it is. I'm well invested. Yeah. Right. And at the end of the day, if we're really good, we go in and we put that into our accounting software and we we tag it as expense of some sort. If I tag that as, um, entertainment. Yeah. Right. You tag it as groceries. Right. Okay, great. Until we just hand that over to an accountant or a financial advisor And they look at that and they look at yours and go, man, you need to find a cheaper place to buy groceries. You're spending a lot of money every day on groceries, right? And they're like, wow, Clay, you're doing a lot of just fooling around entertaining. Yeah. They don't know it's coffee, right? There's meaning to that. Yes. And the transaction is actually identical in this case, but the meaning is, is different. And I think any advice that that. So if AI looks at that and it makes advice on that or even a human in this case, it ignores the difference in those two. It's going to be confident, it's going to be confidently wrong about where that money is going. So put that take that coffee. What is coffee in an enterprise, for example? So, you know, a hospital, a bank, an insurance company, what are what are some examples to help people connect that analogy with the real world? Well, I think if you think about it and you mentioned insurance. And I think it's just an ideal example to start with that. Every property and casualty insurer, for example, has a claims department. Mhm. So if you were to ask AI, you know, how do claims get processed. It will give you a very confident sure answer. I was taking AI training class recently, and it talked very specifically about the fact that that confidence sure answer will come through regardless of its confidence percentage, let's say. So as you read it, I mean, you would read it as well. It's it's certain and to your point about, you know, entertainment versus groceries. Well, you know, every insurance company is going to tell you that they process claims differently and that the way they do them is part of what differentiates them. If you think about, you know, kind of that same analogy, what makes individual insurance companies different and competitive is, first of all, how do they assess risk, right? So what company A might see as a risk, which is I'm only going to do high end luxury vehicles because I believe that caters to a demographic that has very low risk of accident, etc.. And I start going through and that's my bias, right? And so that is when I go to price your policy, I'm going to price it based on the fact that I think you're very low risk. I'm going to give you a better deal. There are whole companies in insurance built on the high risk. They they, you know, assess that risk. They price for it. They, you know, they write their policies with all those things in mind. So if you said to AI and you stood up in both of those companies and you asked the same question, you know, how do you price for risk or assign me a price and I can even feed it, hey, here's all my existing policies. Does my pricing look correct? AI is going to give you an absolutely certain answer and might be wrong in both cases, because it doesn't know anything about us and how we assess risk and how we price for it. And that's meaning that's exactly what we're talking about. It's exactly what it is. And then the financial implications and the risk implications of putting all your eggs in that basket. When AI says that without it understanding the meaning, well, I sent you guys here, at KeyMark I sent you guys a quote yesterday and I, I, I brought it with me because I'm not smart enough to remember it. Um, and it was a two parter and it said, this is the asymmetry at the center of professional AI use. The time saved is small and visible, and the cost of an unverified error is large and arrives later. The cost of a missed error is not paid. When you save the time, it is paid later by someone whose trust you needed. Well, and that you know, when I read that, it really resonated with me because it's exactly right. So we we ask our question, you know, how do you price for risk and all those things? It comes back and says, this is it. And I'm an analyst. Nobody told me to question the AI answer. And we pass it through and it goes into our model and it becomes the price. And then, you know, whatever happens happens. And now all of a sudden, how do you map back to something that's simpler and it's not. And this is, I think, something you and I are conscious of, particularly with this podcast. It's not that AI is wrong or bad. We it knows what we tell it, right? And if we don't teach it that meaning layer, then it's going to say, well, I know about claims based on all the information I've gathered from the internet for all these years. Yeah, that's what I know. That's a great example. So we're going to get into ways to handle that. But before we do that, I wanted to address the elephant in the room. So often is the buzzwords or the alphabet soup. And there are plenty of them, but we're going to probably use a few of them. And I think it's helpful for context. So I brought a list like you did, because there's so many of them. There's no, there's no chance of remembering. Yeah, yeah. So, uh, Ontology. Yeah, yeah. Taxonomy. Hierarchy. Yep. Knowledge Graph, RAG, Graph RAG Yep. Mesh, MCP we're down with MCP. We've said that in the last. Yeah. I'm getting a t-shirt made A2A so it's and that's just scratching the surface right. There's, there's a lot of them. Um, and so I wanted to break some of these down a little bit and then we can we can talk about them. But so let's start with with hierarchy and that I think most people understand that. But in the context of what we're talking about, it's where everything lives. It's, it's