 And now I'm going to introduce, we're going to have a little panel discussion and so I'd like to introduce Ashok Balani who's a senior vice president at Schlumberger and Patti Papi who's the CEO of CMS Energy based in Michigan. Thank you Jeff. So Patti and I both agreed that for us to be on the stage with Jeff Dean was to get some bragging rights. So thank you Jeff. Take lots of pictures and post them on social media. So Patti runs one of the the largest utility in Michigan and they provide electricity and gas to the state of Michigan and so she'll represent the power and the gas side of energy and I'll chip in with my background in oil and gas. I come from Schlumberger and between the two of us we are the ones who are going to get transformed by all the work that Jeff and his people do. So we'll listen to how he's going to transform us. So with that. And two proud Stanford alums. That's true too by the way. We are both Stanford and the brain comes from somewhere else. Minnesota. Go for it. That's why it's a big game. So let's start off. This is a digital session. So let's start off with Patti. Could you tell us a little bit more about the practical aspects of how data and some of this new learning is trending or beginning to transform how we think about efficiency and utilities. Yeah. You bet. We are so excited about what's happening and I think I have to give a little bit of precursor of what drives our thinking. I'm an industrial engineer by training and grew up. My first part of my career was in the automotive industry implementing lean and quality systems and elimination of waste in automotive. And then I moved to energy. And so once you learn to see waste, you can't teach yourself to ignore it. And as I look at the power and our load distribution across the calendar year in Michigan, we see this dramatic peak in the summer months in Michigan. Our demand is almost doubled for a couple hours a year and because of residential air conditioning. And so that system then is the entire system, not just the power generation, but the power generation, the substations, the distribution system, the entire system is designed to serve that peak load. And what we think is extraordinarily exciting is because right now, today, we are retiring our coal. In fact, we've retired a gigawatt of coal generation more than any other investor owned utility in the nation. And we are, we publicly announced this year the remaining coal retirement dates and we're replacing it. Right now we have the choice of replacing that, you know, megawatt for megawatt or actually shaving the peak. And so the idea that we can actually, for the first time, truly shape the demand curve in conjunction with new distributed supply resources, it's actually happening. And you might not expect it to be happening in Michigan, but it's happening in Michigan. And we're so proud to be leading the industry in this way. But the amount of data and the data analytics and the way to optimize the use of energy with customers is completely new ballgame for us. We're great at running big stuff. We're really good at that. Now to figure out how to have a partnership with a customer and allow them for them to allow us to to manage their heating and cooling and to optimize the whole system and to manage the grid is like this great operations research project that's actually going to come to fruition in Michigan. So we're very excited about it. Maybe we could branch off into talking about what are the challenges you have in terms of getting people who can do this new kind of work and maybe you could tell us, Jeff, after that, what you're doing to make this easier for for companies like us to actually absorb as a technology. Yeah, so obviously that's a huge challenge for us. Our engineers, our system planners were trained in a particular school of thought and, you know, their livelihoods have been determined on making sure that on that peak day, we have enough resource and their, you know, professional reputation are based on being able to deliver in that moment. So for them to turn it over to a bunch of computer science guys feels very terrifying. It seems like, you know, fantasy land. So getting our system planners trained in data analytics and doing the modeling and being able to trust machine learning and AI is a huge learning curve for us. And I would say that probably is the biggest challenge that we have. Yeah, I mean, I think, obviously, as machine learning is now apparently to many people a really important thing that has actually created an excess of demand for these kinds of skills. That is, you know, even computer scientists, you know, who graduated maybe 10 or 15 years ago didn't necessarily take a machine learning course. But every machine learning sort of computer scientists student who study now will almost certainly have a machine learning course. So in some sense, academia will produce more people trained in these skills than than it has in the past, because demand is much higher. But at the same time, you know, even at Google, many of our engineers weren't familiar with machine learning. And it's really important that I think every person doing computing today have some understanding of what are the possibilities of machine learning? What are the limitations? What can't it do? What are the pitfalls? And so we've actually created internally a crash course for our own engineering staff that about 22,000 or something like that of our engineers have gone through. And we've actually produced that course externally, so we've made that content available. You can search for the Google machine learning crash course. It's the same course that we offer internally for our own engineers. And I think it's really important that we realize that, first, people should have a better understanding of it than they do, and that is partly a sort of on-the-job training kind of thing. Second, we also need to make the tools easier to use, right? Like, you shouldn't need a deep level of PhD level understanding of machine learning to solve machine learning