7.1 What's coming next in marketing analytics? - Video Tutorials & Practice Problems
Video duration:
17m
Play a video:
<v ->So let's take a look at what's coming next</v> in marketing analytics. So maybe the biggest trend that's happening right now is the one towards big data. So what do we mean by big data? Well, big data is actually a fairly nebulous concept. What it's really talking about is that the volume of data that we have available to us today is far greater than it was even a few years ago. But here's the thing that's funny. A few years from now, it'll be even more. So when you look back on today, you say, "Wow, that wasn't big data. Now we've really got big data." The real thing you care about is that data is increasing all the time, and there's really five V's that you care about for big data. The first is volume, so you probably figured that one out. That's the big part. Next is velocity. So what's happening is that the data is coming at you faster and faster. Where it used to be that you would get a report of something that happened in the last month, now you're getting a stream of data telling you what's happened minute to minute. Another V is variety. It's not just that the analytics are changing in their numbers, but the type of data that you're collecting starts to change. A few years ago, we weren't collecting likes or shares because those things didn't exist. As marketing changes, the nature of data also changes. Variability. The numbers seem to move up and down almost at random. And why is this happening? Well, because when you collect enough data, when you look at small samples, they're going to jump around. So you remember when we talked about how you need to be really careful of what's in your sample? You need to be sure that you've collected enough data that is statistically significant. Well, that's what will reduce your variability. If you can figure out that you're getting data that's not as variable, that's going to help you to draw better conclusions. The last one is actually the most important V in big data, and that is veracity. That's where you come in. You, the marketer. Not the data scientist, not the marketing analyst, not the web metrics person. What you need to do is understand that the data can be used for certain decisions. You need to be the one who decides if that's a good idea. So you need to use your wisdom and your understanding of marketing to decide whether the data is accurate enough for the decision you are trying to make. And that's something that computers aren't going to do anytime soon. So let's look at the progression of big data in marketing. So if we take one example, suppose we take the example of what actually appears in media. What is, quote, on the news? What is showing up in the media that we consume all the time, and that marketers spend a lot of time figuring out how to advertise on? So let's look at television as an example. In the United States in 1995, about a thousand people made the decision of what was going to be on the television news. And so if there was a story that came out about your company, you basically had days to respond. It was pretty slow back then. In fact, some of you might remember the Tylenol scare back in the '90s. It took Tylenol days to announce that they were recalling their products and taking them off the shelves. And if you remember the story, there was some nut who was actually putting poison in the bottles. And so, can you imagine a company taking days to make that decision? Nowadays, they would consider the most heartless company ever. And so that's was just a different time back in the '90s, because people expected decisions to just move more slowly. Let's move forward just 10 years to the advent of blogs. That meant that there were a hundred thousand people deciding what was newsworthy, and you were expected to respond really quickly. In hours, not days. What happened just a few years after that? Social networks. Now there are over a billion people who are deciding what's on the news. Don't think I'm right about that? Next time, check out your local news and see how many of the stories were actually shot by somebody who was just taking pictures with their camera. Citizen journalism, they call it now. And that's what's really happening now, and your response time is now minutes. Now, one of the things you might notice about this progression is it can't keep accelerating at this pace, because we run out of people and we run out of time. But this is what's happened the last few years as marketing data has exploded in volume. And so big data is driving two other trends, data science and artificial intelligence, because data is actually the fuel of artificial intelligence, and data scientists are the people who are putting together the engines that use that fuel. So when we think about big data, we think about email, and photos, and Twitter, and that's true, but what's happening nowadays is that there's omnipresent security cameras that are taking footage all the time. And you have all sorts of sensors that are on every item in the supply chain. And so the amount of data being thrown off now far dwarfs anything you could think about from email. Now, a few years ago, IBM had the statistic that said 98% of the world's data has been created in the last two years, which is kind of stunning, cause that goes all the way back to Gutenberg's Bible, when we think about stored data. But it's accelerating so fast that I think you could ask the question if someday we might say, "Well, 98% of the world's data have been created in the last two months." That's how fast big data is accelerating. So who do we need to help us to corral all of this big data? Well, that's the role for the data scientist. So what is data science you might ask? It's actually a field that pulls together several different disciplines. So computer science, math and statistics are two of them, but the third one might surprise you. The third field that pulls together is actually business knowledge. And so for you, that's knowledge about marketing. So data scientists often call this domain knowledge, and the reason they have to call it that is because they don't have it. They have to call it something, and they need an expert in whatever the particular problem is that's being solved, or they're not going to be able to help, because data scientists can't be experts in all the different domains. That's where you come in. So you need to help the data scientists understand the problem. You need to help the data scientists take a guess at what the factors might be that could solve that problem. You remember when we talked about the lead scoring algorithm? You actually have some good ideas for what the characteristics might be that should be positive or negative in that algorithm. That's a mini version of what data scientists need you to do. They are doing something called feature analysis, where the features or characteristics of the problem, they actually need help from the domain expert, you, when it's a marketing problem. They need help from the domain expert to be able to know which factors they really ought to look at, and which they don't need to. And while we're talking about data science, you might be asking yourself, so what exactly is AI? Well, AI, the standard definition is actually something a computer can do that normally requires the intelligence of a human being. But I think that's actually a goofy definition because what really you want to know is not what the definition of artificial is, but what's the definition of intelligence? And that's what makes AI so hard to nail down. But what we do know is that more and more, there are decisions that a computer can make, that it used to be only human beings could make. And the computers make them faster, and they make them