 From Las Vegas, it's theCUBE. Covering InterConnect 2017, brought to you by IBM. Okay, welcome back everyone. We're here live in Las Vegas at the Mandalay Bay, theCUBE's exclusive three-day coverage of IBM, InterConnect 2017. I'm John Furrier, my co-host, Dave Vellante. Our next guest is Harley Davis, who's the VP of Decision Management at IBM. Welcome to theCUBE. Thank you very much. Thanks for the time today. You've got a hot topic. You've got a hot area. Making decisions in real time with data, being cognitive, enterprise strong, and data first is really, really hard. Oh yeah. So, welcome to theCUBE. What's your thoughts? Because we are talking before we came on about how data we all love, we all data geeks. But the value of the data is all contextual. Absolutely. Give us your color on the data landscape and really the important areas that we should shine the light on, that customers are actively working to extract those insights. So, you know, traditionally, decisions have really been transactional all about taking decisions on systems of record, but what's happening now is, you know, we have the availability of all this data, streaming it in real time, coming from systems of record, data about the past, data about the present, and then data about the future as well. So when you take into account predictive analytics models, machine learning, what you get is kind of data from the future, if I can put it that way. And what's interesting is how you put it all together, look for situations of risk, opportunity. Is there a fraud that's happening now? Is there going to be, you know, lack of resources at a hospital when a patient checks in? And how do we put all that context together, look into the future, and apply kind of business policies to know what to do about it in real time? And that's really the differentiating use cases that people are excited about now. And like you say, it's a real challenge to put that together, but it's happening. It's happening. I think that's the key thing. And there's a couple of mega trends going on right now that's really propelling this. One is machine learning. Two is the big data ecosystem, as we call it, the big data ecosystem, has always been, okay, head dupe was a first wave, then you saw Spark, and then you're seeing that evolving now to a whole nother level, moving data at rest and data in motion is a big conversation, how to do that together. Not just I'm a batch only or a real time only, the integration of those two. Then you combine that with the power of cloud and how fast cloud computing with compute powers accelerating. Those two forces with machine learning and IOT is just amazing. It's all coming together. And what's interesting is how you bridge, how you bridge the gap, how you bring it all together, how you create a single system that manages in real time all this information coming in, how you store it, how you look at history of events, systems of record, and then apply situation detection to it to generate events in real time. So one of the things that we've been working on in the decision management lab is a system called decision server insights, which is a big real time platform. You send a stream of events in, it gets information for systems of records. You insert analytics, predictive analytics, machine learning models into it, and then you write a series of like situation detection rules that look at all that information and can say right now, this is what's happening. I link it in with what's likely to happen in the future. For example, I can say, my predictive analytics model says, based on this data executed right now, this customer, this transaction is likely to, 90% likely to be a fraud. And then I can take all the customer information, I can apply my rule, and I can apply my business policy to say, well, what do I do about that? Do I let it go through anyway? Because it's okay. Do I reject it? Do I send it to a human analyst? We got to put all that together. So that use case that you just described is something, that's happening today, that's the state of the art today. Now, so one of the challenges today, and we all know fraud detection has got much, much better in the last several years, used to take, if you ever found it, it would take six months, right? And it's too late. But a lot of still, a lot of false positives that will negate a transaction is now that's a business rule decision, right? But are we at the point where even that's going to get better and better and better? Well, absolutely. I mean, the whole, there have been two main ways to do fraud detection in the past. The first one is kind of long scale, predictive analytics that you train every few months and requires lots and lots of history of data, but you don't get new use cases that come up in real time. Like you don't have the Ukrainian hacker who decides, if I do a payment from this one website, then I can grab a bunch of money right now. And then you have the other alternative, which is having a bunch of human analysts who look for cases like that guy and put it in as business rules. And what's interesting is to combine the two to retrain the models in real time and still apply the knowledge that the human analysts can get in real time. And that's happening every day in lots of companies now. And that idea of combining transactional data and analytics has become popularized over the last couple of years. One obvious use case there is ad tech, right? Making offers to people, marketing. What's the state of that use case? Well, let's look at it from the positive perspective. What we're able to do now is take information about consumers from multiple sources. You can look at the interaction that you've had with them. Let's say you're a financial services company. You get all sorts of information about a company, about a customer, sorry, from the CRM system, from the series of interactions you've had with them, from what they've looked at on your website. But you can also get additional information about them. If you know them by their Twitter handle or other social media feeds, you can take information from the Twitter, the Twitter feeds, for example, do apply some cognitive technology to extract information from that, do sentiment analysis, do natural language processing, to get some sense of meaning about the tweets. And then you can combine that in real time in a system like the one I talked about, to say, ah, this is the moment right here where this guy's interested in a new car. He's been, he just got a, we think he just got a promotion or a raise because he's now putting more money into the bank. And we see tweets saying, oh, I love that new