 Hi, I'm Peter Burris and welcome again to another CUBE conversation from our wonderful studios here in Palo Alto, California. Another great topic to talk about, we've got Robert Schmidt, who's the chief IoT technologist at Deloitte. Welcome to theCUBE, Robert. Thanks for having me. You also have your own video cast. So, why don't we get that out of the way? What is it? Yeah, every Friday at 9 a.m. Pacific, I do a show called Coffee Chat with Mr. IoT and Misconnected. I just actually added a co-host. I thought I needed someone to help me. And we talk about IoT, it's on YouTube, you can find it on the channel, and it's really odd for me that you're going to ask me the questions and I'm going to have to answer. So, I'm going to try to eat my own advice here and be short. Well, maybe someday you can add one of the Wikibon folks in your podcast, or video cast, we'll have to do that. All right, let's start here though. Deloitte's a great name, been around for a long time, associated with customer value in very profound ways, complex applications, that certainly characterizes IoT. What's going on with IoT and Deloitte? For us, we started a whole practice around IoT and I'm leading that practice, but the thing for us was, there were a lot of science experiments going on around IoT, technology-based, but we really wanted to bring it to, what's the value behind IoT? So, we really focused on use cases and today we see that the most focus is on industrial IoT, though we spend a lot of time around connected products as well. I personally actually today work on a project in a factory in Chicago, on a shop floor, connecting machines and measuring data and providing value. I work with an airline at an airport around their travel, so really helping guide you throughout the day. Interesting fact, you know, we swipe away a lot of notifications without actually doing anything with it, but when an airline tells you, please come in 10 minutes early, the TSA wait time is long. I know you and I- You pay attention. Yeah, we got to be there early, so we actually react to those notifications, so I work on that and I work with high tech companies around their platforms, how do we make their platforms better, so. You know, you've raised a lot of really, really important issues, but let's start with this notion of use cases because a factory floor with a lot of PLCs, spinning out information, mediated by individuals or users and the data, where does it end up, and that's real different from an airport where a lot of the data's being generated by a human being as they move places or as intended to be consumed by a human being. What kind of common patterns are you seeing in these use cases that brings them all under this notion of IoT? I always think of IoT as taking sensor data and making decisions based on those. And what's interesting to me is that it creates this really interesting dilemma that we thought we knew what goes on with users, how they work and what they do. We do surveys just to find out what they're saying. The survey is actually probably not what they do, but now we sense as we know what they do, all the way to machines where we have decades of people having experience about, this sounds a little odd. The machine doesn't sound right, but then they don't know what to do with it and now we can measure that. Because really at the end of the day, vibration isn't anything else but sound, right? And so for me, this is all about and what's common about this is that we really take that we think we know to we actually know, because we cannot measure with sensor what goes on in that area. So it's almost like taking a lot of that time, motion, analysis, operations, research that we used to do periodically, episodically, with human beings doing their best to record stuff and then bringing a lot of that discipline continuously and in real time so that it can better inform overall decisions, right? Yeah, I mean, almost near real time in many of these cases and that's a really interesting scenario for me, right? Because now you can actually see what happens in the factory when I tune the mix or the blend of my raw materials, what happens to the product that gets made at the end of that. So as we think about the challenges or the changes that we foresee going on, is there a difference in thinking about humans as users or humans as consumers of a lot of this data and machines? I know there is, but how is this, because kind of the machine side has always been associated with SCADA, OT and the disciplines and the approaches for that side seem a little bit different than what's coming out of the mobile world, which is still very, very closely associated with how we utilize or how we deploy these systems to inform decisions in either case. Is that right? I don't really know if we do so much about decisions for machines. I think at the end of the day, many of the decisions are still made by humans. So I mean, I think of this like, we have an overheating element running over. At the end of the day, it's still as a human that goes and sort of like says, yeah, let's turn that off. But there's still automation that takes place. Absolutely, there's automation. But automation takes place today. Sure. None of this is particularly new. I mean, OT has done automation forever, right? I think the interesting part is now, taking the learning and connecting the different data points together. So I talked about the factory floor. I just showed actually at the show, we created a virtual factory line, life size. You can download it. It's the virtual factory by the Lloyd. If I get my phone going, I can show you, but it's not right here. I call it the Internet of Rubber Ducks. And the Internet of Rubber Ducks. The Internet of Rubber Ducks. It's kind of cute. You have these little yellow ducks and if you load the app, you can see them being made. But it's actually really what goes on at the factory and it really shows how when you change the blend at the beginning of a production line, how it affects at the end of the factory line, the outcome, how much scrap you have, so what's the scrap, what's the overall equipment efficiency, OE, and so forth. So what happens is now we can connect data from the very beginning of the factory line with the very end of the factory line and then combine that with contextual data, such for example as temperature or the vibration on the machine or the current, which we haven't done before. So this whole time series of data that we not correlate becomes really critical and I don't think that's something we've done really as much before that has not driven automation. So if we think about it, we're talking about sensors, which as you said, GATE has been around for a long time and it tends to automate very, very proximate to where that sensor tower might be, but a lot of the information that went into decisions was actually then generated by a person, perhaps a shift supervisor or somebody else or a machine operator said, I heard a rattle, but there's no time, so it's difficult to correlate and now we're talking about up-leveling a lot of that information, so it becomes part of the natural flow out of the machine, but still for human consumption to make decisions. Yeah, very much like that. As I said, I talked about the blend of the materials that go in and then now we can correlate that particular part of the sheet. We can look on video and see how it looked and check the quality and then see at the end how many pieces of productivity we produce. Actually, in that