The IoT Show S1Ep4
ALLAN BEHRENS [00:00:28] Good morning and welcome to another edition of the IoT show. So the IoT show is designed to provide some valuable insights from industry experts on topics around the industrial Internet of Things and the Internet of Things. We try and pick engaging hot topics and we asked for our expert guests to give us advice, some feedback and genuine information they picked up from their clients and others in the industry. And we’re trying to treat this as more of a learning experience for those who aren’t deeply involved in the IOC as opposed to purely a sales medium for particular companies. Today, we’re going to be talking about analytics, A.I., machine learning and insights for manufacturers. I’m joined today. I’m very pleased to be joined today by Lisa Seacat DeLuca from IBM. I’ve got Timo Elliott from SAP and Jordan Janeczko and CO from Atos. So hello. What I’d like to do first is just allow our guests to introduce themselves. Lisa, why don’t you start us off and introduce yourself and what you do?
LISA SEACAT DELUCA [00:01:47] Hi, guys. My name is Lisa Seacat DeLuca. I am a distinguished engineer at IBM within our Watson Internet of Things division. So, of course, we’re constantly looking at what we call cognitive computing and artificial intelligence with machine learning for all of our IoT solutions. So this is a great topic for us.
ALLAN BEHRENS [00:02:05] Great. Thank you, Timo.
TIMO ELLIOT [00:02:07] I am a global innovation evangelist for SAP. My background is in analytics, so I am passionate about big data, artificial intelligence and machine learning.
ALLAN BEHRENS [00:02:20] Great. And over to Jordan.
JORDAN JANECZKO [00:02:24] Yeah. Hi, my name is Jordan Janeczko with auto sem seminar of excellence and also our scientific community. Been looking at trends in innovation for quite a long time now. I wear a couple of different hats in this company. I’ve been working in big data and analytics for six or seven years now and I’m also globally responsible for the mainstream application development that we do here at Atos.
ALLAN BEHRENS [00:02:45] Great. Thank you very much, everybody. So. So opening this discussion up. What is it and why? You know, what is analytics and these types of technology like AI, machine learning, etc, what are they in the context of IoT and why are they so special? I’m just keen for the listeners to get a sense of the difference between these terminologies as used in the world in general as opposed to just IoT so Jordan, give us your views on that.
JORDAN JANECZKO [00:03:16] Well, I guess there’s a big difference between what you see going on, for example, in smart homes and more of the industrial IoT sort of space, it has to do with how expensive the end devices are. It has to do with how much control you have over the data you get. And obviously that’s something that helps quite a bit with the way that you deal with advanced analytics and machine learning. You have a much better ability to gather the information that’s being used. You have a much better interest in making sure that things are either running operationally efficiently or that you’re best taking advantage of what you’re doing, either with the raw materials that are coming into your processes or the quality of the products that are leaving your your plant or your facilities. And I think that’s that’s sort of the groundwork of what goes on. You’ve simply got more information that’s available to you. And it’s probably more important once you can get the insights and the information out of that raw data.
ALLAN BEHRENS [00:04:07] Right. How about you, Timo?
TIMO ELLIOT [00:04:10] Well, it’s coming up for the big picture. I think there’s a massive change that’s happening now because of the wider availability and power of machine learning in particular. I think it’s safe to say that every product, service and internal process will automatically get better over time as people use it thanks to machine learning. We tend to use analytics in the form of traditional business intelligence. I have a process, industrial process and then I do analytics to try and figure out how to improve that with machine learning. A lot of that work can stop being done by the algorithms themselves. And the more machines you have, the more data you have, the better the algorithms can start working and learning from any remaining exceptions. And we’ve seen some fascinating examples of that with our customers.
ALLAN BEHRENS [00:05:00] Great. And Lisa.
LISA SEACAT DELUCA [00:05:03] Yeah, I mean, just to echo what everyone else said IoT devices in the data alone are not enough. Right. So analytics is the visualisation of that data and making it so that you can find the insights that you need that are important to your business to help drive decisions based on the data.
ALLAN BEHRENS [00:05:20] Right. And what types of analytics and insights are you finding within your customer base?
