 Hello and welcome. My name is Shannon Kemp and I'm the Chief Digital Manager of Data Diversity. We'd like to thank you for joining the current installment of the Monthly Data Diversity Smart Data Webinar Series with host Adrienne Bowles. Today Adrienne will discuss transform industries with AI, manufacturing and retailing. And retail, just a couple of points to get us started. Due to the large number of people that attend these sessions, you will be muted during the webinar. For questions, we'll be collecting them via the Q&A in the bottom right-hand corner of your screen. Or if you'd like to tweet, we encourage you to share highlights or questions via Twitter using hashtag Smart Data. If you'd like to chat with us or with each other, we certainly encourage you to do so. Just click the chat icon in the top right-hand corner for that feature. And as always, we will send a follow-up email within two business days containing links to the slides, the recording of this session, and any additional information requested throughout the webinar. Now let me introduce to you our speaker for today, Adrienne Bowles. Adrienne is an industry analyst and recovering academic, academic, providing research and advisory services for buyer, sellers, and investors in emerging technology markets. His coverage areas include cognitive computing, big data analytics, the Internet of Things, and cloud computing. Adrienne co-authored cognitive computing and big data analytics published by Wiley in 2015, and is currently writing a book on the business and societal impact of these emerging technologies. Adrienne earned his PA in psychology and MS in computer science from SUNY Binghamton, and his PhD in computer science from Northwestern University. And with that, I will give the floor to Adrienne to get today's webinar started. Adrienne, everything's looking good. Hello, and welcome. Hi. What is that? Well, thank you, Shannon, and thanks everyone for joining us today. All right, let's go on here. So in about 45 minutes, I'm going to give you my three steps for transforming your enterprise with artificial intelligence. Any enterprise, any industry can benefit from applying these steps. I would tell you now, but you're not ready. So let's start with some context. We're going to start by thinking about something that you bought recently, or maybe something that you're planning to buy soon. It could be as major as an automobile. As minor as toilet paper. Just think of one object, a physical object that you either just bought or you're planning to buy. Now, think back to a time 20 or 30 years ago, if you're young enough that that's a pain, then think about what your parents were doing in those days. Where did you or your parents buy the same item or its equivalent? Basically, buy a car 20 years ago or a toilet paper or a hair dryer and apply it, something like that. Why don't you think about the difference in experience? Think about what it was like in 20 or 30 years ago. Where was the item that you're thinking about made? How was it made? It was something that was made by hand, by machine, something that's mine. And how did you know then where it was made or how it was made? How did you decide then versus now on how you were going to acquire it? Which product? Assuming there are alternatives. You're buying a car, there are a lot of different kinds. How do you make that decision versus how it was made a couple of decades ago? How long did it take you to decide on what you're buying now? How long did it take you or your parents to do it 20 or 30 years ago? Did you comparison shop? All those things from that. How long did it take you to get what you wanted once you knew what you wanted? Today, we're going to look at how some manufacturers are using artificial intelligence to make the right products for retailers and how retailers are using AI to bring them to you and to your families or to your business. To get started, I want to provide some additional context, if you will, and use a framework that I've been using for a very long time. As Shannon mentioned, I'm a recovering academic. I use this in the very first classes that I taught. We're going to use something from Christopher Alexander. You're not familiar with Alexander. This comes from a book he wrote in 1964. It's a paper back around 1976. I picked up my first copy of this in 1978. I know that because I still have that copy. I've had dozens since then. I picked it up at the Harvard Coupe in Boston in Cambridge. Alexander, in notes on the synthesis of form, wrote about the process of designing. It's an architecture professor known better in computer science circles for work he did on patterns. Patterns for architecture that reformed the basis for a lot of what we do when we're writing software patterns, but this predates that by about a decade. Alexander talks about the idea of building things, physical things, buildings and communities. The form is the part of the world over which you have control. It's what you're building. It's what you can shape, why you leave the outside world, everything around it. I'm going to turn this into a Venn diagram, as it is. We're building what's inside that circle, but it has to fit in the context of the world around us. You can't change the context. You have to be aware of it. The goal is to put the context and the form into what he calls an effortless or frictionless coexistence. When we think about building things, sometimes we forget that whatever we build has to fit in the outside world. I'm going to use this concept of form and context to think about how we create products. This is going to be important for manufacturers. How we put