Porsche 911. Can't wait to go look at it in the showroom. If we can put those things together in real time, why not send him a proactive offer for a loan in a new car or put him in touch with a dealer? No, and sometimes as a consumer, I want that. You know, when I'm looking for, say, scarce tickets to a show or a playoff game or something, and I want the best offer, and I'm going to five or six different websites, somebody remembered where to make me an offer. Hey, here are better seats for a lower price. I would be thrilled. So geographic information's interesting too from that. So let's say, for example, that you're traveling to Napa Valley and let's say that we can detect that you just, you know, took out some money from a bank from your ATM in Napa. Now we know you're in Napa. Now we know that you're a good customer of the bank and we have a deal with a tour operator, a wine tour operator. So let's spontaneously propose a wine tour to you, give you a discount on that to keep you a good customer. Yeah, so relevant offers like that, as a consumer, I'd be very interested in all too often, at least lately, I feel like we're in the first and second innings of that type of, you know, system, where many of the offers that you get are just, wow, okay, for three weeks after I buy the dishwasher, I'm getting dishwasher ads, but it's getting better. You can sort of see it and feel it. You can see it getting a little better. I think this is where the combination with, the combination of all these technologies with machine learning and predictive analytics really comes to the fore and where, you know, the new tools that we have available to data scientists, things like, you know, the data science experience that IBM offers and other tools can help you produce a lot more segmented and targeted analytics models that could be combined with all the other information so that when you see that ad, you say, oh, the bank really understands me. Arley, one of the things that people are working on right now, and most customers, your customers and potential customers that we talk to, is I got the insights coming and I'm working on that. We're going peddling as fast as we can, but I need actionable insight. This is the decision-making thing. So decisions are now what people want to do. So that's what you do. So there's some stats out there that decision-making can be less than 30 minutes based on good data, the life of the data, a short of six seconds. This speaks to the data in motion, humans' side of it. I might be on my mobile phone, I might be looking at some industrial equipment, whatever, I could be a decision-maker in the data center. This is a core problem. What are you guys doing in this area? Because this is really a core problem. Well, this is all about leveraging event-driven architectures, Kafka, Spark, and all the tools that work with it so that we can grab the data in real time as it comes in. We can associate it with the rest of the context that's relevant for making a decision. So basically when we talk about actionable insights, what are we talking about? We're talking about taking data in real time, structured, unstructured data, having a framework for managing it, Kafka, Spark, something like decision-server insights in ODM, whatever, applying cognitive technology to turn some of the unstructured data into structured data, applying machine learning, predictive analytics, tools like SPSS to create a kind of prediction of what happens in the future. And then applying business rules, something like operational decision management, ODM, in order to apply business policies to the insights we've garnered from the rest of the cycle so that we can do something about it. That's decision management. That's how it's an advanced cycle. So you were saying earlier on the use case about I get some event data, I bring it in to systems of record, I apply some rules to it. I mean, that doesn't sound very hard. I mean, it's almost as if that's happening now. It's hard. You know, it's hard. I want to get, this is my whole point. This is not possible years ago. So that's one point. I want to get some color from you on that because this is un-gettable. Most of the systems didn't even go back 10, five years ago with siloed. So now rule-based stuff seems trivial tactically. Okay, I'm applying some rules. But it's now possible to put this package together. Now I know it's hard, but conceptually those are three concepts that someone would say, oh, why weren't we doing this before? It's been possible for a long time. And we have, you know, we have plenty of customers who combine, you know, who do something as simple as, you know, when you would get approved for a loan. That's based on a score, which is essentially a predictive analytics model combined with business rules that say, approve, not approve, ask for more documentation. So that's been done for years. I think, so it's been possible. What's, you know, even more enabled now is doing it in real time, taking into account a much greater degree of information. Or data sources. Data sources, things like social media, things like, you know, sensors from IOT, you know, connected car applications, all sorts of things like that. And so more than done before. And then retraining the models more frequently. So getting better information about the future faster and faster. Give an example of some use cases that you're working with customers on, because I think that's fascinating. And I think I would agree with you that it's been possible before, but the concepts are known. But now it's accelerated to a whole new level. Talk about some of the use cases and to end that you guys have done with customers. Let's think about something like an airline that wants to manage its operations and wants to help its passengers manage operational disruptions or changes. So, you know, what we want to do now is take a series of events coming from all sorts of sources. And that can be, you know, basic operational data like, you know, the airplanes, what's the airplane? Is it running late? Is it not running late? Is the connection running late? Combining it with things about the weather. So information that we get about, you know, upcoming weather events from weather analytics models. And then turning that into, you know, predicting what's going to happen to this passenger for his journey in the future so that we can proactively notify him that, you know, he should either, we can either rebook him automatically on a flight, we can provide him, if we know he's going to be delayed, we can automatically provide him amenities, notify the staff at the airport where he's going to be blocked