LISA SEACAT DELUCA [00:05:29] Yeah. So last time.Yeah. I mean, we’re listening to what our customers want, right? So you can take what they need and what they’re hearing from their devices and the types of business outcomes that they want to see and then derive different insights from that data. So it really depends on the need of the customer. And in the manufacturing use case, for example. We’ve got some cool image recognition techniques where we use a robot to see and take pictures and train the system on what’s normal and what’s abnormal so you can alert when something is coming across the manufacturing line. That’s not necessarily normal.
ALLAN BEHRENS [00:06:08] Jordan, what sort of examples are you using within your customer base?
JORDAN JANECZKO [00:06:12] I think what we see a lot right now is the idea that the data sources are coming from different areas. So it’s not only grabbing information from shop floor devices or other sorts of things. It’s actually integrating that with information coming out of your ERP system or your CRM system, maybe your supply chain management system and taking advantage of all of that information and putting it into, again, these different algorithms, these different analytics activities. And I think that one of the things we see a lot is the idea that a lot of customers are understanding that they need to spend more time thinking about the way the process is change and not only view this as a technology, they need to slightly change the way that they know departments work together. They need to understand a little bit of the differences if you’re getting real time information as opposed to exchanging information in a spreadsheet once a month.
ALLAN BEHRENS [00:07:00] Right. Thank you. And Timo.
TIMO ELLIOT [00:07:03] I think there are some incredible new Internet of Things, opportunities, Internet of Things. It’s really always been about revealing processes that were previously invisible, using sensors to gather data that you can now see and optimise a process that you couldn’t before. But there’s been an explosion in the types of sensors you can use. So, for example, we are working right now with a company that is one of the leaders in creating palm oil. And we’re working with them MI project where they’re flying drones across their plantations of palm trees. It takes pictures and then using image detection machine learning, you get to turn that into valuable information about how fast each tree is growing, whether it’s got a problem, if it changes colour compared to the other trees, you know that it’s a problem and you can send people to try and figure out what it is. So it’s it’s providing a level of visibility that just was never possible before. So it’s introducing a lot more companies to this notion of Internet of Things beyond traditional manufacturing.
ALLAN BEHRENS [00:08:08] Fascinating. What about some of the decision making factors on what to employ and how to employ it? So, I mean, are there any general guidelines or advice you can offer others as to how to go about thinking about employing these types of technologies? And if there’s a difference between the different technologies, just analytics or A.I. or machine learning algorithms. Let’s ask a timo of first.
TIMO ELLIOT [00:08:34] So start with the different terms. Artificial intelligence is essentially what you might call a socio technical term. In other words, it’s a marketing term. We generally use it to mean computers doing things that up until now only people could do is not really associated with the technology. There are various technologies that are associated with artificial intelligence, but the one that is generating most of the real possibilities right now is the advances in machine learning, machine learning and turn is any algorithm that improves itself based on the data. And what can companies do is really not so much about technology. The biggest thing that we’re finding is important is that companies it’s great time to take a step back from whatever you’re doing and rethink some of your processes. We had a lot of success using that design thinking methodology to rethink the customer experience. Looking at every aspect of that customer journey and looking at how new technologies can help.
ALLAN BEHRENS [00:09:37] And the Jordan. How about you from your side?
JORDAN JANECZKO [00:09:41] Yeah, I certainly support with team. I’ve just said I think a lot of it is. I mean, again, you’ll need data scientists. They do their data exploration phase. They need specific tools to help them do that. But then when they’re done, a lot of the people that are using the algorithms are the things that have been created with machine learning or whatever technology you use. They don’t obviously need the same sort of skill sets. They need to understand. Okay. How would this change my process? How would this change my daily activity? And I think also to get into a learning organisation, the question is how do you gather their ideas, their endpoint input about trying to use these different sorts of things? So I think that’s something to keep in mind that some parts of organisation need to understand a little bit more about the details, what they use. Other parts need to understand how it would change, how they do their daily business.
ALLAN BEHRENS [00:10:28] Lisa?