products into our lives and this goes into retail. Basically what we want to do is look at that bright red line or pink line around the form and think of that as the fit. The best thing that we can do is not what's inside our form. It's to make the interface such that we make life easier for the people that are using it. Those that are in the context. This gets into customer experience, the whole area of customer journeys that's so popular right now. When we look at manufacturing, every manufacturing process represents a form. It's something inside and that has to fit with the larger context around the world. That's part one. The next thing I want to do is just two quick slides on machine learning. If you've been with us on some of the other webinars, we go into a lot of detail about machine learning. Today I just want to have enough context about machine learning to make sense in the comments I'm going to make about manufacturing and retail. Because every example that we're going to use today is going to have some reference to machine learning. We don't need to go into a lot of detail, but core concepts, two things here. When we're dealing with a system, machine learning overall refers to a system that can improve its performance over time by having experience with data rather than explicitly reprogramming. So that's the basic for machine learning. And that definition goes back to the 1950s with the early checkers playing programs. But the two major areas for machine learning are supervised and unsupervised. The difference there is that supervised, when you're building a system, you have to know a lot about the types of data that you're going to be dealing with. And you have to provide some examples, actually depending on the application, quite possibly a lot of examples of the data, and train it to perform. So we have training data sets and then we have to evaluate it versus unsupervised, where basically you're asking the system to identify things, their novel. And I generally like to think of the distinction between those two as dealing with children and you're trying to educate them. The supervised approach is you tell them what you want them to learn. You tell them about things. So you say, here's a picture. It's a dog. And you know it's a dog because it has these properties. Versus, I'm going to give you some latitude here. Go out in the yard and see what you find. And you find a dog and then you come back and tell me about it. So it's telling them what they're going to look for versus telling them to look for something and tell you about it. In the middle I have reinforcement learning. The idea of reinforcement learning is very much, if you've ever taken a site course in the lab where you're doing rats as the simplest form of animal that you're going to deal with in the lab, you reinforce or reward behavior that you want more of and you have negative reinforcement or withhold that reinforcement for things that you don't want. Reinforcement learning and machine learning, I tend to put it towards the supervised side because you're giving the systems that feedback as you're training it. The other two dimensions, general versus deep, any one of those supervised, unsupervised or reinforcement can be done with techniques that are sort of one step or deep learning, which is the biologically inspired approach that looks like this. And here the idea of a deep learning system is that we've got multiple levels or layers, which represent the depth of the model and we start out with something that's very concrete and we try to successively refine our understanding until we get to something that's sufficiently abstract. The example here is, let's say you have a picture. It's a JPEG file. So inside that JPEG file, if you just click on the JPEG file on your computer, you're going to see a picture, but inside that file is the digital representation of the image at a pixel level. So every pixel has a brightness, it has a hue, it has all that good stuff. And if you're looking at it at that level, you don't see a picture. You see digital representation. And so the idea is that in a deep learning system, we can go from that really concrete pixel by pixel to trying to figure out what's in the picture and more abstract representation. The first step in video analytics using deep learning is generally to find edges so that we can find shapes and you find edges by looking for distinctive properties. So if you have an edge in a picture, in general, color changes along a path and you can start to identify that once you've found the edges. You can go through it again and find shapes. Maybe you go from just the pixels into identifying using rules of geometry and say, oh, that's pretty close to a right angle. Maybe I've got a rectangle or a square. You start to pull it all together until finally you figure out what the object is. And the reason this is important is because as we start to look at different ways of changing businesses and we'll start with manufacturing, there are applications where we're going to want to hand this off to a machine learning system to give us information based on successive refinement going from concrete to abstract. Or there are going to be applications along with that. There are going to be applications where we have a lot of training data and we want the machine to take over and start performing these tasks, performing them more accurately, of course, and more faster than we could. But we just need to understand the idea that this is something that with data and with the appropriate algorithms, we can get there. And that's as much detail as we need for the rest of the session today. So