Okay, so you've had some time to think about Mary vs Sharmila. So which one of them would be the more valuable customer, where value is defined as likely to make more donations in the next five opportunities? In fact, I'd like a show of hands from you over here. So, if you think that Mary is the one who will be more valuable in the future, raise your hand. Great. And for those of you that think Sharmila will be the more valuable one. Let's see 'em. Okay. And any of you think it'll be a tie? Any of you think that Mary and Sharmila will be worth pretty much the same as each other? Anyone go for a tie? No one likes a tie. No one ever goes for a tie. And I mean this seriously, that if we're in this analytic space and it's our job as a consultant or a vendor firm to score the customer database to say who's the most valuable. Who's the least valuable and how everyone ranks in between. If we said these two are a tie, it would make it sound like we don't know what we're doing here, so it's interesting that we don't like ties. All right, so let's talk about Mary versus Sharmila. And it's interesting to see that I saw not all, but most of the hands go up for Mary, although for those of you who did vote for Sharmila, I can understand why. Cuz if we look back at Mary versus Sharmila, it seems pretty obvious why Sharmila would be the better customer, because if you count up the number of donations that they made, Mary made four, Sharmila made five. If you went to this nonprofit, or any company, and say hey look. Here's a customer who made four purchases, here's a customer who made five purchases. Which one do you think is more valuable to you? People are gonna give you a funny look. Well, it's kinda obvious isn't it, I'm gonna take five over the four, right? Yet, most of you didn't vote for Sharmila, most of you voted for Mary. And why is that? How can it be the case that someone who actually missed two donation opportunities might be more valuable than someone who missed only one? And of course the answer to that question, if you can say it to the camera right now. Cuz it's not only a function of how many donations did they make, or how many opportunities did they miss, but it's also very important to take into account when they made or missed those opportunities. You see with Mary, yes it's true that she missed two times, but in both cases she came back afterwards. But for Sharmila, that zero at the end there, that might mean more than just oops I forgot. It might mean that she no longer likes our organization and she doesn't want to deal with us. It might mean that she no longer has any needs at all for this particular service that we provide. It might be that she moved to some other geography. It might be that she died. I mean, I don't know, but there can be all kinds of reasons why this zero that we see for Sharmila might not just be an oops. It might be a she's gone and she's not coming back. So for those of you voted for Mary over Sharmila, you're looking at that zero at the end there and saying that zero, that single zero for Sharmila carries a lot of weight. In fact it carries more weight then the two zeros that we saw for Mary. And if you think through that logic, if you voted for Mary over Sharmila. What did you just invent in the process? Well, if you think about what we spoke about a short time ago, remember RFM? Let's put the M part aside, since we don't know how much they're spending. But what you were doing is you were making a trade-off between R and F. And you would basically say, if you voted for Mary, that recency trumps frequency. I would rather see someone who I know was with me more recently, because there's a better chance that they're still alive than someone who did more with me but there's some reasonable chance that they're gone. And that's why our forefathers in direct marketing called it RFM. Because they noticed time and time again that recency was more important than frequency, which was more important than monetary value. And basically all you folks who just voted for Mary, you just basically reinvented that wheel. And that's a great thing. It's a recognition that recency and frequency are these two key summary statistics. In some sense, I don't even care about the particular combination of the, of the zeros and the ones. If you could just tell me when the last one occurred, and how many ones there were, that's all I need to know. That's gonna be enough for me to make some trade-offs between the Marys and the Sharmilas. And if we go a step further with that, this is a great example of where we're not necessarily trusting the data. We're looking at the data as an indication of something that's going on below the surface of the data. In other words recency and frequency are nice to know, but more importantly they're indicators of the true, underlying, unobservable, propensities of people to make purchases, to drop out. To do other kinds of things. And it's those underlying unobservables that are gonna help us make better predictions. Especially over the long run. And that's exactly what I'm talking about here. Look, we look at someone like Mary and say, she just made a donation. Of course she's alive. We expect there's a pretty good chance that she might make a donation in the next period, because she was around last time. But with Sharmila, we see that zero and we ask ourself, "did she just miss one, or is she gone for good? Is she a dormant customer who might come back, or is she a lapsed customer who really is gone for good?". And this is where we can't necessarily trust the data to answer that question for us. Again it can get us some indications but we need to tell a story about the stuff that's going on below the surface of the data, in order for us to really understand and make the kinds of long run projections that we want to make. Let me give you one other example. So and I promise this is the last one. Let's compare Mary to Chris over here. And I want you to do the same kind of thing that you did with Mary and Sharmila. I want you to look at their histories. I don't want you to get caught up on any one of them. I want you to think about that general type of pattern overall and who would be more valuable. The Marys or the Chris's. So take a moment and think about that. Look over the pattern, think about what's relevant there, and tell me. Would you go for Mary, in terms of which one's likely to make more purchases in the future? Would you go for Chris? Or would you just say, eh, they're the same? So think about that for a second and I'll jump right in and answer it. So what do you extract from each of these patterns to determine how they compare with each other? And I'm going to read the minds of somebody right now. I bet that some of you looked at the Chris pattern over here and said, well since Chris was acquired, he's gone zero one one zero one one. So as I look into the future, I think Chris is gonna go zero one one zero one one. And you know what? You fell into my trap. So again, I don't necessarily trust the data. Do you really believe that Chris is, every three years is thinking. Hmm, is this the year that I skip and then I'm on for two more years? Probably not, I think that there's just much more chaos and noise when people are actually making these kinds of decisions. So I don't really wanna look at the very specific pattern, I wanna draw the appropriate summary statistics, that it's gonna tell me something, not everything, but something about what's going on below the surface. Of course the other thing that many of you did, I hope that all of you did it, is you looked at Mary and Chris, and the first thing you looked at would be recency and frequency. And you looked at Mary and Chris and said, they are the same. Now some of you might have stopped right there and said, look, Pete told me recency and frequency is all that matters and even though Mary and Chris have different patterns, their R and F is the same, so I'm gonna treat them as the same. And even though they don't like ties, in some cases if I don't have any really good information to the contrary, I'm going to go with that tie. But I bet that some of you weren't willing to stop there. And some of you said, well okay, the recency and frequency is the same so let's find a tie breaker. let's look one period back from the most recent period and what do we see? Huh, they both made a donation. We need to go into double overtime now, and what do we see? Oh, Mary made a donation, Chris didn't. I'll go with Mary. And to you I say, "shame, shame shame." Because when our forefathers in direct marketing invented RFM, it was all about recency and frequency, it wasn't about third order recency. Look, the fact is, that Mary and Chris both gave most recently. We know that they were alive last period, we know that they were alive at the period before that. So in some sense I don't really care where their third to last purchase took place. All I care about is, say it with me, recency and frequency. That's why I was joking around earlier about ties. Again, we tend not to like them, we tend to view that as a sign of weakness. That we can't do the analysis and sort these customers out. But in some cases, if we don't have really good information to do that sorting out, we might as well go with a tie, and I think that would be appropriate in the case of Mary and Chris. Okay, that's enough examples for now. I think by now we've done enough both to motivate this data set and the kinds of things we wanna do with it, as well as some of the early insights. What I wanna do next though is to talk about a very specific model. A story about what's going on below the surface of the data that's gonna let us make accurate projections about Bob and Sarah, Mary, Sharmila, Chris, and everybody else. That's coming next.