LISA SEACAT DELUCA [00:10:30] Yeah. I mean, we’re just getting started with AI so until we have a broader consumption AI need to improve domain expertise, prove the accuracy of the systems and trust in these A.I. models while also reducing costs and bias as well as the time to value a lot of the systems today. For example, take a lot of time to train the system. A lot of human work to get it to the point where it’s a valuable system.
ALLAN BEHRENS [00:10:56] Right. So let’s talk about the considerations before the investments that one makes in these areas. How does one go about deciding how to invest the money and what sort of trajectory to have on the investment side? Timo. Give us your feedback. If you can.
TIMO ELLIOT [00:11:17] There’s a lot of hype around machine learning, analytics, artificial intelligence. But the key is to start small at one level. The easiest applications of machine learning are wherever you find complex, repetitive decisions that are carried out hundreds or thousands of times a day. So obviously things like preventative maintenance, predictive maintenance. But any or logistics problems, is my supply going to arrive in time or not? I can use the data. I have to work that out. So those kinds of processes where you are just trying to use machine learning to increase efficiencies, to take decisions that used to be handled by people and give them over to these new algorithms machines, that’s typically the biggest opportunity.
ALLAN BEHRENS [00:12:08] Lisa, how about yourself?
LISA SEACAT DELUCA [00:12:11] Yeah, I mean I definitely agree with all of that as well as safety of people within the manufacturing floor, for example, as well as, you know, understanding where your assets are, that track and trace notion of that equipment and as well as the individuals within the floor, any safety issue is always a great use of artificial intelligence. Anytime you’re augmenting, like I said, that data industry experts knowledge like going through, poring through large amounts of data that they wouldn’t be able to do on their own, but with the help of a computer, you can give them answers in real time that improve their job. That’s what we’re looking to do.
ALLAN BEHRENS [00:12:50] Right. What about some areas that might be sort of somewhat easier or more difficult to start on, I mean, have you within your experience, do you find that there are some things that your customers find a much more successful when they are trying to get going in the these areas? Lisa?
LISA SEACAT DELUCA [00:13:14] Yeah, it kind of goes back to what Timo said about starting small and looking at those smaller pieces that you can improve through technology, the manufacturing use case of just using pictures to understand this is normal, this is normal and then all of a sudden the red flag that something is abnormal. That’s an easy thing to train the system on. Something that you can learn as a you know, that’s part of machine learning is the learning piece. You have to have something that you can learn through technology. And if you don’t have that piece, it’s hard to make the connection to where you could use AI.
ALLAN BEHRENS [00:13:46] Okay. What about learnings from your customers? I mean, we obviously always learn lots from our customers. Jordan, have you picked up any interesting snippets or observations on things that perhaps you weren’t aware of when you when your business sort of invested in this area from your customers?
JORDAN JANECZKO [00:14:10] Oh, of course, of course, we’re always learning from the customers. It’s again, very creative ideas about where you can apply different sorts of different sorts of technologies, where you can where you can start using machine learning or process mining or other sorts of tools. I think customers have a very advanced way these days of trying to approach these sorts of things and sometimes for companies starting out. One of the key things they need to do is identify people who are interested in this sort of area. And so what we’ve seen, for example, are some people want to focus on predictive maintenance. Others, if they have a process that’s running very smoothly, don’t see a lot of value in that. But they would like to find different sorts of sensors that they can start using or different sorts of quality measurements they can use on raw materials or other sorts of things dealing with things outside of the standard factory and expanding that into different sorts of areas.
ALLAN BEHRENS [00:15:03] Right. And Timo.
TIMO ELLIOT [00:15:06] There’s a couple of things that jump to mind. The first is gathering and combining all of the right data. Now, this has always been hard, but the volumes of sensor data and the data quality problems. And then above all, trying to mix that sensor data with information from more traditional enterprise systems in a way that is appropriately governed is really quite hard. So there are new data pipeline systems where you can introduce governance more easily consistent, auditable ways of moving information from one place to another. And then the other thing that jumps to mind is that ever walked to customers find difficult is once they work through the technology issues, they realise they have a lot of business model issues and those sometimes can be actually a lot harder to work out. Getting people to pay in new ways with new partners, figuring out who owns what. That actually takes up a lot of our customer’s time.