now we're going to go into manufacturing. So, and we've done this in the past in the webinar series with a couple of other industries. What I want to do is look at some of the major processes that are common across manufacturing enterprises and then look at the way some companies are solving them and my examples today will be from mostly very large manufacturing operations because I think it'll be easier to see. So, if you're going to manufacture something, we're talking physical objects here, there's a fair amount of planning that goes into it. We need to do product design. We need to figure out what the product is going to be that we're going to build just because we built widgets last year. It doesn't mean we're going to build widgets this year. We need to do some research. But once we know what it is we're going to build, let's say it's a car, it's a refrigerator, whichever, we have to design it, produce it, ship it out to the rest of the world and while it's in operation, we have to be able to maintain those. We also have to do inventory management, not only the products, but all the parts that we need to build the products. So there's a lot of steps to go into it. What we're going to look at today is some of the options that AI enables in particular. The idea of digital twins, automations, this intermediation. I used to hate that word but then I got put into the habit of using it because I talked e-commerce and some strategy in business schools. It's a very commonly used word there. Getting rid of middlemen or improving processes to take these steps out. And then finally, how do we optimize our global supply chain using AI? So we'll start with General Electric, one of the largest companies in the world. It's a diversified manufacturing and engineering company. This was from a blog by Michael Su. The chief digital officer at GE was also the CEO of GE Digital. The article is about the idea of a company like GE which is about 100 years old. Fakes everything from locomotives to engines, very large things to turbines, down to consumer appliances. What if you woke up one day as a software analytics company and that was driven from the top. The idea here is this is from their corporate blog that you can't just connect the machines. There's a big movement and Shander mentioned the book I'm working on. I'm really focused on the integration of the IoT, the Internet of Things, and making systems more intelligent. It's important to just make sure we have the common definition. When we talk about intelligence, intelligence is broadly speaking the ability to acquire knowledge and to apply it. And by apply it, meaning in a new context, and I generally say that at the higher levels of intelligence you can not only acquire and apply it, you can create new knowledge based on inferential reasoning. So what's happening at GE is they've gone from making large industrial machines, we'll focus on those for many of them. They make home appliances, washing machines, et cetera. To looking at the world is where all the machines should talk to each other. The whole idea of the Internet of Things and the industrial Internet of Things. They should be able to report on what's going on inside them to another machine or to a person who would be able to use that data. But they should also be able to have some level of intelligence ability to use that data at the machine as the data is coming in. And so this is a big shift. Certainly a company like GE has gone through automation continuously since the inception of the assembly line you know that they've gone from changing the way processes are laid out to bringing in robots to building robots to help build machines. But now it's not enough to have the automation. They want to have the intelligence distributed to the tools they're helping them make the products but also distribute them, distribute intelligence within the products so that they can communicate with each other. So in the blog here they talk about the idea of connecting this and then you can take the intelligence that you're getting and now we're talking about intelligence also in the military intelligence usage of the word we're trying to get understanding of what's happening and that intelligence represents collecting data about systems as they're being built and as they're in operations at enough detail or at enough of a detail level that that data can then be used as input to refine the operation of the system itself, the machine itself, so a system can self heal when we get to complicated enough systems but also that we'll start to collect data from all of the instances if you will of what we sent out there. All of the washing machines are going to send data back and that will improve our process going forward so we can start to identify trends and do predictive maintenance, something that is going to say as important on a washing machine as it is on a jet engine I guess there's different levels of importance there but it's certainly for any machine that has moving parts. There is a maintenance cycle which in the past was typically done by time. If you think about how your car, modern car indicates to you that it's time for an oil change, it used to be a simple calculation that was either X months or X miles, whichever came first, it was time to change the oil. Now we have usage based maintenance and that's the type of thing that's going into manufacturing at GE and certainly at other places. That was the point here that everything that you build needs to create data that can be used in the aggregate to improve the process going forward. It can also be used for predicting failure but at some point we