 Welcome back everybody, Jeff Frick here with theCUBE. We're in the Palo Alto studio talking about customer journeys. We're really excited to have our next guest on from Vivint. We have Mickey Seltzer, she's a data scientist. Welcome, Mickey. And with her also is Alvara, a senior data engineer at Vivint. First off, welcome. Thank you. So for people that aren't familiar with Vivint, what is Vivint? So we are a home security and home automation company. We've been around for 20 years. We like to make people's homes safer and smarter and we're trying to do that in a way that customers can just use their home as their normally would and we learn from what they do and make their home smarter. Okay, so I won't call you nest of security but probably a lot of people say nest of security because we always think of nest, right? It's kind of that first kind of smart home appliance that learns about what's going on. So what does that mean when you say that we learn about what you do and how you I guess move about your house, probably your patterns, what does that really mean when you talk about learning about a person in their house? Well, we have a lot of different devices in the user's house and we can tell when they come home, how they like their thermostat set. And so all of those things, sometimes you have to do that manually. Sometimes people have to come home and they set their thermostat to 72 and when they go to bed, sometimes they have to set it cooler because they want to save money when they sleep. But with Vivint, you can set all those controls to happen automatically and Vivint can detect patterns and know you tend to like your home cooler at night and you want to save money during the day because a lot of people, a lot of times people aren't home during the day and so they don't want to run their air conditioning and cool down a house that's not occupied. So we like to use all those patterns and just make your home smarter so that it knows how to save you money and how to make you safer. So that's a lot of data ingest. So what are the types of sensors, appliances, inputs that you leverage to feed the front end of that process? We have motion detectors, there's locks. There's the main panel that you use to interact with the system, the thermostat, the cameras. Smoke alarms, carbon monoxide detectors. We've got a whole host of things. And then when people put Vivint in, do they usually want to put it in because of that whole array of stuff or do they usually start with the doorbell camera or a thermostat or carbon monoxide detector? How does that engagement work and does it grow over time? Well, I think the thing that's really important about Vivint is that we're kind of a one-stop shop solution. So a lot of these products are coming out where you can get a thermostat on its own and you can get a doorbell camera on its own and you can get a security system on its own. But the good thing about Vivint is that everything is integrated and an installer will come to your house and do everything for you. And so there's no configuration that has to be done. It's kind of a, we come in, we set everything up and you're good to go. And a lot of times people will sign up just for security and then find out that we have all these great products and all these smarts that go behind it and it just makes the product that much more valuable to the customer. Right, because I would imagine the more of the pieces that you integrate, the more value you get out of the whole system. Absolutely. One plus one makes three type of scenario. And then what's the business model? Do they buy the gear, kind of a classic security? You buy the gear and then you have some type of monthly subscription for the service or how does the business model work? So right now we are moving more towards a you buy everything up front and then you just pay a monitoring fee going forward. So you will own all of your equipment. Okay, great. So that's on the data collection side. Now you guys are pulling this back in. You both are data scientists, data engineers. So what are some of the unique challenges you have pulling all this desperate data? I guess the good news is it's all coming from your own systems, right? Or are you pulling data from other systems as well? It's a lot of the sensor data that we have and I think a lot of the challenge on that is understanding the data, how it behaves and creating the metrics out of billions and billions of rows of data for all the customers that we have. So that's one of our challenges. And we do have other sources from CRM data sources to NPS and other systems that we use that we combine with all of our data from the sensors just to get a better view of the customer and understand it better. Okay, what's NPS? You said NPS. Net promoter score. Net promoter score. Net promoter score, okay. And then do you use other external stuff like the weather and, you know, I'd imagine there's other kind of external factors, public data sets that impact whether you turn the furnace up or down. Yeah, absolutely. We have a whole host of data sources that we use in order to power the smarts behind our products and weather is absolutely right. That's one of them. We also need information on people's homes in order to figure out how long it's going to take to heat or cool their house. Because somebody who lives in maybe a condo it's going to take shorter amount of time to heat up their house than somebody that lives in a 3000 square foot house. Okay, so then you guys get the data. You can analyze the data. You're both smart people. You both are data