 Morgan, we're here with Tim Estes, who's a CEO, and Rob Metcalf, who's the president and COO of Digital Reasoning. Welcome, gentlemen. I thought maybe Bob Metcalf was coming on, but then my friend Bob Metcalf that I used to work with. That's how you help with that mic, actually. Hey, so congratulations. Saw you guys up on the big screen on the keynote. So that's exciting news. So take us through. Michael Olson, you know, gave him some good love to Digital Reasoning. So, you know, it was a kind of complicated slide. It's a complex solution. So, you know, one, tell us why you guys are on there, and what's the story? Yeah, well, thanks for having us on. And I think that, you know, Mike has been, I consider him both a friend and a partner. So we've had a great relationship with Plowdyr that goes back into, I think it was probably Summer of 09. So rather early in that process, and we started migrating on to Hadoop in 09, and for an unstructured data play, that was pretty early. So it's a little tricky from the standpoint of technology that moves beyond counting things that are well structured, to being able to figure what you're measuring, like what is in unstructured data, who are the entities, who are the people, or the places. It sort of presumes pattern recognition. It presumes other analytics to even make the building blocks that can't be counted. So that's been a big gap. So that slide was talking about how do you take loose, noisy information, and our pedigrees in the defense space and the intelligence space. How do you take loose, noisy information that's disconnected, unstructured, and then connected together so that you can then apply analytics to it in a business valuable way? And so Mike, I think, wanted to show what was possible. He knows about some of our implementations in other places. Well, let's jump into that in a second. But back up and tell the folks about digital reasoning, the company, what do you guys are all about, and specifically educate them on, and then we'll go into the Hadoop side. Yeah. So I think that to understand digital reasoning is to understand that we wanted to create a way to take human communication and use algorithms to make sense of it without having to have a human design on ontology or design some other structure a priority. And so basically digital reasoning is a 30-ish person solver company growing quickly right now, principally in the defense intelligence area and moving into the markets, financial services, enterprise risk. And I think that kind of the transition from what we've been doing technically to sort of where the business value, I want to introduce Rob and kind of have him jump into this since he came off of Lexus. And we really wanted to bring in that kind of knowledge about how to apply information to business value in the DNA of our company. So the ontology thing has been a hard nut to crack for years. It's always been kind of an academic thing. But now with machines, if you can build a good algorithm or seed, you can actually scale. And the reasoning has been one of those things around with now this metadata available, just give us to kind of walk us through kind of like what's under the covers with digital reasoning and some of the tech involved. Pretty on that. Oh, yeah. So, yeah, I mean, it says the founder and inventor at the institution is going to get thrown right back at you. And so I was a philosophy major by background. And so what? Oh, that explains it. Exactly. Exactly. So the new data science mix of disciplines. Like an intelligence analysis, a lot of people are in the liberal arts area that also do good math, if you will. And so those two skill sets allow for creativity and then computation to come together. So for us, what's underneath the covers, if you will, the fundamental difference is we look at a way a word is used in context, a symbol. So like toothbrush. And you can figure out what toothbrush means without having an ontology of toothbrush, being under hygiene or some other. Because where do you really put toothbrush, right? Is it a stick? Is it basically something that's under human hygiene? So where do you fit it? Well, it's known by how it's used. So it tends to be associated with toothpaste and teeth and generally mornings and hopefully evenings if you're very hygienic. So what the system does is it looks at patterns of the way a word is used in context to ground a word in terms of its surrounding material. And that's how a human baby does it too, right? I mean, they're exposed to, here's the taxonomy that came down from Aristotle. And now we know what this means. They're exposed to invariant patterns. And what we do is we build that from the ground up, which allows us to handle data that's traditionally defy knowledge engineering, and that those algorithms have been now ported to work on Hadoop and work at scale and on some of the hardest data sets in the world that are inside certain agencies. So is it a sort of modern form of what I think of as classification, which is a brute force miserable exercise. It's sort of an automated approach toward drawing inferences for this large corpus of data. Is that fair description? I think the way I would classify it is, it's a bad pun if it was one, is as a clustering algorithm. So you look at the similarity of things and you build hierarchical similarity from the bottom up so that you can make things relate to other things with no a priori set of categories, right? Yeah, okay, great. And so we do have the capability to do entity extraction and sort of pattern recognition type approaches, classification approaches through training. But we see that technology while we would say we're comparable to the best in the world at that, especially one that runs on Hadoop. We think that technology is very useful at the high level, but it gets really hard at the details, right? Because a massive human investment is necessary to apply that at low granularity categories. So we try to do the hard low end stuff, the bottom of the ontology wheel, from the bottom up algorithmically through clustering and do the top down through classification. Yeah, okay, so now Rob, let's bring you into the discussion because I know John, you can go all day on this topic and we probably will, but Rob, take us through the discussions you're having with customers in terms of, all right, how are they applying this? What do they want to do with this cool secret sauce? Yeah, so across a number of different areas you've got customers have a large amount of unstructured data that could be inside the firewall, that could be outside, things they want to understand, but today they essentially have to read or at best they have to put forward a solution which gives them a subset of documents to read, sort of search and recall. What they're ultimately trying to do though and what we're all trying to do is identify actors and actions and facts and patterns and put those things in place and time. And we want to be able to do that in the most efficient, automated way possible. We don't want to have to go read and then load things into fact bases. We want a computer to do it for us. And so when that's in the government scenario that can be defined bad actors. If it's in a legal environment, they're always fundamentally trying to answer the questions around who knew what went. That information may be in