would like to be able to share the data among the machine needs. It's a washing machine in one place that discovers that certain load configurations cause it problems. Right now it will communicate back to the manufacturer and say this is what's happening and maybe the actual analysis is done remotely, it's done in the cloud but at some point when the system gets complicated enough like a jet engine or a locomotive that's in their product line, you want the system itself at the instance level, at the locomotive level to be able to do some self healing but also to communicate that either through a hub at the manufacturer say this is what's happening and then the manufacturer will send out the update to all the other locomotives or what we're starting to see is that sometimes depending on the machine, machine can talk to another similar machine directly and usually right now that's only happening in the actual manufacturing process but the way that happens is if you have robots that are assembling components that end up in a complex system manufacturing robots today are in many cases and certainly there are simple robots that have a single function but there are robots that can be reconfigured to do different tasks, there are robots that can actually we'll see an example in a minute when I talk about automobile manufacturing but robots can be set up to bid on tasks as if they were humans going for ebay for example. So if a line manager either human or a bot knows what is needed, I need model X locomotives coming up and we know what that requires in terms of production, the robots that are capable of doing those tasks can actually bid on it. We can think about that as being like uber. So I get to the airport and I need to ride home. Right now the way uber is configured is I send out my request, uber tells me how much it's going to be and then different drivers chime in and one of them gets the job. We can combine with machine learning, something like this where I don't have to take the uber price, I can put out the price that I want to and others can bid on it. It's the same today with the most advanced robots that are building these systems. So a company that you think of as a manufacturing company is in essence now an analytics company. That's what's driving it. Continuing with GE for a minute, the idea of digital twins is one that's taking on widespread popularity. I first started looking at it with SAP, building digital twins. But here is GE talking about the idea of building digital twins where it's a software model, logical model that uses data that we have about the physical asset. So we have a real wind turbine and a real engine, whatever it is, a conflict system and now we have a logical simulation model that's built using data about the real thing. And we can build the digital twin actually before we build the real model. If you look at it just as that part, as something that's done before the physical model is made, that's pretty comparable to simulations that have been done in manufacturing for decades. Wind tunnels, for example, the aerodynamic simulation. The idea of a digital twin is you build the twin, do the simulation, and then build the actual physical asset. But once you deploy the asset, again, it can be something, a wind turbine in a remote area where you're not going to want to go maintain that, go actually see it any more than you have to, have the physical asset fully instrumented with sensors that are connected to the IoT. Those send data to the digital twin and the digital twin can then run simulations based on the current real data and do what if scenarios, analysis of various operating conditions. And then if you're getting information from all the digital twins, now we can, again, take a look at this and say, all right, what's the pattern? What are we seeing? And that gets into the idea of supervised versus unsupervised machine learning. We may train the system by training the digital twin to see how it learns on this training data. But we may also have unsupervised learning where in the real world these deployed physical assets are getting information or producing information that's going out in the sensors and saying, this is what we're really seeing. And goes back to something that Gauze and Weinberg wrote in a book about requirements where the map disagrees with the terrain, believe the terrain, well, your digital twin can be really good. But it's the sensors that are providing information off the physical asset. If that doesn't agree with what you've modeled, then you need to change the model, you need to change the twin, and that's why these things need to aggregate that data. But it all comes down to having the ability to do machine learning once you're generating and gathering all this data. Take a look at another company and there's a cheap in here that I will acknowledge in a second. So Boeing, one of the world's largest manufacturing companies around the world, headquartered Seattle, but pieces for aircraft are made in a lot of different places. The typical jet engine today has about 5,000 sensors on it and each of those sensors is reporting on a continuous stream. Some of that data is held on board until the flight, the mission, is completed and then it's downloaded and batch. Some of it is reporting information directly into the cockpit. There are things that a pilot would want to know about right away. And it's up to the manufacturer again, we're talking to the manufacturing industry here to determine how we want to first of all which data we're going to get and how it's being reported. So some will be streamed, some will be batched, but the area that I