I'm so glad to have Paul Hunter, who's the head of operations for Dunnhumby with us today. He's going to talk to us about shopper cards and how they use that data to better provide experiences and value for customers. So glad to have you here, Paul. Glad you can be with us here today. Thanks for having me, Raj. I look forward to it. So let me get started straight away. Can you describe to us in very basic terms the kind of data that marketers are using from shopper cards? Let me start by talking about what - what data is captured by a single transaction. So imagine if you will, if you have a - a loyalty card, or a shopper card at maybe a hardware store or a grocery store. Any number of retailers have loyalty cards; in fact, it's estimated that over 85 percent of U.S. retailers have some form of a loyalty card. But imagine you walk into a store and you buy a basket of goods. The type of data that's captured is - is quite exhaustive, but it will capture at the Tlog - and Tlog is short for transaction log - but for each Tlog this captured, it will capture your customer number, your loyalty card number; it will capture the product sold and how it captures the product sold as its - the barcode that is on the products that you have purchased. It will capture the number of units sold by barcodes, so you may have bought maybe six yogurts, or six of a particular unit, it will capture that. It will capture the actual price paid. Many loyalty cards will charge two different prices. If you don't have a loyalty card, it might be $2, where if you have a loyalty card, it's $1.70. So the Tlog, the computer transaction log, is capturing also the actual price paid. And depending on the retailer system - and some are quite exhaustive - they will capture the use of coupons. So maybe use a coupon on that particular transaction. It will - and it - and depending if you're, let's say, a grocery store or a drug store that has loyalty card, it will capture if there's government subsidies that is associated with the transaction. So you might be in the WIC program - Women, Infant and Children program - that will tell us that. It will tell us if we're using food stamps. It will tell us how you paid for your transaction; did you use cash, credit, debit? What type of credit card did you use? It will tell us in - in a summary what's the total number of items that were in the basket. Was it 20, 30, 80 different items that were in that basket? And the - the checkout; the Tlog will capture the actual checkout lane that you used and we can tell you down to what checkout lane did you use. Did you use self-scanning? Was it a regular checkout lane? Was it a click-and-collect type of transaction? And depending on the retailer, they may have captured, as part of the Tlog transaction, whether the items that you have purchased in your basket, were they merchandised? So we can append did it come from a secondary display location or did it - was it not merchandised and was it in its regular stocking location. And then finally, at the end of the day - at the end of the day, what it also captured is the time stamp of the transaction. So lots of different variables from the time stamp to the units sold to how you paid is captured in a typical Tlog transaction. Wow. That's a lot of data and a lot of information that we're collecting from customers on the shopper card. What kind of regression models are people are using and how prevalent is this use of regression models in the industry to gain insights from the data that's collected from the shopper cards? You know, Raj, you are right. It is a lot of data. In fact, I'd like to think it's more data than you could shake a stick at. But, you know, with the advent of computers, storage has become quite cheap. The power of databases have allowed us to manage this data and make sense of it. With respect to your question about regression, it's extremely prevalent. We use it daily and we use all different types of regression. But typically, we - we will not necessarily run regression models in totality; so across the entire store. We do and can do that, but we also run regression models by customer types. So if we recall the type of data we have, such as, you know, government subsidies or did people use coupons, it's quite easy for us to begin to segment groups of customers in terms of, let's say, their price sensitivity. And so if we are looking at making an investment to attract, you know, more price-sensitive customers, instead of doing a regression model on the total store, we can do a regression model on just price-sensitive customers and look at the specific items and categories that would allow us to pull levers; invest in price, reduce price in certain items that would bring more price-sensitive customers into the store. In that particular incidence where we're looking at, let's say, pricing models or advertising models, we will - we will typically run multiple regression or what we would call also Bayesian regression models where we're looking at modeling sales and it's the sales of maybe a customer segment as a function of what was the price, was the merchandising conditions present? We're going to try and control for as many factors as we can so we can truly isolate the price elasticity with that multiple regression model. Other types of regression models that we will use is if it comes time to maybe not just invest in price, but maybe send these households - because we have their e-mail address or their mailing address, we know where they - how to reach them - we might even be able to reach them at till. Many different ways to communicate with customers with all this data, we would now like to maybe activate against them. And what we mean by activate is send them relevant offers. And you know, essentially people are using the term coupons; we like to think of it as almost like personalized pricing. So what's the best offer to send to an individual household to bring them into our stores and typically to determine what offer's best? What we're looking at is the probability that they would redeem it. The probability that they would come into our stores. And in that case, we're going to use logistic regression. So logistic regression allows us to model the probability of the activation taking place as a function of some of those other factors that I talked about. So Bassi regression, logistic progression, we use daily. They're invaluable tools. Without them, we wouldn't be able to deliver insights and performance back to our retail partners.