In an earlier video we learned about some of the classic business problems that can be solved using data analytics. Our objective was to start shifting our focus from analytic methods back to the business context and the issues that are driving our analytic process as a whole. We can then start thinking about all the things we need to do after our analysis is complete, to drive action in a market place. In this video we're going to spend just a little time rounding out our perspective by looking at some emerging problems that have stimulated new types of analytical methods and solutions. Mainly areas involving the web, mobile, and social media. Unless you've been hiding under a rock somewhere these technologies probably need no introduction. But just in case let's outline exactly what we're talking about. When we say the web, we're talking about how people navigate and use the Internet. More specifically, we're often focused on how organizations interact with their users or customers' via website. With Mobile, we consider where people are and what they are doing by focusing in the presence and usage of mobile devices. Whether they be mobile phone, tablets, laptops or other connected devices that can move with a user. We can actually expand this a little to include objects that are connected to mobile networks. Whether they be mobile like a car or truck or not like the water meter at your house. Social media speaks to how people interact and share information with each other in an online connected environment. This is actually a pretty broad concept which incorporates a number of ideas, as illustrated in this nice figure from Brian Solis and JESS3. In our discussion, we'll focus on a couple of aspects of social. Namely, how organizations look at what users communicate and how users are connected to each other. Okay, now that we have the basics defined, let's outline some of the business issues associated with each area. And how we use analytics to address those issues. Let's start with the web. Just about all organizations have one or more websites that they use to interact with their users or customers. One of the most common needs that organizations have, is to understand how users interact with the website itself. Questions we might want to answer include the following. How many users are coming to the website, and where are they coming from? How long do people stay on the website? How do people navigate through the website? Are people getting stuck somewhere, or is navigation fluid? How often do users access help or FAQ sections of the site? How often are certain actions, like purchasing, happening? When do they leave the website and where do they go when they leave the website? Understanding these types of patterns helps us assess whether our website is well designed and whether we are stimulating the right volumes and types of activity. For example, let's say the we're an online retailer. We might find the traffic to our site is very strong and that people spend quite a while browsing products and regularly add products to their online shopping cart. But for some reason those products aren't being purchased. By looking closely at exactly where customers are falling out of the process, we might find that a point in the navigation path is broken or confusing. So how do we typically get this type of information? It turns out that most websites have measurement mechanisms built into the site itself. Usually each page or element on a page contains a small piece of code that generates a message when your user arrives or clicks on something. These messages are written to logs and can be interpreted using a web analytic services like those provided by Google's, Adobe, IBM and Web Trends among others. Or driven directly into an organization's data environment, we can incorporate the data from website into a variety of descriptive productive prescriptive analysis. And there are a number of common web analytics matrix that are used to characterize behavior. Two very common types of analysis you might see are funnel analysis and path analysis. Funnel analysis traces how many customers move from one major stage of a process to the next. For example, from the website homepage to a shopping cart, to a registration process, to an actual purchase payment. The visualization of this type of analysis tends to resemble a funnel. Hence the name. Path analysis is a bit more involved and quantifies the ways that customers navigate through a website which is usually not linear. Here's a simple example of how path information might be represented based on web data. The other major way that organizations use the web is for online advertising. We may want to ensure that our website is listed prominently when people search the web. Or we may want to drive people to our website from other sales channels by using ads on third party websites. The way we measure the activities stimulated by online advertising is similar to the way we measure other web activity. Namely through the same snippets of the code that fire when someone clicks on our ads or someone arrives at our web site from specific third party locations. However, the way we figure out where and how to advertise involves some special types of analysis, including Search Engine Optimization. Where we try to understand the algorithms used by search engines and set up our web site to optimize when our web site shows up. The web is obviously massive and complex. And there are many more ways in which web data can be used in analytics. But again, these are the most common ones you'll see in most organizations. Let's move on mobile, first many of the same ideas that apply to that web also fly to mobile. Users can access web sites on mobile devices. And we like to understand that experience and potentially how that experience differs across different types of devices. We can often use web analytics techniques to get the answers. Mobile also adds another dimension to our advertising use case. Companies are increasingly using available