ALLAN BEHRENS [00:16:06] Right, Lisa? Any anything to add to that?
LISA SEACAT DELUCA [00:16:10] You guys cover most of preventive maintenance is the big one, of course. And then worker safety. Anytime you can keep your machines running and not have that downtime, that’s a huge money saver.
ALLAN BEHRENS [00:16:20] Right.
ALLAN BEHRENS [00:16:46] What about surprising situations and, you know, just opportunity areas or hurdles, anything that pops out from your sort of experience with your customers or installations to dates? Lisa?
LISA SEACAT DELUCA [00:17:04] I think that the biggest thing is kind of what we talked about before on using your existing sensors and equipment and what you can get from that, as opposed to having to create more infrastructure and bring in more equipment to make it possible. So any time you can take advantage of what you’ve already got in place, rather than dumping a lot of money into growing, that infrastructure is really important for our clients.
ALLAN BEHRENS [00:17:28] Right? Right. Timo?
TIMO ELLIOT [00:17:31] What I’m fascinated by is the. This self improving notion that I talked about earlier, for example, working with the city of Nanjing in China. It’s very fast growing. We first worked with them on a project to gather information from all of their taxis and use that to optimise the flow of taxis throughout the city. That was successful. They’ve now extended it to all the other forms of transport. So the city is just full of cameras and sensors and all of that information gets fed into a big machine learning model and then is used to optimise the traffic flow in real time and again because it’s using machine learning. It’s getting better and better at it over time. So it can make adjustments to the timing of traffic lights, for example, based on the time of day or whether there’s a large sporting event or whether it’s a school holiday. And so the system is optimising itself to a to an amazing extent. And that’s a glimpse of the future for every kind of business.
ALLAN BEHRENS [00:18:35] Right. Right. Jordan, any any points you want to add to that?
JORDAN JANECZKO [00:18:39] Yeah, I think so. I guess I mentioned to sort of things that that we’ve seen. One is that data governance is becoming more and more important. There are obvious advantages. If you’re grabbing information coming out of machines, it’s formatted for you nicely. You know, when maybe you’re missing some information because of bad connectivity or whatever it was. But data governance between organisations is very important, especially when you’re trying to bring information out of different systems, some of them that aren’t simply times here, information that’s being collected through or OPC UA or whatever your protocol of choices is. Sometimes we’ve had to interrupt the data science work that we’ve been doing with customers and we actually go back into a consulting exercise about which departments are the single source of truth for information coming out of other sorts of systems. Who has the responsibility and accountability for making sure that you’ve got good data from other systems and that people are using it properly so I think that’s one set of things going. And the other thing that’s going on is that people are realising that algorithms themselves or the results of machine learning activities are assets in themselves. So you used to focus primarily on the physical devices, but now more and more as they’re playing a significant role in sort of making automated decisions. You realise that actually the the the algorithms become assets. You need to have an asset management system that includes those. You need to have a good versioning system for what’s going on. And people need to understand again how that’s impacting the process.
ALLAN BEHRENS [00:20:09] Hmm. Fascinating. So I’ve got two last question. The first one is so obviously new technologies. We’re talking a number of the companies that are watching this, all new starters, perhaps some of them are on the fence. What would you say to those companies that are on the fence and making that decision? Lisa?
LISA SEACAT DELUCA [00:20:34] Just do it right. It’s a low barrier to entry. There’s a lot of API is available. You can search the Internet and see it say API is for AI and machine learning and there’s a million things out there to just get your hands dirty a little bit and try it out. And then it starts you creatively thinking about how you can apply some of these technologies to your own business.
ALLAN BEHRENS [00:20:55] Thank you, Timo?
TIMO ELLIOT [00:20:58] Echoed that sentiment. You definitely have to do enough research to know what you’re doing. If you’re deciding not to move forward right now, it’s not something that I’ve seen from any customer. Everybody I know is moving forward with some project along these areas. The technology is absolutely mature enough to make big business opportunities right now.