scientists. How do you package that up in a way for the consumer? So I would imagine the consumer interface clearly doesn't have billions of rows of data and doesn't incorporate that. So how have you guys kind of done, I don't want to say dumbed it down, but dumbed it down to the consumer so they've got a much, I guess, easier engagement with the system? I think we basically work with each business, a person that and from their request we start working with them, understand what they want to measure. And usually as with big data happens you kind of create a story with metrics for them. So we start with that. It's mostly on a request basis. And we have some automations just to keep track of some metrics that we like to keep historical measurements. But it's mostly we talk with the business people that to see what they want to track and kind of create our own story with the data that we have. Okay, and then I would imagine over time the objective would be for the system to take over a lot more of the control without engagement with the consumer in their home. Ultimately you want to learn what they do and start adapting your patterns to how they act so that their direct engagement with the system decreases over time. Yeah, so that's the ultimate goal is that we can infer all of these data points without having to confirm with the customer that yes I'm not home or yes I do want my home would be cooler. So that is something that we're working towards. So you've been at it for a while, 20 years the company's been around. That's pretty amazing. How have the challenges changed over that course of time? Are you looking at things differently? Are you pulling in more data sources or has it changed very much in the last 20 years or have you just added more to the portfolio I guess which adds more data input which is probably a good thing. Well, the journey that we've been on really started in about 2014 when we launched our own platform for security and home automation. Because at that point that's when we started getting the whole fire hose of data. Okay. And so at that point, that was the beginning of our data journey. And when that happened we kind of had to harness all of that data and figure out what do people want to know? Like what does our business need to know about how people are using the system? Right. And so at the very beginning it was simpler questions but now that we've kind of evolved more we can answer kind of the more complex questions that don't necessarily have straightforward answers. So it's kind of evolved from 2014 when we were able to get all of that rich data from the platform and it's evolved to now where we can use all of that data to inform the smarts for our products. And I love that we said that it's not necessarily, it's not necessarily an answer, right? It's very nuanced, right? Everything's got some type of a score variable or some type of a trade off. Have you created your own kind of scoring and trade off tools internally to help make kind of those value decisions? Yeah, so it's really all driven by context. So a lot of our data without any context, it doesn't matter, it doesn't provide any use. We're in a unique situation where we define our own success metrics. So a lot of times we'll monitor things like how, what percentage of the time is a camera connected to the internet? Because if it's not connected to the internet, then you can't view it from your phone or from your computer. So, I mean- Let's have a tight relationship with Comcast, hopefully. We won't go there. Yeah. Okay, so there's that and then again, how much of that stuff do you display back to the customer? How much control do they have? How much control do they want? Those are all kind of squishy decisions as well. All right, so you're here on behalf of Datamir, so you chose them. So what was it that attracted you to the Datamir solution? I think it's the fact that just interacting with your big data is way simpler than going to either, even if it's an SQL environment like Hive, it takes a longer process to get your data out and it's more visual, so you're seeing the transformations that you're doing in there. And I think it allows people with a more analytical skill set to get into the data and go through the whole journey of from knowing the data from almost raw to getting their own metrics, which I think it adds value for the end product or end metrics or reports. So more value for the people who have the knowledge and the data science shops. And how many kind of hardcore data scientists do you have in your team? On our team, I think we have about five or six hardcore data scientists. We're kind of split into two different teams. One team does real-time streaming analytics and our team does kind of more batch analytics. So we're all using a whole host of different machine learning and data science techniques. But on the batch side, we use Datamir a lot to be able to transform and pull insights out of that raw data that would be really difficult otherwise. And then what about for the people that aren't in your core team, that aren't the hardcore data scientists? What's been the impact of Datamir in this type of the tool to enable them to see the data, play with the data, create reports, ask for more specific data? What's been the impact for them to be able to actually engage with this data without being a data scientist per se? They can go into Datamir and get answers quicker than like I mentioned, just writing something that will take longer time. And also we also feed data to them because we have more access to historical