Maintenance data collection

Big Data and Predictive Analytics

Joe Berti, Clockwork Solutions: Applying Big Data to Cut Costs, from Field Service East 2013

Video transcript:

So I am here to talk about a different technology. I am here to talk about Big Data and Predictive Analytics. It's new; most of us don't know what it is. So I am going to actually take you guys through this. It's going to be a simple message but it's also important to actually understand what this is and why it's important. So there's a set of emerging technologies that come over time that you, that I believe we need to pay attention to as they apply to this particular industry.

So what is Big Data? Why is it important? I am actually going to take you through some real world examples. Really the take away, I am going to give you the conclusion first. I want you to actually start to build the strategy around Big Data and understand why it's important to field service. At some point of time if you haven't already, you are going to hear it with your IT guys or your CEO or your business people or even analyst talking about this area. And so it's good to be prepared about what it is and you'd say, hey, I heard about the conference and the guy was completely blown up. So you know, everyday we are creating 2.5 quintillion bytes of data. So it's something that EMC and others here in the room love to hear. It's basically 90% of the data in the world today has been created in the last 2 years and it's growing, and growing, and growing. The problem is current data base is in systems can't actually handle that size of data and so there's a new set of tools coming out and then there's something on top of it that's called predictive analysis that's trying to take that data and predict the future, so it's actually taking that data and making it intelligent. So it's not just analytics, some people said, hey, aren’t they just my analyst, looking at the data. It's actually completely different, there's a technology behind it. It's also not a Hadoop technology projects. You here people talk about Hadoop, but that's actually their regional technology that powered Google. I don't even know Google still uses Hadoop. So but it's, you are hearing people talk about Hadoop as it relates to this.

So there is technology behind this. It's actually a new set of tools and it's still emerging and you are going to see it continue to evolve and get better and better over time. It's a way to deal of large amounts of information, it does include standard databases but it actually includes other things. It does … it could include things like Google Big Query, Column-Oriented Database or different type of way to query data. It could actually involve Hadoop, it does … Hadoop does have some application within this. It includes potentially in memory databases to eliminate discreets for certain pieces of the data, certain indexes within it. It includes algorithms and heuristics that derive conclusion to help with this data and they do it quickly and I will give you some examples of that. It would also involve techniques to generate and store massive quantities of data. So think about the impact of that.

So, it’s in a way the Take Away in an emerging area, the technology is rapidly improving. The biggest big data project you could think of is Google itself. When you Google, you get immediate results. So think about taking your business data and having some sort of Google but with prediction engine on it. So think about Google, trying to you know generating predictions in the future. So also associated in these areas are new role, Clockwork as team of data science have been around for about 10 years and actually the number 2 job was CNN. This past year was a data scientist, so they are very excited. And they didn't even know that they were going to be one of the top roles in the country. A lot of people still don't even know what a data scientist is. The way I am defining it is someone who can derive information from massive quantities of data but not just as an analyst, they are doing in a much more sophisticated way and they are drawing conclusions and making decisions from this information. So that typical profile someone at work, bringing in as a data scientist, someone who’s a graduate degree or a Ph.D. in Math or Physics is often the role. They are oftentimes an analyst, they are oftentimes are programmer, so they actually write code in the fly and come up with solutions, and the scientist all in one so they can actually apply math and heuristics to real problems. They have the ability to do Applied Math. I took a lot of Math and I was always annoyed by the fact that I couldn't take majority of those algorithms and figure out what their used for. And so I would you know, I would probably drive my professor nuts to say what do you do to this? What is this for? Where do you use this algorithm for? And sometimes they didn't know, and sometimes they did. But I was always looking for ways to actually apply the math and make it useful and so that's a different breed of person, it's not everybody can do this. It's actually a real hard to find skill. The other thing is they don't wear a crash helmet, so they can actually walk and talk and speak to an executive. So some high level companies are now having side by side with the CEO and CXO team, they are having data scientist who work on the side and solve problems. So they are kind of a think-tank that are doing things to quickly figure out how to solve a particular issue.

The take-away is the reason you need data scientist, as why can't you just software to this as the data is never perfect and it's never going to be a 100%, right. If you look through your data, the history on it you can always derive a prediction from it because the history may not be accurate and they have been improperly recorded. It may have lost some of it along the way. There may be noise in it like, for example, if you are getting data for censors, there may be messages to it. So you need a data scientist who can look at it and help associate with it.