ALLAN BEHRENS [00:21:20] Great. Thanks and Jordan.
JORDAN JANECZKO [00:21:23] I would agree there’s big business opportunities. But that doesn’t need to be the only thing you can be doing. The idea of just starting with something that you’re very convinced about learning as an organisation, how to deal with these sorts of things is also something that needs to be put into that sort of return on investment calculation. I think everybody is. Everybody I know at least is convinced that this is the direction that everybody is headed, trying to bring these new sorts of systems and and algorithms into the way they do daily business. And so part of it is learning how to deal with that. And part of that needs to be a start and learn. And after you learn what’s going on, make sure that you spread that within the different organisations or the different parts of your organisation and keep learning and keep building upon what you’ve done for second and third.
ALLAN BEHRENS [00:22:10] Great advice. Great advice. So finally. Oh, sorry. Timo, go.
TIMO ELLIOT [00:22:15] I’m sorry if I could just add to that. One of the big new opportunities, as I mentioned, is we design thinking is approaches that are really oriented around fast, agile prototyping to test things out with small groups of people. You can do that much cheaper, more agile way than ever before. So you don’t have to embark on a huge, big, expensive project. You can come up with an idea and test it quickly with a small audience relatively cheap.
ALLAN BEHRENS [00:22:43] Good advice. Good advice. So final question. Where can people go for help and advice on this, Lisa?
LISA SEACAT DELUCA [00:22:51] Well, that’s very similar to last month, definitely. The Internet has more resources than we can ever learn about. I like to go on Twitter and follow influencers around AI and machine learning and just see what their latest and greatest things are that they’re tackling. I just retweeted a message about the 26 useful interesting news cases around AI and machine learning. So it’s kind of fun just to read about what other companies are doing to help you get ideas for how you can apply it in your business.
ALLAN BEHRENS [00:23:21] Great. Jordan?
JORDAN JANECZKO [00:23:23] Yeah, I’d say obviously you can go. I mean, our company has consulting services. A lot of companies do. I think in spite of that, I’m going to say something counterintuitive. In addition to taking advantage of the wealth of open source tools that are out there, go into your own company and find people that are interested in what’s going on. It’s such an important topic. It’s difficult to imagine a large organisation with with technical savvy that won’t have people that are personally interested in this and personally willing to invest some time into what what can be done.
ALLAN BEHRENS [00:23:57] Right. Timo?
TIMO ELLIOT [00:23:58] So first I would definitely urge people to go and look at best practice across how their industry is. One of the interesting things is there’s much more blurring of industry boundaries that year. If you’re an oil and gas company, you can probably learn something from a luxury retailer. Now it’s fascinating to what extent the technologies are transversal and people are doing very similar things ultimately in retail or healthcare. So I’d go and read widely. Don’t just stick to your own industry, of course at SAP. We have a number of IoT accelerators that we have created with our customers around high opportunity business areas. And then I would echo Jordan’s comment that you look internally and find people who are interested in moving the company great forward. Creativity, internal hackathons have been really interesting and every company that I’ve seen that’s run them. So you find it open to anybody who’s interested. You look around for data sources. You ask people to suggest data sources and suggest uses for that data and you bring them all together with some agile platforms. Typically cloud based to allow them to hack that data together from devices display to small applications and just see where it goes.
LISA SEACAT DELUCA [00:25:18] And to add to that, I just want to say that when you are coming up with these teams, make sure they’re diverse because one of the problems that you’re going to see with A.I. is they’ll come with the biases of the people that are creating and training the system. So get a unique and diverse group of people that are helping to create your A.I. models.
ALLAN BEHRENS [00:25:36] Great. Excellent advice. Thank you very much for that. So just for our listeners delight. We’re gonna have some key takeaways. If you have a look at the link on the on the Web page following the the production of this episode, you’ll see some key takeaways. Feel free to go and download them. Obviously, much more interesting to listen to our guests, the fascinating conversation. I’d like to thank Lisa, Timo and Jordan for your time. And I hope you all thought that was really interesting. Thank you. I certainly did.