because he's our platinum customer. We want to give him lounge access. We want to give him his favorite drink. So combine all this information together and that's what's going to happen. That's life. That's life. That is happening. I don't want to fly that airline. Okay. So we've been talking a lot about it. American Airlines? Okay, we're going to push it in the spot there. Hold that. It'll get you in trouble. Well, I guess we had to experience the other day where it's a real land use case. If you ever say, oh, hey, you're not going to make your connection. Like, yeah, thanks for letting me know. Okay. Okay, we were talking a lot about the sort of way things used to be, the way things are and the way things are going to be are actually are today in that last example. And you're talking about event-driven workloads. One of the things we've been talking about it, SiliconANGLE and Wikibon on theCUBE is workloads. We had Batch, Interactive, Hadoop brought back Batch. And now we have what you call this event-driven workloads, we call it kind of continuous workloads. It's all about data and motion. There's, you know, we all call it different things, but it's the same thing. And when we look at our forecasters, we're like, wow, this is really going to hit, it hasn't yet, but it's going to hit the steep part of the S curve. What do you guys expect in terms of adoption of those types of workloads? Is it going to be niche? Is it going to be predominant? I think it should be predominant. And I think companies wanted to be predominant. What we still need, I think, is a further iteration on the technology and the ability to bring all these different things together. We have the technologies for the different components. We have machine learning technology, predictive analytics technology, business rules technology, event-driven architecture technology, but putting it all together in a single framework. Right now, it's still a real, it's both a technology implementation challenge and it's an organizational challenge because you have to have data scientists, work with IT architects, work with operational people, work with business policy people, and just organizationally bringing everything together. That's an organizational gap. Yeah, that's what you're talking about. Yeah. But every company wants it to happen because they all see a competitive advantage in doing it this way. And what's some of the things that are barriers being removed as you see them because that is a consistent thing we're hearing. The products are getting better but the organizational culture. The easy thing is the technology barriers. Yeah. That's the thing, you know. That's kind of the easy thing to work on. How do we have single frameworks that bring together everything that lets you develop both a machine learning model, business rules model and optimization, resource optimization model in a single platform and manage it all together? That's, we're working on that and that's going to do it. All right, so I'll throw a wrinkle into the conversation. Hopefully a spark, pun intended. Open source and microservices and cloud native apps are coming that are with open source. It's actually coming in and fueling a lot more activity. This should be a helpful thing to your point about more data sources. How do you guys talk about that because that's something you have to be part of enabling the inbound migration of new stuff. Is that? Yeah, we have, I mean, it's everything's part of the environment. I mean, it's been, I think it's been the case for a while that open source has been kind of the driver of a lot of innovation and we assimilate that. We can either assimilate it directly, help our customers use it via services, package it up and rebrand open source technologies as services that we manage and we control and integrate it on behalf of customers. All right, Harley, last question for you. Future prediction. First five years out. What's going to happen in your mind's eye? I mean, we're not going to hold you, I mean, IBM to this view personally. You know, just as you can see some of the stuff unfold with this machine learning, we're expecting that to crank things up pretty quickly. We've seen cognitive and cognitive to the core really rocking and rolling here. So what's your, how do you see the next five years playing out for decision making? First thing is I don't see Skynet ever happening. I think we're, we're so- Mark Binyiv had a nice reference in the keynote about Terminator. I'm like, no one picked up on that on Twitter. Yeah. So I think that's nearly impossible as a scenario. But of course what is going to happen and what we're seeing accelerating on a daily basis is applying machine learning, cognitive technology to more and more aspects of our daily life. But I see it, it's in a passive way. So when you do image recognition, that's passive. You have to tell the computer, tell me what's in this image. But you, the human or the developer or the programmer still has to kick that off and has to say, okay, now that you've told me there's a cat in an image, what do I do about that? And that's something a human still has to do. And that's, you know, that's, that's the thing that would be scary if our, if our system started saying, oh, I, you know, we're going to do something on behalf of you because we understand humans completely and what they need. So we're going to do it on your behalf. But that's, that's not going to happen. So the role of the human is critical paramount in all of this. And it's not going to go away. I mean, we decide what our business policies are in. But isn't, well, autonomous vehicles is an example of that, but it's not a business policy. It's, it's the car making a decision for us because we can't react fast enough, presumably. But the car is not going to tell you where you want to go. Oh, absolutely. You're not going to, I mean, if it started, you know, if you got in the car and it said, I'm taking you to the doctor because you have a fever, maybe that will happen. Yeah, maybe. That's kind of Skynet like, I'll be worried about that. It might make a recommendation. Yeah. A notification. Hey, you ought to go to the doctor. Thank you. I'm good. I really don't see Skynet happening, but I do think we're going to get, you know, more and more intelligent observations from our systems and that's really cool. It's very cool. Carly, thanks so much for coming on theCUBE and sharing the insights. Really appreciate it. theCUBE getting the insights here at IBM Interconnect 2017. I'm John Furrier with Dave Vellante. Stay with us. Some more great interviews on day three here in Las Vegas, more after this short break.