All right, enough of this abstract talk about period one, period two, period three. I wanna get real. I wanna talk about a real data set, with real customers, real behavior, and real decisions on the part of the organization that actually provided the data to us. I'm gonna actually start with a non-profit context. And you might find that appealing, you might not. In some sense it really doesn't matter. Let me just give you a few details about it, and then when we look at the data set as whole you'll see that it's applicable to a broad variety of different settings. Profit, non profit, domestic, international, B to B, B to C. I think everyone will find something appealing about this basic kind of data. I want to talk about a specific non profit organization that's looking at its donors over time. So with so many companies, they're gonna take their customers, their donors and we're gonna find a whole bunch of them who were acquired at the same time. I think this is actually a very important lesson beyond the analytics, beyond the predictive models. Very important for companies to be able to look at their customers on a cohort basis. Let's identify all customers who we either acquired at the same time, we acquired them through the same campaign or maybe through the same channel or maybe they all made the same first purchase with us. Well, in this case it's gonna be a time-based cohort. We're gonna look at a bunch of donors who made their first donation to this non profit organization in 1995. So, in other words, with knew nothing about these people. They didn't exist to us at all until 1995, they made a donation. Like, hey, we have a new donor here. And so we're gonna follow these folks for the six years after 1995 and then we're gonna wanna make predictions about them for the five years after that. So we're talking about a very long data set. Again, the kind of data that companies might not have had available to them at all in the old days. But today, because of digital everything, it's becoming increasingly easy to create these long histories for customers, and more important than ever, to be able to make statements about who's gonna do what, when in the future. So you can picture the data structure right over here. We can see we have 11,104 customers, 11,104 rows in a spreadsheet and then we have these bunch of columns where each columns represents a year and we're simply asking, did they or didn't they make a donation? For right now. We don't care how many donations they might have made, although rarely do people make more than one a year, and we don't even care about the M part of RFM or what the size of those donations were. All we care about is did they donate or not. I'm gonna give a 1 if they did make a donation and a 0 if they didn't. So if you look at the data structure here, first you see this whole column of ones. Because that's telling us that all of these customers were acquired in 1995. And now you see a bunch of zeroes and ones to the right of that, basically saying did they or didn't they make a donation. Over each of the next six years after being acquired. So, let's first stare at that data and really get to know it better, and then we're gonna wanna make some statements about who's likely to do what in the future. Before we even do that, though, just look at this data set. There's a bunch of customers doing things or not doing things over time. So in this particular case, it's donations. But you can imagine this very same data structure for, oh I don't know, a hotel chain. So we have a bunch of people in our loyalty program, did they stay in one of our properties or not in a given month? It might be for a credit card firm. Did these people have revolving interest charges or not on a month-to-month basis? It could be for a mobile gaming application. Did people play the game or not on a day-to-day basis? So I hope you can see that this data structure is very general. It's very practical. I don't really care what the ones and zeros represent. Whether they're purchases, whether they're posting social media, staying in hotels, interest charges, whatever it might be and I don't even care what the time period is. So here we're focusing on a year as the period, but it absolutely can be a month, a week, a day, even a second depending on what the situation is. So before I go on I just want you to appreciate just the, the generality of this kind of data structure. It's data that many of you have contributed to as consumers, as companies try to track and anticipate what you might be doing. And also on the business side, you might be working with data structures like this. To help you become just a little bit more comfortable with it, let me just pick out a few examples here. And I'd like you to actually make a prediction for me based on what limited information you have, what you see in front of you on this slide. I'd like you make some predictions for certain kinds of customers. So for instance, let's start with Bob over here. So look at Bob, you can see that after Bob was acquired in 1995, and again, we're not gonna predict the 1995 purchase cuz we know nothing about Bob or anyone else in this data set until after that purchase occurred. Look at Bob, out of six opportunities, he donated all six times. He went six out of six. So let's make statements about Bob for the next five years. So in fact I'd like you to really write that down. What's your best guess about how many donations Bob will make out of the next five opportunities. And if your answer is I don't know, I don't have any idea, then you're absolutely right. Because, to be honest with you, I don't necessarily care about Bob. I don't necessarily care about this one customer. What I really care about in my ability to make good predictions about, wouldn't be Bob per se, it would be the Bobs. Tell me about all the customers who share these same characteristics. And so while it might be hard for me to make statements about any one of them, once I start grouping them together, and in this data set, as you'll see, there are about 1,200 Bobs, I wanna be able to make statements about the Bobs on average. So if we think about the Bobs, who have gone six out of six since being acquired, how many donations on average will they make out of the next five opportunities. So any one of them will make zero, one, two, three, four, or five donations. When we look across all of them, some will make zero, some will make five, some will make two or three, what's the average across all of them? I want you to write that number down. So that's Bob, and now let's look at Sarah over here. Since being acquired in 1995, what has Sarah done? Nothing! She's the opposite of Bob. She's gone O for six, she's done absolutely nothing. So I want you to make a prediction about Sarah, or again, the Sarahs, all of these customers who have done nothing since being acquired. Now, you might be wondering. Why should we make statements about Sarah? She's done nothing for six years. Let's just get rid of those customers. They can't be worth much to us. And the fact of the matter is, any one of the Sarahs probably isn't worth much. But why is it that we're so interested in tracking Sarah? Because there's so many of them. Even if you look at this slide right over here, look at how many Sarahs you see with the zero since being acquired. Turns out that the Sarah, that is to say customers who do nothing after being acquired over some observable period of time, are very, very common. In this data set, they represent about 33% of the customers. I see in many data sets they are often over 50%. So while any one of those Sarahs might not be worth much, collectively to the extent that there's any value in them at all, when we add all that up, the Sarahs can actually still be a fairly large chunk of value for the company in the future. Especially when we look over the long run. So I want you to write down your prediction for Sarah, or for the Sarahs. How many donations on average will they make over the next five opportunities? And I have one more quiz question for you right now. I want you to compare Mary and Sharmila. Okay, I'm not even going to say anything about them. You can see the numbers for yourself right over here. So I want you look at Mary and Sharmila and in fact I want you to answer two questions. Question one. Which of them will be worth more, on average, over the next five years? And question two. By how much? So take a minute, look at the Marys, look at the Sharmilas, and think what it is that makes one of them more valuable than the other. And I want you to write that down, and make that prediction about which one's gonna be more valuable, and by how much. We'll take a break and we'll come on back and talk about some of the logic that you might have been using to make that decision between Mary and Sharmila.