data and aggregations like probabilities and those type of metrics we can create for them and they can utilize that in their more real-time environments and use or probabilities metrics for reacting or, I forget the word. Predicting. Predicting actions that the customer are going to take. Right, right. And I wonder if you could speak a little bit about how the two groups work together between the batch and the real-time. Because a lot of time about real-time it's the hot, sexy topic right now. But the two go hand in hand, right? They're not either or. So how do you see the relationship between the two groups working? How do you leverage each other? Kind of what's the business benefit that you deliver versus the real-time people? How does that work out? So when you're doing real-time and streaming analytics you really need to have your, you need to have your analytics based in something that's already happened. So we inform our real-time analytics by looking at past behaviors and that helps us develop methodologies that'll be more, be able to go really quick in real-time. So using past insights to inform our real-time analytics is really important to us. Which is a big part of the ML piece, right? The machine learning. You build a model based on the past. You take the data that's streaming and now make the adjustment to continue to modify. And I'm just curious to get your take on the evolution of machine learning and artificial intelligence and how you guys are leveraging that to get more value out of the data, out of your platform, deliver more value to your customer. One of those interesting little examples I always joke with people. They think of these big things and they're like, well how about when, you know, Google, take, read your email and put your flight information on your calendar. I think that's pretty cool. That's a pretty cool application. I mean, are there some kind of cool little ones that you can highlight that may not seem that big to the outside world, but in fact, they're really high-value things. Well, I think one of the biggest challenges for Vivint is something simple like knowing whether there's somebody home. So occupancy has been a big challenge for us because we have all these sensors and we can easily tell when somebody's home because they'll have a motion detector and we'll be able to see that there's somebody moving around the house. However, knowing that somebody is not home is a lot, is a bigger challenge because the lack of, you know, motion in the house doesn't mean that somebody's not home. They could be taking a nap. They could be in a room that doesn't have a motion sensor. And so using machine learning algorithms and data science to figure those problems out has been, it's been really interesting and it seems like it is a relatively simple problem but when you break it down it gets a little more complicated. Check their Instagram feed probably to get a good starting point. Or if the dog is running around setting off the motion sensors, I'd imagine it's another interesting challenge. That's also a big challenge. All right, good. So as you look forward to 2018, I can't believe this year's already over. What are some of your priorities? What are some of the things that you're working on and if we were to sit down a year from now what would we be talking about? I think create something that is more approachable as in people can get their own value from it rather than doing one-off time requests as when we're moving from on our data journey. Right, so basically democratizing the data, democratizing the tools, letting more people engage with it to get their own solutions. Yeah, because like Mickey said, the data that we're getting, it wasn't available to us until like 2014. So people are just realizing that we have this amount of data and first the questions come and they're kind of specific and eventually you start getting similar requests to the point that to speed development on other reports, we want to be able to provide some of the more important metrics that we have received in the past years to a more automated way so that we can keep track of them historically and for people that need to know those metrics. And Mickey? Yeah, as Raul said, we're trying to move more towards self-service in the past since our data is constantly evolving. There are not many people who know the context and the nuance of all of our data. So it's been really important for us to work with our business stakeholders so that we know that they're getting the right data with the right context. And so moving towards having them be able to pull their own data is a really big opportunity for us. With that context overlay, so they know what they're actually looking at. People so under report the importance of context to anything, without the context. Is it big? Is it small? Well, what are we comparing it to? Exactly. All right, well, Mickey and Raul, thanks for taking a few minutes of your time and sharing your story. Fascinating, we'll look into more about Vivint and I guess you just need to get more motion sensors around the house, on the bed, keep an eye on that Instagram account or are they taking pictures down the way? Let's not be creeped out. Let's not be creeped out. Well, that's why I look at a great line, right? Data science done great is magic. Data science not done well is creepy. So there's a fine line. So thanks again for sharing your story. Really appreciate it. Thanks for having us. And I'm Jeff Frick and you're watching theCUBE. Thanks for tuning in. We'll catch you next time. Thanks for watching.