The other part that I thought was very interesting about some work we had done here is what's called the traveling salesman problem. Now this is probably a problem that's not familiar to many of you. But it's an old problem that has been around since the 1950s where a salesman has to visit a bunch of different cities. So on the map of the United States I'm showing you, each of those red dots represents a different city that a salesman who covers the United States has to visit. And the traveling salesmen problem says, what's the shortest route the salesperson can take to visit all of those cities. Cuz obviously the salesperson doesn't wanna spend time traveling, they wanna spend time visiting clients. So you might say what the hell does this have to do with analytics and what does it have to do with retailing. Well, it has everything to do with it. So what we were interested in is do people who move more efficiently throughout a store, do they buy more or less? In other words, do you want efficient shoppers, or do you want people that wander more? And again, the reason back to my five point framework, think about this. This is data we never had before. So that's phase one. Number two, we're now able to create things like heat maps that allow us to explore the data, to look at what fraction of the store is being cover. Step three, now we're gonna analyze the data under the framework of a traveling salesman problem. And we're gonna look to see, are certain types of shoppers, people that are more efficient, in and out shoppers, are they more valuable to the firm or less valuable to the firm? Step four, then I'm gonna talk to you about who you should be targeting, and step five, I'm gonna show you about decision making of the firm. So this is a great example that has all five points I was talking about. So again, did you can see here on this picture, let's imagine the green circles represent the products that the person bought. The red line on this picture represents the shortest possible path the person could have taken to pick up their cart, buy those products, and check out. Now let's be clear, this is not the path any given person took. This is the shortest possible path they could have taken and we're interested in understanding are people that took shorter paths or not, more valuable to the firm. So for example, the path on the A, if you like, person on the left hand side of the picture here. This is a more efficient shopper than person B. If you'd like, person B got all wiggly and jiggly. Person B didn't just visit parts of the store where he or she bought stuff, but did a lot of other wandering around the store. Now the question is, which custom would you rather have, A or B? Obviously, before I flip to the truth on the next couple slide, each of you should be thinking, do you want to make things efficient for customers, or do you want people to do some degree of wandering. So this is what's called a triangle plot what's actually interesting here is that, you can see on the slide it says the average TSP optimality is 28%. Now, what does that mean, it means that about 75% of all movement inside a store is not required. That means people could be spending one-quarter essentially the amount of time in the store and still buy the goods that they bought. Now the question is, is that good or bad? So what's neat in this triangle plot is each of these dot represents one of about a million customers that we analyzed and you can see here that on the top we have TSP optimal. So if people were actually moving in the shortest possible path you'd see all of these dots at the top of this triangle. You don't. If the right travel deviation, what does that mean? It means, for example, in a store, imagine you should pick up the milk first and then the cheese and the bread, but you went bread, milk, cheese. That's called travel deviation. That means you went in an efficient way, but you went in the wrong order. So think about it as you have to buy products A, B, and C, you should go ABC, but you went CAB. That's order deviation. Then travel deviation has to do with how much jiggliness. So I'm walking from the milk aisle to the bread aisle, how much did I jiggle? How much did I wander off my optimal path? What you can see here is that most people aren't optimal shoppers. And this last piece is now steps four and five of the framework that I gave you. The key number is what I'm telling you in yellow. Group 4 are people that go in the wrong order. And they have the most jiggliness. And notice those people buy on average about ten items every time they go to a store where the people that are most efficient, which is group 1, only buy about half as much. So notice how I've turned new data, which is where people go in the store, I've explored that data to what parts of the store are most valuable. I've now taken that data and I've analyzed it through the traveling salesperson model where I've now looked, now I can categorize each person as, are they efficient in how much they jiggle between points A and B and do they go in the right order? And now I've taken it to the action step where I can see which customers are more valuable and it turns out, you don't wanna rush people through the store, you don't want them to be that efficient. Now the challenge is, you have to be able to identify which types of people are the more jiggly types, and which people are not. Unfortunately, this shows you a limitation in our dataset we didn't have demographics about the people. So we didn't have age, race, income, gender, urbanicity, do they live in the urban neighborhood versus rural. If you did, you could then start to say now I see who to target. Now I know what types of customers to target. Unfortunately in this case, we had good data. We had data on locations, and we had the data on purchases, but we didn't have data on demographics. There's a great example of how, without that information. I've come to a business decision. I need to find customers that wander more around the store. That's great. That's nice to know. The next step would be, of course, knowing who are those customers. What are their types? Can I identify them? Can I target them with marketing in a different way? Cuz I'd rather have more of those customer types than the other customer types. The next kind of data I'd like to tell you about, which is again kind of the future of marketing science, is eye camera data. Now you might say, eye camera data? You mean people walking around the stores with eye cameras on? Well, many of you may remember, and you could call it a failure or not, on Google Glass, that was Google's foray into the area of let's have people where glasses where everything they're looking at is being tracked. Now if you had those glasses on, you could actually track where people are looking in the store. We actually did a study of over 1,000 people where we had them put on these types of glasses. And so this was an experiment, it was a field experiment, where we actually put these glasses on