problems. And that's partly why we're creating open-source tools like TensorFlow, higher-level APIs that allow people to more concisely and easily express the kinds of machine learning computations that are really practically important in really easy ways for people who don't necessarily have that deep level of research training. So maybe if I may, I'll give you a little bit of perspective on the oil and gas business. In the oil and gas business, we work down in the subsurface, so we're working at, say, 20,000 to 35,000 feet in the ground. So we never actually see our object or touch our object that we are trying to imagine or simulate or model. And so, obviously, we are used to working with data for a long period of time and relatively large amounts of data, let's say. So a company like Schlumberger or a corresponding company somewhere probably was, let's say, go back three or four years going to 40, 50 petaflops of power on CPU and GPU for high-performance computing for seismic problems or Resva simulation problems. And then over the last couple of years, we've switched completely to working with GCP at Google or with Azure and Microsoft and we do most of our problem solving in the cloud now. And in the cloud, as you solve these scientific problems which require a large amount of computing, you discover that there are many little steps in the solutions of these problems which are better run by machine learning. And so the interactivity with the computer increases significantly and you become significantly more performant. So we've had problems that we would work on, let's say, a 50,000 square kilometers subsurface imaging problem which would normally take seven or eight months to process, can be done in something like two months with a lot less interaction with the computer. So machine learning is a key component of solving those kind of problems. And then what is true is that you really do have to get new kinds of people who think completely differently about the problem than we used to before. So I understand that a lot of work and actually you talk about in your talk as well, there are some simulation problems that you actually change to big data type solutions with machine learning. Could you tell us a little bit more about that because it's very relevant to oil and gas for sure. Sure, so the chemistry, you're referring to the chemistry problem that I framed with a simulator of the chemical molecular properties. We've also seen this in a few other scientific domains. For example, one, we had a visiting faculty member from Harvard who is an earth scientist come and he had a machine learning model, a sort of traditional high performance computing model of the fault dynamics of earthquake faults and wanted to simulate kind of what happens when things shift. And he replaced the inner loop of that system with actually the world's tiniest machine learning model, like 10 neurons per layer, four layers, which is incredibly small and found that he got about 100,000 times speed up in a simulator and couldn't tell the difference in accuracy. So again, this is a pattern that we're seeing where you have some quite complicated traditional simulation code and you can actually use that and try to replace that or replace piece of it, a piece of it, the computationally intensive piece with a machine learned version of it. It's just a pattern that seems to crop up a few different places and I think will crop up even more over time. I would say that we've had a discussion with Urs from Google about the fact that you can do these kind of things and not have too much domain knowledge in the process and he gives the example of the fact that you did your deep goal without any knowledge of what the game is when it is winning against anyone. So far, I think where we are at least, we think that most of the initial networks have to be done by people who do understand the domain and who do understand how to sort of zero in on the problem. But with time as we get more and more data, I think the learning will become more performant. Yeah, I mean, I would say the best kind of way to tackle these problems is to pair someone with machine learning expertise with someone with domain expertise and get a team together that collectively you can solve this problem and individually neither person has the complete set of skills to do this. And I've seen this repeatedly in my career where I get together with people who have very complementary kinds of skills and collectively we make progress on something that none of us could do individually and that's great because you also then take a bit of that domain knowledge or other person's knowledge and it rubs off on you and you go off to your next project and now you know a little bit more than you did before. So Patty, switching to, you have this amazing experience of leading some of the thinking on electric vehicles and the mobility transformation. Could you help us understand a little bit where this is going? Yeah, you bet. Partly because I'm from Michigan and I was actually with a group of CEOs from Michigan this morning and getting them pumped up that we are still the home of the automobile industry. I know there's a dispute out here whether that's true or not. But as the co-chair of EEIs, the Edison Electric Institutes, that's the utility CEO's group, I'm the co-chair of the Electric Transportation Task Force and as a result I've had the privilege of meeting with automakers across the country and around the world and talking about their plans for electrification of transportation. And as a result of that, I do have a strong point of view that it is, just as you said, Jeff, going to happen faster than we expect and it's going to be driven, I would say, predominantly by autonomous vehicles and the desire and the consumer demand for autonomous vehicles. I've driven an electric car for a decade and I always get the best parking spot in the lot and I'm the only one who drives it, you know, so I've never... Not in Palo Alto. See, that's how it is in the rest of the country. And