emails, that information may be out in discussion boards, it may be out in other areas to be able to fuse those things together. And this is where the sort of the technology and what Tim just walked us through in terms of the ability to take on a lot of different domains and be able to understand those things without having to pre-model really matters. You need something that can link across various data sets and surface the key people or entities, organizations and things and enable you to start to draw connections, conclusions from that data. And that matters we think in, whether you're a bank or financial services business, trying to make sophisticated intelligent trades, whether you're doing electronic discovery and trying to much more efficiently search and understand large amounts of data, we think it matters down the road in health as well, where you've got large amounts of unstructured data and health records that can really only be understood today if someone has to read it. So we touched on about four use cases if I got it right, government going after bad guys, trying to figure out what the chatter says and what kind of patterns and where the bad guys are. NSC must love that. Legal IT, I guess, the e-discovery space, what largely has been email archiving, which is this little slice of enterprise risk, information risk I should say, and then analytics, making money on Wall Street and then eventually saving lives. And today if I understand it, you guys are primarily working with the government and looking to obviously take that to the commercial sector, right? That's right. How's that going? That's going well. I think that we're kind of on the bleeding edge of what's going on in Hadoop, as Mike made in his keynote. The year ahead of us is probably the year of applications. We believe that part of the application area that's been very limitedly exploited, analyzed, understood is the unstructured data. And we think that fundamentally, that's where most of the value is still latent. So the structured data, that we already have patterns for handling that, we just haven't been able to scale it very well. And unstructured data is a different problem because we don't even have really the patterns at hand to understand what we can do with that when we do. What's the go-to-market and growth strategy for you guys? Obviously, this is a hot area. Reasoning has been a concept that is really ready for prime time with Hadoop. It enables so much more capabilities at scale than you guys are doing. There's different use cases in terms of your customers. I mean, they want to do, they embed it, are they licensing it? And then as the market grows, how do you guys grow your business? I mean, the surveillance tech that Hadoop enables is pretty compelling. So things like we were talking yesterday, the transactional side of the business is not there yet with Hadoop. So it's still fast, I mean low latency with HBase, but it's not still not transactional in the nanosecond that Wall Street, for example, needs right now. So how do you guys see that playing for you right now? Are you still in that VI business intelligence-like category? Is transactional workloads something you're doing? What's the story there? Yeah, I don't see it being transactional because essentially what we're trying to do is lower the human effort to understand and read things. Got it. So we talk about automating understanding, that's sort of like a phrase we've seized on in this. It's really about how do you take on a class of data where we no longer can scale the human resources to understand it. Because what's happened is data's gone about 100x in the last maybe 15 years to 16 years, about 10x in the last six rides. So you can back up to the web round 0102, get another 10x there. Human attention has not expanded. So what's ended up happening is we do a search and you get maybe the top 1% to 5%, depending upon the amount of data involved, 10 to 15 years ago. Now we're getting the 0.5%, the 0.1%. Or we're getting worse than that in the larger datasets, 0.01%. And the psychological thing happens when you're dealing with such a small sample that you feel you have no confidence at all anymore. So that already has been hit by the Intel community. And so they've had to make a preemptive investment in that arena, which we're part of. And I think that it's true in extra financial services and people that have large scale data problems. So web 2.0 community that has this data comments those pieces as well as financial intelligence. It's natural they're gonna have to do that next because they can't go hire 100,000 people to read this stuff. So information risk is inherently distributed by its nature. So, and if I understand it, you're not trying to bring all that information into one big data temple. You're going to where the data is. Can you talk a little bit about how you see that working, maybe specifically in a commercial case and even in a legal scenario? How will a customer actually practically exploit your software to solve problems related to who knows what, when, where? Sure. I mean, I think what you'll have is a, essentially as data is either consolidated in certain locations or identified in various ones, you'll have to go and run an algorithmic process to get at the key facts and relationships. And you'll have to store and process that information across the various data stores. So rather than sort of necessarily consolidating everything in one place, what we think about is there's data and then there's the analytical or understanding layer. There's the entities and the relationships, which is a spring. A metadata. You can think of it as very, very... A lot of it. Everything is metadata in the model. So it matters not just that a particular person has talked about, but who they spoke with. That's, think of that as metadata, but that's ultimately a very, very sort of feature rich environment which you can run statistics on and to the extent that you can start to summarize those entities and the things that they're doing, you can run the statistics either locally or, you know... We'd like to do some follow up with you guys. Love the, I mean, we can talk for an hour on this, talk more about the technology and the solutions in place because that's the future what you guys are doing is the future. We're covering it, so we're out of time, but I want to follow up with you guys for sure. So... That'd be great. Just real quick, real quick, we can. So 30 people funded? Not yet. Not yet? Big data fund, hello. Looking for money, or actively, or can you say? Word is not going to comment right now. Okay, that means yes. Translation, everyone wants to fund them, so we don't. Good for you. Auction, one word, auction. I like them a lot already. Learned from Frank Quattrone, baby. Well thank you very much for having us. It's been a real pleasure and we're happy to come back in any time. We think it's going to be a very, very big change in the way computers are used when we can harness some structured data over the next decade, the way we've been able to harness structure for the last 20 years. Yeah, excellent. Any VC watching, you got to fund these guys. It's the future, love it. All right, Tim Estes, Rob, Madcap, thanks very much. Digital Reasoning, watch these guys and we'll be watching, thank you. Thank you very much. Okay, our next...