So let me briefly, for the next five or so minutes, tell you about some sample business problems that why people are so exited about analytics. And then I'm gonna get into kind of the heart and meat, which are some applications of some firms today. So here's probably, as I travel around and talk to companies all over the world, kind of one of the most important problems that people want to answer is the advertising attribution problem. Or if you'd like, how much is each piece of advertising worth? Or what is the ROI of advertising? So imagine you're sitting here listening to this marketing analytics course on Coursera, and imagine you say, I have a budget of $100,000 to spend on marketing for my firm, do I wanna spend it on TV? Do I wanna spend it on radio? Do I wanna spend it on banner ads? Do I wanna spend it on search engine marketing? Well, how do you figure out the ROI of each of these? And this is a classic example of an applied area that people are spending a lot of time thinking about. Unfortunately, the state of the art out there today or not the state of the art, kind of common practice out there today is what's called last-click attribution which is wherever the last thing you did, that gets you the credit. In other words, if the last thing I did before I bought your product was hear a radio ad, well, it must have been the radio ad that did it. But if you actually think about that logic, what made you listen to that radio ad as opposed to changing the station? Or let's imagine you see a banner ad, you don't click on the banner ad and buy, but you click on the banner ad, and then you buy a day later. Or imagine you do a Google search, if any, use your favorite search engine, imagine use a Google or Bing search, and now all of a sudden, you know the set of alternatives out there. And so what's really important for attribution is understanding the entire path of things that people are seeing. So here is an example of a data set that I've worked on before. Let's actually start at the end of the date set, and let's work our way forward. Imagine you have a conversion at a site. So for example, on the example I've given you here is an automobile site. And so imagine someone has converted, meaning they've actually set up an appointment to test drive a car. Unfortunately in the United States, you can't actually buy a car online, but you can actually design your car, and you can send that for a quote to a dealer. They consider that a conversion, but imagine I knew everything about your path that you did prior to converting at the site, like I knew you went to Edmunds.com and viewed an ad, I know you went to CNN.com, and you were targeted with an ad, I know you went to Kelley's Blue Book and saw an ad. I know you went to a Google search, then I know you went to an advertiser's site, and finally, you actually converted at that site. And so at the end of the day, last-click attribution would say give all the credit to the click-through at Google, but that might not be right. All of those things, all the ads that you saw before that may be what has led you to actually go to Google, do that search, go to the advertiser, and eventually click. So again, that's one of the most important problems. Another problem I loved studying was, I actually spent about a year working on a project with ESPN, the worldwide leader in sports, and one of the things, this had to do with the 2010 World Cup, the soccer tournament, which was the largest, as people know I think, the World Cup soccer is the most broadcast and watched event in the world. And one of the events they wanted to answer through analytics was, if they launch a mobile platform, would that cannibalize their TV viewing, and would it cannibalize people streaming online, and would it cannibalize those people going to ESPN.com? Now the reason that's an important business problem, and I'm sure many of you face this problem as well, is if one channel cannibalizes the other channel, that may not be a bad thing if the other channel's more profitable, but mobile ads are much less valuable than other ads. You can charge a lot more for TV ads than you can charge for mobile ads. And if mobile ads are cannibalizing, then that's a bad thing. So this was a question that ESPN wanted to know, if they launched a mobile channel to watch the World Cup, would that cannibalize? This is the basic problem of what's called GRPs in marketing, what are called Gross Rating Points. So the idea that ESPN wanted to understand is they have two things in mind, reach, they wanna reach as many customers as possible, in this case with World Cup content. And frequency, they want those customers to watch as much content as possible. So reach is, if you like, what fraction of households are gonna watch any content, and frequency is the volume of content. You multiply those things together, you get GRPs. Now why does ESPN care about GRPs, because they charge money for advertising and the more GRPs they have, the more they can charge for advertising. So this was an analytics problems that was how does the addition of another channel affect the total number of GRPs that ESPN can serve, and here's the findings. It's one of those findings that's obvious after the fact, but maybe not obvious beforehand, which is, yes, if you launch a mobile channel for your firm, doesn't have to be ESPN, you will definitely cannibalize your other channels. But let me just say, and by the way, this could be mobile sales. If you sell more on mobile, will you sell less online? The answer's probably, but let me say two things about that. Number one, it's better for you to cannibalize your own sales than for someone else to cannibalize your sales. And secondly, what they found is, even though, and ESPN found this, and I have found this more generally true, through analytics, even if the mobile channel does partly cannibalize your other channels, the total volume actually goes up. And the way it worked for the ESPN World Cup is there was intraday cannibalization, meaning, if I spend more time on my mobile phone on a given day, I am gonna watch TV less on that day, but overall, I'm