so the rest of us who drive electric cars realize the huge advantage of them and I'm a big fan but I never felt like just being electric was enough of a driver to move the market across the country. But when you combine it with autonomous vehicles and when, in talking with the automakers and I know Mike Abelson from General Motors will be here tomorrow, I think they're doing some extraordinary work and we have the privilege of working with them on it. But here's the conclusion that...the conclusions that I have drawn in this. Number one, it will happen faster than we think. And number two, if the societal cause to eliminate emissions from vehicles is the driving function. So autonomous will be the consumer demand but the societal good comes from eliminating the emissions from transportation. Then we should make a commitment to do that at the lowest cost possible because there is a huge existing infrastructure today to fuel vehicles. And we're saying we're going to replace that with a different fuel source, therefore entirely different infrastructure. So here's the neat thing. There are times when there is excess power and every region has different load profiles. Michigan's is very unique and it's very different than California's but with Michigan's load profile and demand for electricity we have a lot of windows where there's lots of excess power and infrastructure available to fuel those cars or trucks or buses. The only objective then that we have as the energy provider is to say let's fuel all those vehicles off-peak and as long as we're fueling them off-peak we're not going to add infrastructure to fuel them. We've got existing infrastructure to fuel them. That is the lowest cost societal transformation from internal combustion engines to electric transportation and that then reduces the unit cost of energy for everyone. And so for us that's the big societal call and so when GM publishes their triple zero, zero crashes, zero congestion and zero emissions it inspires us to say let's have zero emission vehicles fueled by zero emission power and that's the ultimate and so that's what we're really excited about bringing to market. Thank you. So you say balance of system is the most important thing to work on to get the right energy efficiency. Yeah which puts the utility in this integrator role that Isabelle from NG was talking describing that this role of integrator optimizer it is a lot of multivariate equations happening simultaneously and to make sure that there is a system integrator that can assure that the load is as smooth as possible and then the supply of that load is equally flexible. And so I do think it is a huge transformation from tradition and just let me be clear about where like the real world is. Those of us in the Midwest I guess. You know literally a year ago I had people marching through backyards to read a meter on someone's house and find out how much energy they used in the last 30 days. Now today that's not true anymore. I have smart meters fully deployed. We in fact did a cellular deployment with the first wide scale cellular deployment so I didn't even have to build additional infrastructure for those meters to communicate in a very cost efficient way. But now I went from 36 million data points in a year to 270 million data points in a day. Now I was talking to one of our summer interns, a computer engineer from Purdue. Brilliant guy. He's so excited. It's so exciting to work on this stuff. It is so much fun. But he said Patty the only problem is my computer keeps crashing. You know like you guys are just not prepared. I'm like get this kid a better computer. But this is the point that we need help from certainly the wizards out here at Stanford and Silicon Valley to help make that computing power not so demanding that we can optimize and integrate all of these new hardware and software together to deliver the lowest cost energy and cleanest. We call it clean and lean energy. And when we can have clean and lean energy we have changed the world. One thing I would say is you know obviously as you get more data you start to have these larger computational demands even to sort of do simple processing of it let alone sort of machine learning on it. And this is I think where the general transformation to using cloud based computing is really helpful because all of a sudden you don't need to provision lots and lots of hardware in your own data center or whatever or give your intern a more powerful laptop or whatever it is. You can have the intern be able to spin up 100 computers to in parallel process this data. And one thing we really optimize for in our own research environment is turn around time on experiments. Our researchers don't like to wait for three weeks for an experiment to run. You want instant gratification or at least like hour long gratification so that you can actually iterate more quickly, understand the results of that experiment, go on to the next experiment based on the results of that one. And if you have something that does that in an hour, that's a qualitatively different kind of science or engineering or research that you do than if it takes three weeks or six weeks for an experiment. Yeah, because we need real time. You know, electricity is flowing at the speed of light and we do need to be able to optimize it real time. And so we need that real time secure data because of the infrastructure that we're managing. And so it's very exciting to us to think about how quickly these technologies are coming to market and how to really optimize it for the use and benefit of the citizens that we serve. Sorry, to be clear, I was talking about the kind of offline experimentation of trying different approaches which doesn't have quite the half a second kind of environment. But the cloud-based computing environments are actually quite secure. We have hundreds of security engineers and our staff keeping things safe which is actually a pretty large investment in security that other companies that are not in that space. I'm a believer in that. I absolutely