So now I want to talk about some surprising applications of the models, and I spoke about the Zodiac applications. Yes, we can use them to figure out which kinds of customers to acquire and which ones to give the bonuses too and what other stuff to sell people. Those are great, but I don't think they come as particularly surprising. This pricing stuff happens when we take these models instead of just looking at customers buying things overtime, then eventually dropping out when we go way outside that domain. Whether it's work that we've done, whether its work from other organizations or other academic since. Wildly unrelated fields, want to talk about a few of them because I think that you'll find it motivating to see how regular these patterns are, that it's not just a marketing thing. It's not just a customer thing. It's kind of life in general. Example, number one. Doctors Without Borders, and I'm sure many of you are familiar with this wonderful, wonderful organization. Physicians who basically take assignments in places that are very risky or where they're kind of specialty, isn't widely available. So Doctors Without Borders, I was interested in understanding, given a physician's history with us, given how many assignments she's already done in different parts of the world, how much longer do we think that she's going to continue contemplating assignments with us? And how many more do we think that she will do in the future? Well man, that sounds an awful lot like, badya die, given the transactions of the donations. In this case, the volunteer assignments that you've taken on. Given how many have done frequency given the timing of the last one recency, can we make projections about how many more you'll do and for how long? And sure enough, Doctors Without Borders actually at first they ran a competition through Wharton's people analytics initiative to try to better understand this. To help them better allocate their own resources to acquire and to maintain relationships with the set of Physicians and some of the folks, some of the students. Who actually won this kind of people analytics data thon? Applied some of our by till you die models and work great. I had nothing to do with it but it was wonderful to see and then as we ran Zodiac, sure enough we actually work with Doctors Without Borders and then did an application an even larger scale with some of the other bells and whistles that I referred to before. So if that one you might understand you much. Okay, well, let's just people doing things overtime, and whether it's people making donations to a charity or buying things from an e-commerce firm. Or taking assignments and dangerous parts of the world to save lives, he said. It's just people making decisions. Well, how about some other kinds? For instance, I became of aware of some research from a colleague somewhere in the western US. I'm happy to share the paper if I can find it. If I can track it down because it was about animal tracking. So you have these folks who are basically looking at some part of the country, and they're basically well, attach some kind of tag to a mountain lion's ear, and then they'll have some kind of detector on a tree. And they want to, they know how many times the mountain lion's has gone by that tree in the past and they want to be able to project or they want to be able to say, then Mountain Lion hasn't been around for awhile. I wonder if it's God like is it God to a different area? Is it no longer interested in this tree? Is it perhaps dead? We don't know. And if this sounds kind of like body die and I yeah so. So basically other researchers have taken these exact same models. When I say exact same, I mean exact same. Maybe they don't even have all the. Bells and whistles that have spoken about before exact same models to say given how many journeys past the tree, each Mountain Lion has made. How many more journeys do we think they'll make for how long? And given that there's been a gap of a certain length, what's the probability that will oversee that Mountain Lion again, and that's kind of remarkable, right there? because you would think that it's at the same basic patterns of people choosing to do things wouldn't necessarily apply. Here's another one. How about Library books, I'm aware of some research from some folks in biblio metrics will have applied some of these by dia di models for library books being taken out. Again, this is a very practical issue because these days as space becomes more and more constrained in libraries have to make these decisions. That which books to keep in the main library versus which ones to put in a remote storage facility. They want to basically do that on the basis of how many times do we think this book is going to take get taken out in the future. So can we score? All of the books based on their likelihood of being taken out over some period of time, or the number of times you expect them to be borrowed over some longer horizon. And so sure enough we can look at book borrowing, behavior, recency and frequency, and project for each book in the same way that we would score our donors. We can score all of the books and decide which ones to keep around, which ones that we can put some place else. One of my favorite applications is on Pediatric asthma. See the way it works is when you're very young and your windpipe is very narrow, a lot of children will have the symptoms of asthma and then you got to give them some kind of treatment. An inhaler in order for them to open up their windpipe. Thing is, some children will grow out of it as the windpipes grow. They no longer need the inhaler but some will be chronic asthmatic. This is a very very important issue you can imagine lives depend on it and it's quite important for a physician. To be able to look at someones history with an inhaler and to make a projection about how much longer do we think you're going to need it? Or if it's been awhile since you've had one of those prescriptions, maybe you don't need one at all. This has been a very active researcher among a pediatric asthma experts. I'm not one of them, but I did have a conversation with one of them. Said, I got the perfect model for you. The name might be a little creepy in this context. I'd rather not call it by till you die, but maybe it's going to be something like. Prescribe until you don't need it anymore [LAUGH]. It's a little bit more pleasant and the models worked perfectly again right out of the gate BGBB. Same exact model that you've seen and the results were astonishing. Very impactful work and again very gratifying to find out that not only hey my models work, but it can actually save lives and gives give us better insights than trying to run fancy regression models that might just. Try to capture all of this through lots and lots and lots of explanatory variables. Instead of telling this surprisingly simple story two more examples for you. One of them is one that I'm working on right now as we speak crime recidivism. One of my current students is working on different kinds of programs to try to get people to not commit crimes over and over and over, or to not have to go back to jail or serve some some kind of penalty or something like that. And if you think about it, once again in a positive way. The kind of commit crimes until you stop. And if you think about the story of ARBGBB model, you might say that every period you're flipping your coin. Are you going to do something illegal or not? And at some point either you grow out of it or you move or again, you could just imagine what the circumstances might be, but you no longer do that kind of thing, so we haven't even estimated this model yet. But given my experience with this wide range of other applications, I'm actually very confident. That A, the model will work well, B it will give us some really good forecast so we can make statements about who are the folks most at risk, C it will work at least as well as regression or machine learning. Kinds of approaches were going to talk more about that as well. D, it's going to give us a really nice baseline. So as we start to introduce different kinds of policies as we start to say, hey we have a new program is going to keep people from committing crimes. We know what the baseline is. We have a sense of how many crimes we would have expected that person to commit. Or when we think the next one would have occured. So as we run this new program, we can see that it's working. So really my point here on these Badia di models transcends the idea of customer analytics and I hope that's valuable for you that it not only gives you more faith that these customer analytic models work really well for customers, but they also captured different aspects of life. Sneak preview, the last example I'm going to talk about is Customer-based corporate valuation. So much more coming up on that as well in area, I'm spending a lot of time on these days. I think you're going to find that one fascinating too. So again, gives you a range of the applications but let's dive back in and talk about some more model stuff when we come back.