giving you an additional channel to engage with my firm. This is one of the really strong application areas of analytics today is the fact that you can measure people across multiple channels if you like. TV, mobile, web, streaming video means what matters is the total engagement. You don't worry as much about profit maximization for each channel, and imagine the same thing is true in sales. If you have online and offline sales, you shouldn't be worried about, well, if I launch online, is that gonna cannibalize my store? Well, maybe. But what you should care about is the total profitability, and it's the interplay between the two. And it's only through improved data and analytics can you assess that problem. The next one is something as you know, I'm a big golf fan, and so I spend a lot of time thinking of golf. And this is a question, how valuable are price discounts? And I want to frame my answer to this in the way of golf. But again, this is seminal question that's being asked today through analytics. So here's an example. Obviously, my name is Eric. Imaging you send a price promotion to Eric about playing golf, and let's imagine I'm a price sensitive customer, so I golf more. That's true. So one of the things you could do is, you could measure the fraction of people or the number of people you send coupons to for golf discounts, and you could see whether they play golf more. My view view is, that's exactly the wrong answer and will underestimate the effect of discounts. Now the question is why? Well, most people don't golf by themselves. So for example, if you send me a discount for golf, I might play golf more, and then I'm gonna call three of my friends up and say, hey it's a foursome, let's all play golf. Well, that initial coupon needs to get the credit, not just for me playing golf more, but for my friends playing golf more. And there was an empirical study done about, if you like, the value of your social network, and it was founded that about two-thirds of the value of discounts tends to be for the person you send it to. But about one-third of that value is by the friends that they talk to. Now, of course, how do you know that? You know that because you would have to have data on who Eric talked to once you sent them. Now where do you get that data from? Well, you can collect that information from Facebook today. There actually are firms out there, I know this is gonna sound creepy, that you can put an app on your phone that actually can measure your conversations. And of course, you know that you're signing up for this app. But the ability to measure word of mouth today gives us a much better view of the value of marketing today. And again, there may be customers that are more valuable than other customers, it's not just because they react to marketing more, but because they tell more friends. In other words, they not only internalize the marketing, they externalize it to their friends. So think about it, these problems of who should I give the marketing credit to. Do channels cannibalize each other? And what's the value of a discount? These are seminal marketing problems, that up until recently, through the collection of better data, were unanswerable questions. You could answer them in aggregate, but you could never measure them at the individual level. The last one I want to talk about is what is the ROI of Facebook. And this probably is a problem many of you are dealing with. In other words, is more likes or mentions valuable? And the way I always like to frame it is, of course. Who wouldn't want their product liked more? But I claim that's the wrong business problem. And here's why it's the wrong business problem. Because imagine instead of data set where you not only have the likes, but you also have the online ad someone saw. Imagine you also had the television ad someone saw. Imagine you also had Facebook, what they saw on Facebook, and imagine you also had purchase data. Well, if you run a regression of purchase data on Facebook likes, I guarantee you, products that have more likes on Facebook, will be bought more. But that's not the question, that's not why Facebook is worth, as of today, 70 billion. It could be 75, 80 billion tomorrow, is that most people believe that Facebook information contains data that online and television don't tell you. So that means, if I run a regression of trying to predict purchasing of online or television with also including Facebook data, that Facebook would have a significant and positive coefficient. That means even if I know how much you do online, even if I know how much television ads you see, knowing how many likes you saw on Facebook for a given product helps me predict purchase data more so than just traditional online or television advertising. That's the magic of customer attribution and customer level data. If you have this data, you can start to decompose, is it the banner ads, is it the television, is it Facebook? Over and over again from the data sets that I've analyzed shown that actually Facebook does have predictive power above and beyond just traditional advertising. But the only way to address that again, is through better data. So again Facebook and online have greater short-term effects, and this part's probably not that surprising. The part that's probably interesting to you is viewing a TV ad has longer term effects. What I mean by that is, if you see a Facebook ad and you see a like in your Facebook feed, in your news feed, you're probably gonna say oh, I'm glad my friend likes this product. But five, ten minutes later, you're probably gonna forget about it. TV, on the other hand, has a much longer effect. The part that's the most fascinating to us is that markets are efficient. What I mean by that is if a television ad costs a thousand times a Facebook ad, the economic value of a television ad is a thousand times the economic value of a Facebook ad. Meaning, basically, the amount you spend divided by the amount of impact is basically constant across these different channels. And we were definitely surprised that markets were efficient, we did not expect that to be the case.