believe that. And so it's just finding out a way to help educate other people within our entity that the cloud is safe, that our regulators believe that the cyber security is robust and there's no doubt that people who specialize in it are going to be better at it than we are. And we're a big believer in third-party partnerships with us. In fact, my leadership team and I, what we say, we go back to school a couple times a year to get exposure to the latest and greatest technology. So this year we came here to Stanford in February and met the team here at pre-court and as well as we met with the Nest team. We really liked their barbecue and beer on Wednesday night. We thought that was a very good idea. My lawyers would not let me serve beers. My engineers throw back at home. I don't know why. But we added taco Tuesday though, no beer. That's the best weekend muster. But we came out here as a team to really get exposure to some of the latest and greatest technologies. And we were very compelled by the work that Google and Nest are both doing to create this platform. And we know that you're going to figure things out a lot faster than we're going to be able to design them ourselves. And why would we as an industry demand that every utility figure out this equation by ourselves? It makes no sense. And so we've got some active partnerships that are helping us to accelerate our transformation. And we're finding it to be very powerful. So there's a couple of questions that I want to weave into the discussion here. One directly for you, Patricia, is have you considered the infrastructure required and the process for actually charging the cars off peak hours? Would that require a significant change? Yes. So in fact, we're running an active pilot right now with General Motors utilizing their on-star technology for smart charging. And we haven't found an automaker yet who's interested in having two-way power flows. So it really is just about making sure that we don't charge on peak. By having smart charging applications and running those experiments, the behavior actually is quite easy to manage. People, then the studies are very clear. We've got lots of research here in California about adoption rates and the behavior patterns of people charging off peak. What is unique is that each load center has a different peak. And so it has to be flexible for wherever the car is at a given time. But I'll tell you, here's how it works at my house. So we have time of use rates and I don't have a separate meter. It's very simple. You can tell when I'm charging my car at my house. It's actually not a complicated algorithm. And so we have the vehicle. I drive a Cadillac that's an electric Cadillac. And so I just indicate that to charge off peak. And so whenever I plug in, it automatically defaults to charging off peak. Well, every once in a while I'm in a hurry or I need to do something I charge on peak. And on my energy bill, from myself to myself, it tells me that I've charged on peak. Now my husband gives me a monthly report about this matter. And we have an ongoing debate about really does it matter. But he's quite insistent that I should be charging off peak. So it's very clear that for most people, it's an easy incentive to get people to do the behavioral change necessary to charge off peak. And it doesn't have to be everybody. That's the thing. It's always on the averages. You can optimize the system without having everybody charging off peak. And so it really is just about those signals and the behavioral implications. So given that, that takes a lot less infrastructure. Now if we're going to put in a hub of charging electric buses, of course we're going to need new transformer and maybe even a substation in a particular circuit. But generally the conversion to personal transportation, electric transportation, you can make minor investments in transformers and the system as long as you've got this behavior pattern that you've established to leverage the existing infrastructure off peak. So there's another question which we should take right here is what is the maximum amount of energy saving with machine learning as a ballpark? And I think both of you could take a shot at it. I'm challenging Jeff to a math problem. I don't think that's a good idea. I think it's a big guess problem. Well we can see that our peak is purely driven by residential air conditioning and so the idea that we could shave that peak in half with analytics and smart thermostats is completely achievable. So you could imagine anywhere from a 30 to 40 percent and I'm talking to man-response. I'm not just talking about energy efficiency. Energy efficiency, we have an annual program. It's a hundred and forty million dollar annual program. We've saved over a power plant of energy in Michigan by implementing energy efficiency. So that's sort of your base load that you're reducing but it's really demand response which is different than energy efficiency where you're actually reducing the usage on peak that it has the power to really shift the load curve and the shape of how people use energy and again it depends on the region and what the drivers are. Many regions have a winter peak but we don't, we just have the summer. So I would say 30 to 40 percent easily reduced to efficiency. I mean I'm not an expert in the energy sector but my rough understanding is that seems pretty plausible, 30 to 40 percent and I think it also depends how broadly you define machine learning. If you go back to basic science and use machine learning to for example discover new solar panel materials or something that might also be a pretty impactful thing but would obviously take longer to get into sort of a real impact setting but we should be trying all these things. I saw some work done at X a couple of years ago where I think actually there's actually something about Michigan I think but it was a number of factors less, the efficiency that you could get from all the learning aspects and all the data control aspects. So autonomous driving at that time actually. Yeah