think is interesting is that once we have this data, what do we do with it? Sure, you can use it for predictive maintenance. You can use it for rebalancing your supply chain if you start to see patterns in the data across planes and we can again use machine learning to see what were the conditions under which that engine was operating and start to allocate replacement parts and put them in places where you're expecting the failures or you expect to have a need to replace parts. Once on a jet that had to shut down because the person walked out in front of it and then next thing you know, plane wouldn't start up, we had to wait for parts to be shipped in. Unfortunately, it was only a few hundred miles. But the parts that are replaced less frequently, they don't have as many of those and so if something goes out, it takes a long time. Time is money. The asset, the engine is attached to you without a service. So by using machine learning about that data, that's during the operational phase. It's also being used, machine learning at Boeing for product design now and maintenance. This has been made public recently. This was in DA for a while. But Boeing is working with Microsoft and these mixed reality platform with the HoloLens to allow engineers to walk through a virtual model of aircraft and parts before their manufacturer. It's used to do some predictions. It's used to change designs again before these things go on the line. And then the same type of model is used for training people to repair. So you can actually wear the glasses and see what's inside a more complex component before you start to disassemble it. In addition at Boeing, they are using cognitive and IoT services from Microsoft Azure to do predictive maintenance and fuel optimization. So by looking at the data that's coming in from the planes, you can start to look at it and say, okay, what are the conditions that we're aware of that made this engine use more fuel on this route than another one? Is it weather? Is it time that it was used? Temperature? Altitude? Put all of that in and that's where the combination of the stuff that you know goes in with supervised learning and stuff that you don't goes in with comes out with unsupervised learning where you're finding these patterns. And it all works together. So that's how Boeing is handling it. More manufacturing example. General motors. And this is my, they don't build them like they used to slide. Anybody who's sort of followed what's going on with the automobile industry knows that it's in a massive upheaval right now. Ford announced that they're going to stop making a lot of models of cars for sale in the US. GM has gone through some changes. They're all dealing with the types of manufacturing robots mentioned that improve the operational performance, the actual construction performance, if you will. That's been going on for years. What's new is that now as we move towards autonomous vehicles, the machine learning skills and autonomous vehicles are all used, machine learning, in particular reinforcement learning for a lot of tasks. Now just for fun, I pulled out some employment ads this week. If you see the type of engineers that are being recruited at General Motors, the big push is on people who know analytics by statistics and AI. AI, AI, AI, it's all about preparing to, I don't want to say take the intelligence from the driver and put it in the car to change that dynamic and extend the car able to not only make decisions but to communicate those decisions and to be able to communicate with other systems. And those systems can include cars, they can include smart street signs, things like that. But all of that requires an understanding of machine learning. And this is why it's already been used in production. Now it's being used gearing up for, pardon the pun, the whole idea of autonomous vehicles. And so my thought on this, and this is the hierarchy of autonomous vehicles from the society of automotive engineers, everybody's moving from no automation to 57 Chevy in the background to automated systems that will get us the full automation. So first the robots started making their cars and then they started talking among themselves to get smarter. That's where manufacturing robots can, based on the experience they have in assembling pieces, take that knowledge if something goes in a way that it wasn't, that it hadn't in the past, there's a new pattern to look at. They can share that knowledge either at the end of the shift or it can be in some cases of assuming that it's critical a failure or an anomaly in one robot might be communicated and shut down the whole line. So that's where they go. And now hopefully learn to drive them. So I just thought it was interesting that today in, this week I looked on indeed, six pages of job listings for engineering and learning of engineers in autonomous vehicles. Let's move on to retail. So to search, I'll give you some pictures to go along with the, I think, with the webinar today, I did a search on shopping on one of the photo sites. What I thought was interesting is there are no men. And free picture is of a woman shopping talking about stereotypes. We want to look at the whole retail experience which certainly involves, I would say, every, everyone, every age, every gender, every combination, today acquires some products or services. And that's retail. Well, retail is hard. And here's my sad example. This is the hardware store in my little town. It's been in business for 27 years and it's closing at the end of the month. There's a couple of quotes from the owner. Family friend, nuts and bolts don't pay the rent. Amazon has been a big game