particular case, it's really fascinating. It wasn't so much about reducing cost, it was actually increasing output. For them, each line cost about 10 million and with the findings we have and what we're doing with them, we can actually give them the ability not to build another line, but actually produce more lines because they can sell more, which is a great position to be in. So you can actually impact the top line rather than just the bottom line. Well, productivity fundamentally is a function of what work you can perform for what costs are required to perform that work. And if you can improve the effectiveness of something, keep the cost same, but get more work out of it, that's a big, big plus on the bottom line. And they have to market to sell it into, right? Absolutely. Because even if you just make more and you can't sell it. Well, there's that too. Yeah, which is really the great thing about that. But to talk about how, for example, you noted that they can look at a video of how the plastic or the sheets coming off, the machine off, set of rollers perhaps, but how does AI start to be incorporated into this IoT discussion? And what kind of use cases are you seeing becoming appropriate or more appropriate or made more productive by some of these new technologies as we bring some of the analytics and some of the IoT elements together? So we find that we do a variety of theories. We go in and say, how about this? How about that? And then we have our data scientists go and look at models for that and see what goes on and then put machine learning in. And then we take those machine learning models and feed it back into, we talked before a little bit about this, but edge processing is really something where we now process some of those models on the edge. So the algorithm development and all the analysis, we send that to the cloud. We do number crunching there and we really take advantage of the unlimited capacity. So a lot of training happens up at the cloud. A lot of the training happens in the cloud and then whatever models come down, we load those on the edge and we actually do make decisions right there on the edge or we give the operator the choices to make the decisions right there on the edge. So training up in the cloud, but the inferencing actually is approximate to the actual action. So there's locality for the action based on what's in the model and there's a lot of training that can happen. Quite frankly, we don't have to underwrite the cost of the infrastructure to do it. Exactly. So that suggests that there's going to be a fair amount of change in the industry over the next few years and this notion of moving from O-T to I-T or skated to I-O-T, this is not just a set of technology issues. There's some fundamental other questions that are going to be important. A lot of people just kind of assume, oh, we'll throw a bunch of general purpose stuff at these I-O-T related things and it's going to be the I-O-T industry or the I-T industry all over again. Or is really the expertise associated with the use case going to be more important? How is that use case going to be ultimately realized? Is it going to be a bunch of piece parts or is it going to be more of a holistic approach to really understanding the nature of the solution and making sure that the outcome is the first and focal point? I'm going to come back to your question a second. I just always, I have to smile because so I have a master's in petroleum engineering. And so when I studied, I built really fancy models like differential models, integral models and you know, I simulated fracturing process controls built with that stuff. So I lived a good part of my life in I-O-T and then after I came out of university, I really moved more and more into I-T. So I've spent most of my career in information technology including being a CIO. And I always thought that the most fancy math we'd ever do is percentage calculations and that was pretty fancy. And so now I find myself in this awesome place where I can bring together some of that O-T, some of that real deep data science work that I did early on in my life. Now with some of the process and the system implementation expertise and practice that I come out of I-T and they really come together. I don't think one takes over the other. I think there's this real sort of like meeting each other and going like, wow, okay, I guess we really got to work together. And so that's really fun. About your question around what solutions do we see today? I see a lot of very vertical, very one use case oriented solutions that go all the way from the sensor to edge to cloud to hopefully integration to the back office systems because without that you can't really take good action. But they're very narrow. And so like in the good old cloud days when cloud became really big, there were really good point solutions and the good cloud providers sold to the business user right then and then and ran around I-T. And I see the same in I-O-T happening right now. You get a very good solution for temperature control on a truck, for example, which is a very narrow solution. But the moment you want to start doing something with your warehouse where you have other sensors and you need a horizontal platform, those vertical solutions fall short. And so that's what I think is sort of like the interesting dilemma right now. You have these vertical pillars and you have the horizontal platforms that the big providers have. And so it'll be interesting to see when we're going to see some consolidation in this space when some of the vertical solutions going to get bought up by the horizontals to provide better use cases. It's a little bit like the ERPs who did every industry and then eventually they realized we need industrial, industry focused solutions. We'll see the same in the I-T space. Well, the I-T industry has always supposed that we can transfer knowledge we gain in one domain into other customers, into other use cases. And it almost sounds like what you're saying is we're going to have that vertical organization of expertise, which is absolutely essential to solve that complex core business problem. High risk, high value, high uncertainty, often bespoke, never done before. But over time we will see a degree of experience sharing and diffusion so that over time we might see better, more applicable platforms that are capable of providing that foundation for a broader set of use cases but that's going to be a natural process of accretion. Is that how you kind of see it? Yeah, I mean, we're all going to need streaming capabilities, we're all going to need capabilities for machine learning, for cognitive, for video analytics. We'll all need that but I think it'll be specific to the individual use case in a sense of, I'll give you an example. I just had a scientist show me how he started looking at 20 year old scientific research on gearboxes. What frequencies happen in gearboxes specifically to certain scenarios? That's not replicable from a gearbox to a pump. You know, you have different, so there is specific things and yes, it might be the same gearbox in one factory that produces, I don't know, rubber ducks to another factory who makes metal sheets but it's still gearbox specific, right? And so I think this is the specificity we're going to see around models, around learning and around sensors in a certain extent. Excellent, Robert Schmidt, chief IOT technologist at Deloitte, thanks very much for being on theCUBE. Thanks for having me Peter, it was a pleasure. Thank you.