So you might see this section of the videos as a little bit of a downer, right? I'm telling you, we can only expect so much accuracy from these models. Usually if we can explain 30-50% of variation in whether people leave or how they're performing. We're doing really well. Yeah, sensible question is are we wasting our time with spending all this time wondering about using ai unemployment, but actually when we do so the results are great. I don't think we are wasting our time. My favorite illustration for this comes from Moneyball. I don't know how many of you have either read the book, Moneyball or or seen the movie. It's kind of foundational text for a lot of people analytics work kind of like our Iliad if you're feeling really pretentious, right. It's going to get really one of the first books that thought about how we apply analytics and various different modeling. To understand people, kind of the basic the basis of the book and the movie is about a team called the Oakland A's. They're struggling and they're trying to figure out who were the right people to hire and they're applying analytics to try and figure out which team members to bring in rather than kind of the usual scouting techniques. Why am I bringing all this up? There's this great moment in the movie where Brad Pitt leans back and says, gentleman, we are now card counters at the blackjack table and we're going to turn the odds at the casino, let's be honest. It sounded a lot better when Brad Pitt said it. But, I think the basic point is important. So I don't know how many of you are into card counting, know what it does. It's basically a way of skewing the odds when playing blackjack, which is a card game that's frequently played a casino. Again, when you're betting on cards, what you really, really want to know is what cards you're going to get and what cards the dealer is going to get. I mean, if you actually know that, then you can bet perfectly, because you can put in big bets when your card is going to be better than the dealers and not bet when your cards are going to get worse, right? Now, you can never know that, right? But it turns out because they're, you know, there are only a limited number of cards that are in play at any time if you know what cards have been already dealt. So which cards have been used recently? Well, those cards aren't coming up again, and so that gives you a sense of what cards are left. And so what kind of cards are more likely to be received by either you or the dealer in future. And it turns out that, you know, sometimes when there are certain cards that are more common in the debt coming up, you're more likely to win. And sometimes when other cards are more common in the debt coming up, you're more likely to lose. And so by knowing that you don't know exactly what cards you're going to get, you know what kind of cards you are more likely to get. Okay, and so that can inform your betting, right? So if you implement card counting effectively, you're still going to lose plenty of hands, are going to be plenty of times when you bet. And it turns out your cards are worse than the dealers. But the point is that you will win more frequently than you will lose. Its about skewing the odds in your favor. And if you do so consistently the odds of winning go up enough if you do this successfully you should actually expect to make money on average rather than losing it. Hence casinos tend to be fairly negative when you're actually doing this. They're not big fans, right? But the point of this is you don't need to be right about every card. You just need to be right about what cards are coming up more frequently than you're wrong. And it's a similar idea when it comes to implementing AI in HR decision making right, the models will not be perfect. There'll be a bunch of people who we expect to perform well. They won't perform well. There'll be a bunch of people who we think will perform badly and therefore will reject them and they'll turn out to have been great highest, right? Okay. But we actually don't need the system to be perfect. We just need them to be better than the alternatives. And so again, if we think about hiring, we know that left unaided, there are a lot of people who we hired as we expect them to perform well and they don't and we've got to assume that some of the people we reject would also turned out to have been great highest. So, the question we need to ask ourselves is not is the AI model perfect question we want to be asking ourselves as is it better than the alternatives? Right. And so, in all of these applications, what we're really trying to do is just skew the odds of making the right decision in our favor. Okay. We know that when we use human judgment in making these decisions is pretty flawed, in plenty of applications when we build models, they make better predictions than we do unaided. When we make better predictions, yeah, a bunch of our decisions will turn out to be wrong, but fewer of them will turn out to be wrong than they would have done when we're making these decisions unaided. Okay. And so if we can skew the odds of making the right decision in our favor, we will end up being more successful. And that's why it turns out often to be well worth investing in building these models and applying them to how we manage people.