changer. Everything we sell can be purchased online. So after 27 years, they're going out of business. What can the small guys do? You need to understand what the big guys are doing to get there. So within retail, the basic processes you've got purchasing. You have to decide what you're going to carry, inventory management wants you to have it. Where the products go in the store, customer service. You've got to get people to your store, even if it's a virtual store. This for retail, this works absolutely the same for physical stores and online stores. Go back to John Lawnmaker, well-known retailer started out of Philadelphia in New Century'sville. We said half the money I spent on advertising is wasted. The trouble is I don't know which half. And I would say that today, because we have better analytics, we're probably still wasting half our money, but we know which half, and so next time we'll change it. So the things that are changing the retail industry include changing consumer habits and expectations. You've got more choices, but most importantly, you have the population that has access to more information. The reason I asked you to do that mental exercise at the beginning is to think the difference between buying a car 20 or 30 years ago and today. There's a big difference in the cars that are available. The technologies and AI have improved the cars, but the process, you get a lot more information. You spend a lot less time going from place to place. You can comparison shop online, even if you're going to end up buying the car, the piano, the washing machine, whatever it is, in a physical store. So the question is how do we reach people at the right time and how do we make all these decisions? Where does machine learning come in? I'm going to start. I'm going to have a brief look at Walmart, the largest retailer by volume, by revenue. Walmart has long been known as a company that optimized their supply chain. They shared data. They built systems with their suppliers, with their partners. They were able to drive down their costs by sort of reaching into the IT systems of suppliers to help them optimize what was coming into the store, sharing all that information. Back when I was teaching e-commerce, a big part of the strategy we had to look at is where else can people get things? What options are there and how do I simplify it? They've been doing that for decades at Walmart. They're very good at it. Recently, last year, announced a new venture called Store 8 that is owned by Walmart. And basically, it's the laboratory that they are looking at retailing of the future. They're looking at trends. They've always had that research. But now they're investing in startups that they can ultimately improve the Walmart experience. So it's primarily done in order to get AI talent and to get people in an entrepreneurial space using machine learning. The reason I say that is we start to look at what they're doing for a process standpoint whose Walmart's biggest competitor these days wasn't around a couple of decades ago. It's Amazon. You have to look at what they're doing and how that's disrupting Walmart. So Walmart applied for a patent last year on a retail subscription in the Internet of Things environment. Basically, this is a system that we'll use machine learning. We'll use some artificial intelligence. And it'll compete with Amazon Dash. It has fewer clicks and that's always the goal. And so it's trying to do automatic replenishment using some predictive modeling, using the data that's generated by tracking things once they leave the store, things that you purchased. And just as a side note, one of the things, one of my associates, Nicole Specialli, who's done some research on this recently, said that one of the areas that we could start using machine learning and video analytics is in the area of loss prevention. And certainly every large store has an issue with loss prevention. The idea is that in some places employees, even if they witness theft, are not allowed to approach the person. Have them leave the parking lot and call the police. Well, if you start to have tags at a finer level than the big clothing tags on every item, you can track its movement throughout the store just as you track people at the store. And then if we add video analytics, we can start to capture information about faces. We can compare those to databases. We can start to automate the loss prevention. This one is taking a step further in terms of those tags. Once the item reads the store, they want to still track it so that they know when something has been used so that you can set up to automatically replenish. I thought it was interesting to see Walmart. I was like, Walmart is being distributed in terms of sending stuff out to stores. But I didn't realize that many places they're doing research in these areas. These are all machine learning jobs. A couple of them are duplicates because they're on different platforms. But it's a pretty big push and clearly Walmart sees that as part of the future. Kroger. Kroger has been around for 135 years, biggest global supermarket chain by revenue. This picture there is from an interview I did with Kroger executives a couple of months ago. And we were talking about their use of analytics and where it's getting into machine learning. They're known for their focus on freshness and customer experience. They do a lot of historical data capture and they do customized coupons like a lot of companies do based on that historical data. But what I thought was really interesting and gives them the opportunity to do some new things. They actually track people's movement throughout the store