people and we track what they look at. And so you can see a picture of a person on the left with this eye camera on. You can see on the right what it looks like from the eye camera. And then, the first thing we wanted to notice. And I've always wondered this question. But again without the right analytics and data, you could never answer it is, if I'm showing you, let's say, an orange juice, what's called the planogram of the orange juice, where the orange juice sits in the store, where would you rather be in the store? Would you rather be up high like six feet high at the top shelf? Would you rather be at the bottom shelf, the middle shelf? Well, through eye tracking data and analytics we can now answer this question. It actually turns out as you can see or think of this as a heat map again red hot blue cold. Turns out the optimal height for many product categories around 5 foot 6. You might say, well, why is that? Well, the average female height is around 5 foot 6 and so in many supermarkets where 80 to 90% of the shoppers are female. You wanna put it at eye height. People don't like looking down, people don't like looking up. People like looking at eye height. You may also notice in this picture the left side of the planogram seems to be a little bit hotter than the right side. And you might say, wonder why that's true. Well, we read from left to right. So we scan from left to right. So, if you have a choice between being on the right-hand side of the shelf or the left-hand side of the shelf, you wanna be on the left-hand side of the shelf, because people, when they walk up to an object, we scan from left to right cuz we read from left to right. We scan at our eye level. And then sometimes we go up or down if we don't find the product we want, so kind of, eye height level on the left-hand side is the best place to be positioned in the store and many of you may be working in industries where you have to pay for shelf space, so you wanna know what the ideal shelf space is. It was through measuring one customer at a time and looking at their eye data that we were able to notice. For women it's about 5 foot 6, for men it's about 5 foot 9, its left is better than right, it's through analytics that we were able to answer this question. So the other thing that was interesting through this, and this is now to steps four and five. What's the business implication of doing this? So, what I'm now showing you a picture of is, on the X-axis here, since remember, we were able to remember what it is you looked at on the shelf, not just what you bought, on the X-axis you see three different variables. You see, did someone fixate, look at the product for no times, they never looked at it, once or twice or more. The thing we were interested in is does in store marketing work? Are products that are looked at more bought more? And what you can see here in the green line, what we did was we measured people's fixations through their eye-tracking device that was on their head, and we knew how many times you looked at a given brand, zero, once, twice, or more. And then when you left the store, we surveyed you and say, do you remember seeing that brand? Now let me say the good news from the green line. More views upward sloping, the more you look at something, the more you remember seeing it. I don't know, maybe I'm not gonna wanna know about a prize for that, but no one had ever looked at that data before, so recalls, meaning do you recall seeing the product, remember, that's valuable data because just cuz your not gonna buy that today doesn't mean your not gonna buy it in the future. More views, higher recall. How about consideration, we also asked people what brands did you consider buying? Notice again, an upward sloping curve. The more you saw a product, the more you consider buying it. So this is where things like banner ads and other stuff, just putting your name out in front of people over and over again not only raises recall, but it raises consideration. And lastly, probably the one you might care about the most is purchase. There is an upward slope in curve. It's a little bit flatter. And this is not surprising. Some people who view a product a lot don't necessarily buy it, so recall and consideration are tremendously affected by more views, choice less so. But again, this is the power of analytics because now again you can start to say hey, I see there's potential long-run value of different forms of advertising. It isn't just in choice. In fact, choice may be the last thing to come. Recall and consideration are gonna come very quickly, now that people have seen your ad, seen your commercial. Now they're much more likely to recall it and consider your product. Hopefully, if your product has good benefits, people will actually buy it. So again, this was a great study we did using data at the individual customer level measured in store. So another thing that came from this, since we also had data on what people planned to buy and unplanned, cuz remember from that first slide I showed you on the in-store data, we asked people before they went in the store what do you plan on buying. It turns out, and this is the part you can see here, 4.9 out of 8.2 this the number to me was shocking. 60% of the things people bought in stores, they went in and had no intention of buying when they entered the store. Now again, how did we answer that through, when we go back to my five point framework? We had data, we knew we wanted to measure this, so we measured intentions before they entered the store and we know what they bought in the store. It actually turned out that an also in reverse 40% of the stuff they planned to buy, they didn't actually buy. Now that's kind of a reinforcement of marketing works. In other words, you don't always buy what you intend to buy. This is a great example of how analytics can affect in store marketing. The future of shopping. Many of you may have already exhibited this at your local stores, supermarket stores etc. But if you don't know about it let me tell you what's coming soon. It's something called a media cart. So imagine walking around your store where you're putting your stuff in the basket and imagine the screen in front of you is now giving you product recommendations. Like you just put pasta in the cart and it's saying, sauce is two aisles down on the right. Or imagine you bought chips and you say okay, where do I go get the beer now, and so imagine just like Amazon does product recommendations on the web, imagine product recommendations in real time based on your physical location and what you've put in the cart that's coming. Now, my research would say I wonder if that's a good thing, since this may make people more efficient shoppers and they may not buy more, but on the other hand, it may make them come back to the store more often cuz they can be more efficient shoppers. I think the future is wide open, but you should expect to see things like media carts in the future, and helping you if you'd like navigate around stores.