often more accurately, but they sometimes make mistakes that humans wouldn't make. And so we need to focus with the data scientists on making sure that the mistakes that are made by AI are actually manageable, and we need to make sure that we are understanding what the accuracy is of any AI model, and that we're working with the data scientists to make it as accurate as possible. You might hear other terms like machine learning and deep learning. And so artificial intelligence is any technique that really allows computers to mimic human behavior and judgment. Machine learning are techniques that allow the computer to actually improve as it learns more about the problem. So as it gets more and more data, it starts to understand the problem better. And it's doing that without being explicitly programmed with a set of rules or instructions for what to do. It's basically trying to deliver a certain outcome and predict something that matches the data that you've given it already. Deep learning is something that's relatively new. And what it does is it helps you to broaden your training data. So what do we mean by training data? Well, the way that machine learning algorithms work is you actually give it a bunch of data that has the correct answers. And so it uses that to learn from those examples. Then what happens is you start giving it examples of things that hasn't seen, so that you can then get predictions from the computer. Now, what deep learning does is it broadens that training data. So for example, if you had training data that was trying to help an algorithm to figure out whether social media comments were positive or negative about hotels, well, the training data that you give it is going to have things in it like "Hilton is terrible because they messed up my bill." Or it's going to have something in there that said, "The Radisson lost my reservation." And those would be two negative comments. But deep learning can do something to actually broaden that data, because it can recognize, for example, that Hilton and Radisson are both hotels. And so it can make sure that if a new comment came along that was also negative that matched one of those comments, but it was about the Hyatt, then all of a sudden it knows that because that's a hotel, it's going to see the same kinds of complaints. And so deep learning can actually use multiple types of techniques to take your training data and broaden it even beyond what you have. So that allows you to have more accurate machine learning models. It also allows you to use less training data. And because you're putting together all the training data by hand, that can be a really important thing as you're developing your AI model. But you might be asking, "Hey, that's great Mike, but what can marketers do with AI?" So let's walk you through an example of AI being used in marketing. And let's go back to what we were just talking about in the last lesson, conversion rate optimization. And you might've guessed from looking at that lesson that CRO can be kind of hard, because how do you know what to test? And you could only test what you can think of, right? Suppose there's something you should test that you just didn't have the imagination for. The other problem with testing is it can take a real long time to get results. Suppose you have a really important page, but you don't have that many people coming to it. It might be important because it's for a very expensive product that you only sell a few times a year, but you also only get a few people coming to it a year. And so what could that do? That could mean that if you ran an AB test, it might take you years to get the answer. Well, you don't want to wait years. And so what can AI do to help with conversion rate optimization? Well, there's too much to test, so AI can help you choose what the right test is to run. But you would say, "But how would it help me to understand what things to test?" Well, AI can actually make suggestions of things to test for your CRO. And you say, "Well, it takes too long to get the results anyway." And well, here's something AI can do. It can make predictions before pages are even tested or published to tell you which things you might want to improve. So how would that work? Well, it works the way we talked about before. An analyst, with your help, is going to choose some features or characteristics for the system to examine. So maybe, for example, a characteristic of a page that has a high bounce rate would be it has more than two calls to action. Maybe it's confusing, or maybe your idea is that it uses company jargon and brand terms very early in the buyer journey, before you've really gotten them interested in the problem. Maybe it has a slow page load time, or it has broken links on it, or it's got a lot of fields in the call to action form. These are all theories that you might have about what might make a page something that has a high bounce rate. And so by identifying those characteristics, those features, then what happens is that you can work with the data scientists so that they can test to see whether any of your theories are correct. Are any of those things really things that prevent people from staying with the page that caused them to bounce? And the system can find patterns in those features. And so how do you think the AI model would work? Basically, once you've trained that model using all of the steps we just talked about, you would want to know, "Hey, what is it about our pages that work, and what is it about our pages that don't work?" And you would want to run your web pages through that to see whether it can find pages that might have high bounce rates, or low conversion rates, or maybe low social shares, or very few inbound links, or other kinds of problems that might indicate a problem with the content quality. And so you might end up with a dashboard that before you even test, AI could tell you some things that are wrong, because it's looked at all of the pages out there. It's looked at all of those things that you thought might be the problems, and it's found correlations between those problems and low bounce rates, historically. Now, you're still going to want to look at it as the marketer to make sure you think that it's not just a correlation, it's something that causes the low bounce rate. And so remember, that's still your job as a marketer, but once you do that, you might be able to run pages through this, even before you publish them, and find that, oh yeah, this has slow load time, or it's not written at the right grade level, or the content is really long. And you can get a dashboard that can tell you what things are right about this content and what things are wrong that you might want to fix even before you publish the page. Why is this important? Well, because testing speeds how fast you can improve. So think about what it's like if you're doing AB testing, and you can only run one test a month versus one test a week. Or what if you could run a test every day? Hmm, well, if you could run a test every day and your competitor could only run one a month, you can be 30 times dumber than them and still be in the same place, because you can come up with 30 ideas in the same time they can only come up with and test one. And so this can help you by having this ability to test your conversion rate and optimize your conversion rate without having to spend all the time doing AB testing. And so as this starts to become available, you now can continuously optimize. AI can constantly be collecting data about your marketing and how customers respond to it, and it can automatically make some improvements. So here's just one example of how AI and marketing could help you. And this is just one thing that's coming when we think about what the marketing analytics of the future will look like.