I mean the other thing about autonomous vehicles is you can sort of plan to have the right size vehicle for the thing you're trying to accomplish be it take one person from here to there be it deliver this particular thing maybe you need a very teeny vehicle of some sort and you know I think as you start to see these kinds of opportunities there's big opportunities to save energy that way that's enabled by machine learning and autonomous vehicles but isn't directly sort of an energy optimization machine learning problem. So there's a few questions on data and actually they're very topical I think when we talk about digitalization and we've been talking about how we can move to cloud use some of these new capabilities and new technologies but you have to actually move the data to cloud. So and from my personal experience in my own company and many other companies around us and how are some of our customers are trending in this direction in the oil and gas world this is not such an easy problem to solve so there are very many issues actually in fact just my own company's data we had we made a and this is just the internal operations data of my company not the data we work with for our customers it was about 56 terabytes about three years ago and moving it to cloud we were moving perabytes to cloud at that time through a fat pipe from our computer center to your cloud and that was our scientific data if you want and it took us three years this project was just finished last month to actually move from an outsourced data center where we had 300 applications which were moving along with the data and it's a tremendous job because you have different companies, different licenses different data issues, different repositories how are you going to store in a cloud store such that you can do applications on top of it so it's not such a trivial problem to actually move industrial data to cloud and suddenly start working on it so some of the questions here and I can answer for Elin Gad if you could answer for you so the question is who owns the data and how is this going to trend in the future so utilities first well this does come back to the role of the system integrator and in the absence of the utility in concert and in partnership with our customers owning that data and having customers having the right to opt in to us managing their data we would lose the opportunity to optimize the system and so if you know you're just an individual provider of some service and the data that you need for your service actually would inadvertently cause a disruption to the overall and I'm not talking about like power interruptions but just an unnecessary peak causing or it's not built into the algorithm it makes it very important that the utility plays this very important role of system optimizer and integrator to minimize the total structural cost required to deliver the power that everyone demands at the right time and for the most affordable and safe and secure manner so I do feel that there's a very important necessary role for the utility to provide in concert with our customers yeah I would just add I think you know it's obviously super important that we all kind of work together to reduce energy usage and with appropriate user consent or opt out or opt in abilities we want the ability to use collective data to make better energy efficient decisions and that seems sort of the way we should sort of there's definitely there's no kind of secret conspiracy happening that we're trying to protect the data so people use more energy like that's not the conspiracy I promise our goal is to actually have people use a lot less because of this unique moment in time where we have retired and are in the process of retiring these big central station power plants we can still have financial success without replacing those one for one in fact it's the moment in time right now where we can reduce the usage and reduce the infrastructure and still have a successful financial performance we call it our triple bottom line we don't feel that we need to make a sacrifice between serving people and affordability serving the planet and serving prosperity and able to attract capital to do this infrastructure investment so I do think it is very important that we partner with the right people to get the right information at the right time but somebody has to bring that all together and optimize the system and that's the role that I think the utility has not played in the past in fact I'll just add one more thing you know just an analogy in the past our grid our energy system looked a lot like I-80 running through Iowa have you ever driven I-80 through Iowa? let's just say it's pretty straight and flat and so it's like one you got two lanes you got a couple entrance ramps a couple exit ramps that's sort of how energy has flowed for a hundred years and now we have the opportunity with the machine learning AI data analytics this rich data set of information that never existed before and then distributed energy resources it's going to look a lot more like let's say downtown San Francisco where you've got the whole range of transportation you've got pedestrians you have autonomous vehicles you have buses you have trolley cars there's a whole range you've got the bar so you have that whole combination now that's a much more complicated quantitative exercise that requires a completely different set of skills from the team responsible for keeping the lights on on the hottest day of the year and I just couldn't be more excited about the potential that exists to do that in a clean and lean way and having partnerships with pre-court and others in Google to figure out how to make that possible sooner is our great ambition I would say one thing that happens is at a certain scale human ingenuity and sort of optimization manually can do great wonders but at some point that breaks down completely and you have to approach this as a sort of machine learning computer science optimization problem and once you do that you can dramatically increase