A couple of case studies, so to say what are some examples. Obviously Clockwork does this with assets, we are looking, we are making predictions on the assets rolling up all the history associated with this. But we are also looking at the failure rates of particular components on the rate and we are working with the engineers and we are working with the people on the field and interviewing them to say okay, how does the stuff really fail? What really happens out in the field? Stuff that you guys know inside and out and you are talking to customers and you are looking at this happen overtime. You tend to know sometimes more by just the history of speaking to the customers sometimes than what you can actually derive from the data. So through that, we basically build a model and example is, this is a drill shift fleet where you can actually see, you know the top line is basically, you are looking at a comparison of fleet availability to the cost of spares. And there's actually an inflection point in which you get to the point where putting more spares or more components is not really going to increase your fleet availability anymore. So you can that there's, that there's a curve in which the two inflect. That end, by taking the historical information and predicting into the future. The net result was a $2 billion in additional revenue. So by keeping this fleet up and running, it's going to generate that much revenue. That's a lot of money, right. So you think of ways, okay what is, why is this a big deal? Well, you just took a lot of data, you took a lot of history, you applied some analytics to it, you applied a data scientist to it. You did some interviews, you build a model, and then from that model you'd generate some forecast and it's significant. So based upon the 15 year life cycle, the actual shift.

So another example in this particular case study on battle damage repairs, so one of the difficult things is the way the assets are used. It could be different, like these particulars assets were used by the marines and the, you know, in this particular case there's kits that were derived. You say okay, what are some kits that can be built to actually maintain these vehicles and have less, you know, less visits into the shop and increase the actual uptime. The problem is these vehicles are used in the desert. Sometimes they are used in just normal uses on the road. They are using in high temperature, they are using freezing cold temperature same vehicles and then they are put on planes and moves so the vibration profile on them is different. So we actually went through a bunch of different studies where we looked at all different scenarios, we looked at the actual configuration of each of the individual assets. And so think about that amount of data that we had actually go through to generate these predictions. But the end result in optimized kit saved over 24 … $26.4 million in one year all right. We also have another case study for the army … we are just … we can look at their ground fleet alone and save more $7 billion by optimizing and using predictive analytics for their ground fleet. So that's not, that's not even in the entire army fleet. So there's pretty, there's some real nuggets in the data that you currently own today.

So conclusion, so I will just set a simple message. That read of a simple message is start a study of this area, start to understand it, start to figure out why it's important. Think about the data you have, you have a lot of history you know a maintenance history out in the field. You have parts usage history; you have information in your mechanics heads that you know field service technicians. You've got information that your customers currently have. So think about ways to put that together. Come up with a predictive, a big data, predictive analytic strategy to derive the solutions and then start to look at tools to help you do this as you go through and try to figure out ways to generate value. But do it value based, this isn't about technology, this isn't, this is about applied technologies taking real technology in solving a problem. So figure out ways where you can actually come up with a value proposition to say there's real dollars here, there's real savings and then lot of technologies support that going forward. So I guess real quick

Big Data and Predictive Analytics

Joe Berti, Clockwork Solutions: Applying Big Data to Cut Costs, from Field Service East 2013

Video transcript:

So I am here to talk about a different technology. I am here to talk about Big Data and Predictive Analytics. It's new; most of us don't know what it is. So I am going to actually take you guys through this. It's going to be a simple message but it's also important to actually understand what this is and why it's important. So there's a set of emerging technologies that come over time that you, that I believe we need to pay attention to as they apply to this particular industry.

So what is Big Data? Why is it important? I am actually going to take you through some real world examples. Really the take away, I am going to give you the conclusion first. I want you to actually start to build the strategy around Big Data and understand why it's important to field service. At some point of time if you haven't already, you are going to hear it with your IT guys or your CEO or your business people or even analyst talking about this area. And so it's good to be prepared about what it is and you'd say, hey, I heard about the conference and the guy was completely blown up. So you know, everyday we are creating 2.5 quintillion bytes of data. So it's something that EMC and others here in the room love to hear. It's basically 90% of the data in the world today has been created in the last 2 years and it's growing, and growing, and growing. The problem is current data base is in systems can't actually handle that size of data and so there's a new set of tools coming out and then there's something on top of it that's called predictive analysis that's trying to take that data and predict the future, so it's actually taking that data and making it intelligent. So it's not just analytics, some people said, hey, aren’t they just my analyst, looking at the data. It's actually completely different, there's a technology behind it. It's also not a Hadoop technology projects. You here people talk about Hadoop, but that's actually their regional technology that powered Google. I don't even know Google still uses Hadoop. So but it's, you are hearing people talk about Hadoop as it relates to this.

So there is technology behind this. It's actually a new set of tools and it's still emerging and you are going to see it continue to evolve and get better and better over time. It's a way to deal of large amounts of information, it does include standard databases but it actually includes other things. It does … it could include things like Google Big Query, Column-Oriented Database or different type of way to query data. It could actually involve Hadoop, it does … Hadoop does have some application within this. It includes potentially in memory databases to eliminate discreets for certain pieces of the data, certain indexes within it. It includes algorithms and heuristics that derive conclusion to help with this data and they do it quickly and I will give you some examples of that. It would also involve techniques to generate and store massive quantities of data. So think about the impact of that.