and their proximity to other people within the store and the way they cluster using heat sensors. And so, for example, if you get four people that come in at around the same time and they're all over in the meat section at the same time, their algorithms are trying to figure out if that represents a family of four or if it's a coincidence that you have four people over there. And the difference is if you have a family of four, that's going to be one checkout. And so, versus four if those people are not shopping together. And by family it's a unit that's traveling together. And they use that now for things like allocating people to register so they don't have to wait until there are 10 people waiting to get me another register opening. Okay. Based on the number of people that we estimate are clustering together, we can move people around and be there when you need it, not in response. So it's predictive. You can start to take that with some of the video analytics that are out there today. Some interesting companies doing work on everything from emotion detection to age gender ethnicity detection and start to do predictive offers in the store to people based on their behavior and their demographics even if you don't have personally identifiable information. The north face. All right. Almost done with the retail examples but the north face I had to put in because back in 2013 when IBM first announced the Watson ecosystem that they were having partners build systems, north face was one of the first companies that started using Watson experiments back there. If you go on the site now you can use the Watson powered app to help you select items. And it will ask you things like in this case I would add it for a jacket. What are you going to use it for in natural language you can put in. I'm just going to wear it in the winter in Chicago or I'm going on a camping trip. Where are you going? How long? What season? And have an interactive dialogue where it narrows the field and gives you a more personalized solution. Probably if you've looked at analytics you've seen that on Friday nights I guess it is people by beer tend to buy diapers and all jokes aside there's a lot of different scenarios that end up in that same result. But what's interesting to me is the way companies like CVS are now aggregating data as or using the aggregation of data as the basis for corporate mergers and acquisitions in the retail space. So CVS is merging with etna to be able to provide more personalization and to do things like combining the data from CVS they know what prescriptions you're taking. So that's good data and etna which has a lot of information on outcomes and the combination if you happen to be an etna customer and a CVS customer is that you can again get more personalized attention. So all of these have been pretty big organizations. What if you're just a little three and a half billion dollar bomb and pop outfit. These urban outfitters are urban which is the parent that owns urban outfitters terrain anthropology and a couple of others. They recently hired someone who had been doing machine learning at gate martin to run a machine learning unit at the corporate level looking to do logistics optimization and product recommendations and personalization and of course recommendation engines are things that we've long associated with machine learning but I did have the opportunity this week to talk to the fellow that's doing this job and I said one of the interesting things is you're looking at this within urban outfitters. You give a talk at an event that I attended at Google and they were really focused on things like within a line with analytics and just take urban outfitters as an example. They have very detailed data on different parts of the country, different stores, what type of dress cells based on whether it's a floral print, a solid color or a stripe and sleeve length. So take those different variables and they can start to predict seasonally what they're going to need and what gets shipped from one store to another when it's not selling. But they're looking at how to take this deeper and my first thought on this clearly I'm not a shopper because I'm not a representative on the page with all the people that are shopping but I live in a town that has four different stores owned by this parent company anthropology, urban outfitters, free people and another one called terrain, all of these and they serve different populations but I didn't even realize they were all owned by the same entity until this week. They started talking to the fellow about it. Now we can say, okay, you don't have stores all over the country. The fact that you've got four of these stores in one town of 30,000 people tells me that there's a lot in that demographic that you should be able to start doing using machine learning across store across brand. And right now we're going to advise some clients on using artificial intelligence in the contact center. So you can use information about the person that's calling in and their case and use that to guide which agent is going to help in the contact center. But you can also use that to guide the response that the agent provides since the whole idea of guided agents in the contact center using machine learning is something that's just starting to catch on. And it occurred to me looking at this, if you have a family in a town that has multiple stores like this you can start to collect information and use that to guide them and physically say, you know what? I see what you just bought at one store here. We've got something else at the other store that compliments it and compliments this member of