a tree, right? The folders within the folders that's, that's most people get that. But taxonomy is really about what something is. This is an invoice. This is a contract, right? There's definitions, labels that you've assigned to those, whatever it is in your organization, that's your taxonomy. But then what we're really talking about is this new term, or at least relatively new, at least in, in my mind is ontology. And that's that meaning. So think of ontology as meaning what it actually means. And so the ontology does just doesn't say that this is an invoice. It's saying that this is an invoice that came from this vendor, right? So it's making that connection. It references a purchase order and it sends that to a certain person who needs approval, and that person sends it on and it gets paid right. And it understands that entire process and what all that means. So it's this web of relationships with your and rules with your within your content. So, uh, before we talk about knowledge graph, I think we will dive into that for a minute. But I did want to throw this question at you. I think it's interesting is it is the industry drowning buyers in vocabulary on purpose? Like. Yes. Are they. Yes. I mean, um, it's, there's a, there's a scheme, here's, here's the, here's how I would frame it. Illuminati. Yeah, that's exactly right. Yeah. There's just they're all in a basement coming up with terms. You know, I think it was, it was funny because as you were running through that, I almost interrupted you to say that, you know what, what I think is happening is you're we're getting a lot of new terminology, uh, assigned to ideas that have been around for quite a while. Right? If you think about, you know, the concept of taxonomy in the world that, you know, we grew up in with keymark and this, you know, unstructured content world, we talk about our content customers, you know, that the type of document it is to your point, well, how do you want to search for it? Well, I want to search on invoice number, PO number, total charge. You know, I want to do. So one of the things that is so critical to the beginning of any kind of a project like that is figuring out that taxonomy. It is critically important because as we grow and evolve and change and scale, will it hold up? Have we made it so that it will stand the test of time? How does it interlink with other components of my organization? So great I've got invoice number etc. assigned. Well, finance might not be the only people that ever want to see an invoice. It might be someone in sales, it might be somebody in marketing. Is that taxonomy going to make sense to them? You know, how do we how do we connect all those different dots, that hierarchy, you know? How do those pieces fit together? And the ontology, uh, comment, we used to just call them rules. You know, I mean, really, if you get down to it. Right. We used to call them rules. They're the rules that drive process. They're rules that drive, um, you know, workflow in an organization, not the technology, but the actual workflow of an organization. And, you know, I think ontology, and I'm oversimplifying and I get that, but I think that, you know, these terms can be a little bit, I don't know if it's scare people away, but it's certainly, you know, can have that effect a little bit of a, a bit of a flex. But I mean, these are, uh, you know, kind of new terms in many recycled, albeit evolved ideas. Yeah. Well, you actually did an excellent job of breaking down just the definition of a knowledge graph. And you actually, I want to read, I'm going to read it because it's, it's succinct. So a knowledge graph is just the ontology brought to life with your real data. The ontology is the blueprint. The knowledge graph is the actual building or like the structure you just talked about that it's connecting the dots. It's your real customers, you know, use cases, documents all wired together through like the relational connectivity there. And I think that's that you said it really well. I thought I'd say it nerdy and read it off, but you said it well in layman's terms and helping people understand that. And that's really what we want to do here is help people kind of get, get away from stumbling over these, these words because I, yeah, I don't think they're, they're not intentionally being created to confuse people, but they were trying to name things. We're trying to say, this is what this can feel like barriers, you know, to somebody who's trying to better understand how to make use of this technology. And the last thing you want is, you know, a term like ontology to, uh, dissuade somebody from, from digging in because, you know, arguably the best value you're going to get out of an AI investment is you've got to get the ontology, right? Right. You've got to get the knowledge graph right. If you wanted to make good decisions about you, your org, you specifically, that's where that value is going to come from. So we've got more to cover. I think what we're going to do is we're going to call it for this episode. We'll call this part one, part one and part two. Yeah. Wow. Look at us growing up and being all cool and unstructured, the sequel you. Yeah. Oh, I like that. That's good. The sequel yeah, that might be overselling it. That's something marketing would do. You know, oversell it. Yeah. Um, anyways, so thank you for joining us on this part, one of The Mostly Unstructured Podcast. Come back next time. What we're going to do is we're going to really put this into a real world scenario. We're going to talk about two paths, kind of a fork in the road. Uh, don't miss it. Come back and check it out. You're going to, I think you're going to get a lot from this. We'll see you next time. Thanks.