So, it’s in a way the Take Away in an emerging area, the technology is rapidly improving. The biggest big data project you could think of is Google itself. When you Google, you get immediate results. So think about taking your business data and having some sort of Google but with prediction engine on it. So think about Google, trying to you know generating predictions in the future. So also associated in these areas are new role, Clockwork as team of data science have been around for about 10 years and actually the number 2 job was CNN. This past year was a data scientist, so they are very excited. And they didn't even know that they were going to be one of the top roles in the country. A lot of people still don't even know what a data scientist is. The way I am defining it is someone who can derive information from massive quantities of data but not just as an analyst, they are doing in a much more sophisticated way and they are drawing conclusions and making decisions from this information. So that typical profile someone at work, bringing in as a data scientist, someone who’s a graduate degree or a Ph.D. in Math or Physics is often the role. They are oftentimes an analyst, they are oftentimes are programmer, so they actually write code in the fly and come up with solutions, and the scientist all in one so they can actually apply math and heuristics to real problems. They have the ability to do Applied Math. I took a lot of Math and I was always annoyed by the fact that I couldn't take majority of those algorithms and figure out what their used for. And so I would you know, I would probably drive my professor nuts to say what do you do to this? What is this for? Where do you use this algorithm for? And sometimes they didn't know, and sometimes they did. But I was always looking for ways to actually apply the math and make it useful and so that's a different breed of person, it's not everybody can do this. It's actually a real hard to find skill. The other thing is they don't wear a crash helmet, so they can actually walk and talk and speak to an executive. So some high level companies are now having side by side with the CEO and CXO team, they are having data scientist who work on the side and solve problems. So they are kind of a think-tank that are doing things to quickly figure out how to solve a particular issue.

The take-away is the reason you need data scientist, as why can't you just software to this as the data is never perfect and it's never going to be a 100%, right. If you look through your data, the history on it you can always derive a prediction from it because the history may not be accurate and they have been improperly recorded. It may have lost some of it along the way. There may be noise in it like, for example, if you are getting data for censors, there may be messages to it. So you need a data scientist who can look at it and help associate with it.

A couple of case studies, so to say what are some examples. Obviously Clockwork does this with assets, we are looking, we are making predictions on the assets rolling up all the history associated with this. But we are also looking at the failure rates of particular components on the rate and we are working with the engineers and we are working with the people on the field and interviewing them to say okay, how does the stuff really fail? What really happens out in the field? Stuff that you guys know inside and out and you are talking to customers and you are looking at this happen overtime. You tend to know sometimes more by just the history of speaking to the customers sometimes than what you can actually derive from the data. So through that, we basically build a model and example is, this is a drill shift fleet where you can actually see, you know the top line is basically, you are looking at a comparison of fleet availability to the cost of spares. And there's actually an inflection point in which you get to the point where putting more spares or more components is not really going to increase your fleet availability anymore. So you can that there's, that there's a curve in which the two inflect. That end, by taking the historical information and predicting into the future. The net result was a $2 billion in additional revenue. So by keeping this fleet up and running, it's going to generate that much revenue. That's a lot of money, right. So you think of ways, okay what is, why is this a big deal? Well, you just took a lot of data, you took a lot of history, you applied some analytics to it, you applied a data scientist to it. You did some interviews, you build a model, and then from that model you'd generate some forecast and it's significant. So based upon the 15 year life cycle, the actual shift.

So another example in this particular case study on battle damage repairs, so one of the difficult things is the way the assets are used. It could be different, like these particulars assets were used by the marines and the, you know, in this particular case there's kits that were derived. You say okay, what are some kits that can be built to actually maintain these vehicles and have less, you know, less visits into the shop and increase the actual uptime. The problem is these vehicles are used in the desert. Sometimes they are used in just normal uses on the road. They are using in high temperature, they are using freezing cold temperature same vehicles and then they are put on planes and moves so the vibration profile on them is different. So we actually went through a bunch of different studies where we looked at all different scenarios, we looked at the actual configuration of each of the individual assets. And so think about that amount of data that we had actually go through to generate these predictions. But the end result in optimized kit saved over 24 … $26.4 million in one year all right. We also have another case study for the army … we are just … we can look at their ground fleet alone and save more $7 billion by optimizing and using predictive analytics for their ground fleet. So that's not, that's not even in the entire army fleet. So there's pretty, there's some real nuggets in the data that you currently own today.

So conclusion, so I will just set a simple message. That read of a simple message is start a study of this area, start to understand it, start to figure out why it's important. Think about the data you have, you have a lot of history you know a maintenance history out in the field. You have parts usage history; you have information in your mechanics heads that you know field service technicians. You've got information that your customers currently have. So think about ways to put that together. Come up with a predictive, a big data, predictive analytic strategy to derive the solutions and then start to look at tools to help you do this as you go through and try to figure out ways to generate value. But do it value based, this isn't about technology, this isn't, this is about applied technologies taking real technology in solving a problem. So figure out ways where you can actually come up with a value proposition to say there's real dollars here, there's real savings and then lot of technologies support that going forward. So I guess real quick