your family. All of that is made possible by saying here in the third bullet, use artificial intelligence in real life to guide the customers to where they can see what they want to see and where they can get things that are complimentary to what they have, what they've just shown the preference for. So excuse me. I'm going to give you the three recommendations and then pull it all together with one thought. So the three things you need to understand if you're going to reinvent your business with AI, start with understanding what your enterprise values or where it values natural intelligence. So where do you have people that are key to the company, not because they do something quickly, but because they add insights, they can take information from a source and synthesize it with other information and apply that knowledge in new circumstances. Is it if you're a high-end retailer, it may be the people in the jewelry department. I'm just making that up. I haven't done the analysis. If you're a manufacturer, it may be people that are doing the planning for maintenance costs. There's generally a small subset of personnel in those roles that adds the most value based on the information. The recommendation is to get the data that they have, get more data, use machine learning to augment them, not to replace them, but to make them more effective. So start by figuring out where the natural intelligence comes in, augment those roles with machine learning. Where it's where the company values, where you're rewarding, where you're compensating based on high performance for tasks that are perhaps difficult to implement. Or the data doesn't change very much. That's where you want to start to automate. And we don't have time to get into whether AI is going to create or remove jobs, it's going to do both. But this is the distinction. Where natural intelligence is the priority, augment it. Don't try to replace it. Where repetition, even if it's logical, if it's a call center, there will be some aspects of that that can be automated. Just finished writing the paper on this whole idea of guided support, which would come into play for both manufacturing and retail. You want to be able to handle the mundane things with an automatic response. You want to handle the custom cases, the human response. That's the first two. And then third one is look at where you have data and really do an analysis because a lot of data is out there that nobody is touching. I hate the term unstructured data, but the reality is that in many cases there are notes that people keep, there are records that people keep that aren't used for decisions in real time or in customer time, if you will, because it's too hard to use that data. So identify the tasks and then do that partitioning between augmenting and automating, and then identify the data and create applications using machine learning. It will depend on what the data is and what you want to do with it, whether you use supervised or unsupervised or reinforcement learning. But any data that you're collecting, there's probably value to it that's going untapped. With that, I'm going to close with some thoughts from a session at Google. You may have heard that Google has had a big event going on this week called IO with about 7,000 people there, but the day before IO started, we had a small briefing going over Google's AI strategy. And before I got there, I didn't know how appropriate it would be to the talk today. But if you look at their vertical industry strategy, they're starting with seven industries and that includes retail and manufacturing, which is pretty cool since we pick retail and manufacturing for what we're doing on today's webinar months and months ago. But here's the thing. Their definition of machine learning as a way of creating problem-solving systems is perfectly aligned with everything that we've been saying here. You can use it. You just need to understand the types of data that are available, the types of tasks that you want to do, and whether you need supervised or unsupervised reinforcement learning. Google has re-branded them themselves, if you will. It's not actually a marketing brand, but saying that they are an AI-driven company. They're an AI company now. And their goal is to allow to make every company a machine learning company. And I was so happy to see that. Obviously included is my very last slide here, because that's the point I want to make. If you can do those three things that I just said, identify the stuff where intelligence is valued, identify the stuff where intelligence isn't valued, and then identify new data. You can build a machine learning company or build a company that improves its performance almost immediately using machine learning. With that, I'm going to turn it back to Shannon. Thank you for another great presentation. I just love it. I don't have any questions coming in, but if you have any questions for Adrian, submit them in the Q&A section in the bottom right-hand corner of your screen. And to answer the most commonly asked questions, I'll be sending a follow-up email for this presentation by end of Monday for this with links to the slides and links to the recording. Everyone is quiet today, Adrian. Sorry, I was on mute. I can live without quiet today. But seriously, if you have questions, follow up. Oh, yeah, follow up. Indeed. And so, oh, and I love that topic for next month's natural language processing, always a hot one. All right. Well, I hope everyone has a great day. Thank you for attending today. Adrian, thank you again for another great presentation. And I hope we'll see you all next month. Thanks. Take care. Thanks. Bye.