the scale of problems that you're tackling and so you can for example treat eight million homes as things you're individually optimizing rather than like five power plants or five distribution centers so there's a few questions about oil and gas data so let me quickly take a stab at that in oil and gas there are I think we sort of classify three domains when we are working in the subsurface to characterize the subsurface or explore and there there is competitive issues of the ownership of data then there is drilling where we drill thousands of wells so there is data such that you can automate the drilling process and then there are assets which are being produced over the lifetime of the assets where there's a continuous stream of data coming in from the producing assets where you can optimize with some of the modern IoT solutions so most of the time the question is are people going to sort of share data in the oil and gas business and the answer for the most part is really no because in the case of the producing assets there is enough data streaming in that machine learning over a period of time takes over and does the right optimization when you have the right measurement and the right control set up then machine learning can help to do that we are not there yet because of the IoT functionalities between Edge and cloud are not in place yet for automated implementations but we are going down that path very rapidly in the case of drilling data it's mostly operational data so it's not competitive generally and there are really thousands of wells so we think we can go up the path of automation quite rapidly now such because there are lots of interesting new technologies of being able to get the data in a hyperconverged sense even off the cloud so that you can do optimization or automation I think in my company at least we have this cartoon of a drilling rig operating in a few years from now where there is a driller and he has a dog and the driller is there to feed the dog and the dog is there to prevent the driller from touching anything we allow dogs it works so we think that is possible and in the subsurface generally there will be some sharing but the sharing is forced by the regulatory bodies that there is enough data about subsurface available and you put on top of that the proprietary data of whatever company is operating and there is enough information there to be able to characterize and use all kinds of machine learning techniques and things to make that happen so we generally in the oil and gas are very very excited about all the possibilities but there we have a combination of high performance computing and machine learning and IoT edge cloud type implementations that bring the high level optimization you know one of the things that I look forward to is and the possibility of some sort of platform and there is providers out there that have this sort of model like a neural for example is a company who has a platform that becomes hardware agnostic and so whether it's a nest thermostat or a honeywell thermostat or an ecobee we would still have access then to the universal data of all of our customers with their willingness to opt in and so to your point of this first of all our customers willing to share the data that's a big question for us until now our customers if they live at 123 Main Street they get us that's kind of how it goes we don't have the local aggregation like exist in California and so until now if they lived at that address and they paid their bill we promise to keep the lights on and that's kind of the degree of the nature of the relationship that's as much as we needed to talk to each other and saying customer how would you like me to control your heating and cooling remotely and just trust me I'll do a good job on that that's a different relationship for one to have with a service provider now they might be more inclined to choose Google to do that because Google does such a good job on so many things but if I only have my nest thermostats operated by Google and my ecobees are somebody else and my honeywells are somebody else and I've got to somehow optimize that system that's a different challenge and so the idea that there could be a bridge that has the relationship with Google and nest who has the relationship with ecobee who has the relationship nationally internationally with honeywell then I just have to have a relationship with that person and I don't have to invent all this stuff because it's actually not my strong suit my strong suit is making sure that the lights stay on and so the idea that we could have a partnership with someone who has that bridge feels quite compelling to us as the fastest way to optimize that total system and have customers have choice and then we become hardware agnostic which I think is going to be important for us in order to have customers have all the choices that they want to have and then still have that system that's the lowest cost for everybody so we actually have also in oil and gas a slightly different version of that I think what we would like to open source is the data framework and then everybody can put their data into that framework such that the stack of possibilities on top including your TensorFlow and stuff can work very smoothly on those and make this transition possible if you like. Yeah and I'll just add that for example it's really important to us that TensorFlow be run in lots of different kinds of computing environments so in data centers it runs on windows and Linux and high phones and Android and edge devices and the same computation and can be sort of transparently mapped on to all those different things because obviously you need to run wherever machine learning wants to run and that's everywhere. So if Patty you sign a contract with Google very soon I certainly want to be part of that. That's a deal. We definitely love these guys so it's always possible. Well I'll take the occasion to thank everyone for listening and thank you both and it was a pleasure to participate in this with you. Thank you.