information about a user's location to provide highly localized targeted ads to the right people at the right place, at the right time. However one of the richest areas of insight provided by mobile data is around how people and objects as a whole are distributed geographically and how they move around. It's possible for many companies to get access to this information. Here is some examples. A retailer's mobile application ask for permission to access GPS information on a device. A wireless carrier detects which cell towers connect to a device. A fast food restaurant chain sees when a mobile device connects to a Wi-Fi hotspot in the store. A trucking company installs wireless devices in all its vehicles. There are a wide variety of marketing, sales, and operations problems that can be by understanding location data. In fact, there are entire businesses that are built with location data at the core. Think of your favorite navigation or traffic app on your phone. There are also businesses that are using precisely positioned connection points within their physical locations to track how customers move through a building or store. Broadly speaking the techniques we use when working with location data fall into a class of analytics called spatial analytics. Normally this involves plotting location points on a map of some sort. And using visualization or algorithms to identify patterns and draw inferences on behaviors. We would also use related techniques like boundary triggering, cluster pattern analysis, or hot spot analysis to drive insights around specific problems. Let's look at a simple example of how we might use mobile location data. To identify different patterns of behavior and perhaps assign specific characteristics or segments to customers based on what we see. Supposed we're working in a wireless company and we want to examine where customers tend to use their phones for making calls. We might use location data to identify a few patterns like these. In the first figure, we have a bit of a a barbel pattern with use on a path between two end point clusters. This might represent someone travelling between work and home, or school and home locations. In the second example we see a single dominate use location, this could be someone who works at home like a homemaker or owner of a home business. In the third example we see clusters of use in various locations across the United States. This might be a consultant or travelling sales person. We could use classification based on these patterns directly in business strategy or marketing decisions. Or we could drive these distinctions into other analytical efforts such as broader segmentation or direct marketing activities. The last emerging area we'll cover is social media. As we noted earlier this is a pretty broad space. But let's look at a couple of the most common business problems that social media can help to inform. The first is helping us understand how our organization is perceived in the market. Analysis of the patterns and content of comments made about our organization can help us identify what things are going well. And whether there might be issues we need to address. This in turn can help shape strategy and tactics in the market. So, how do we actually do it? Typically, we start with a tool or service that scrapes relevant content from a set of social media sites, blogs or other sources. Using this data, we can construct simple measures, like how many times people refer to our organization. And where or when people are talking about us. We can also employ this technique called sediment analysis which uses text mining to identify words or phrases that are positive or negative or which refer to specific ideas or events of interest. This type of social media analysis tends to focus on how people relate to our organization individually or as a whole. However, there are some really interesting insights that can be gained by examining how people interact with each other. There's some ways this can be done using publicly available data but the richest insights are possible when an organization has access to a complete set of interaction. Social networks themselves obviously have access to this type of data. But so do organizations like telecommunications companies who can see things like calling patterns across all customers. If we do have access to broad network data, we employ methods like social network analysis to identify interaction patterns, groupings and even the roles of individuals within their own personal networks. Let's use an example to illustrate. Let say we're able to see how a group of people interact with each other. We can use the presence of interactions to construct the following network graph of the group. Right away we can see what looks like two more connected subgroups linked by a few individuals with common connections between those subgroups. Now with only 11 people it's pretty easy to see how the connections exist. But you can imagine how difficult visual interpretations might be with hundreds, thousands or even millions of people. It turns out that we can use some numerical matrix to identify people who occupy key positions within a social network. Here are a few popular measures each of which has a specific numerical calculations behind them. The first is Network Centrality, which is the number of direct neighbors someone has in the network. As the name suggests, these are people who tend to sit at the center of network clusters. So, who do you think has the highest degree of Network Centrality in our network? If you said Jane, you're correct. With four direct neighbors, Jane has more direct connections than anyone else. A second measure is called the Closeness Centrality which is basically how close each person is to all others in the network. This one is a bit harder to do visually. Can you identify a couple of people who might be the closest to all others? In turns out that Frank actually has the highest level of Closeness Centrality, just edging out Ethan. A third measure is Betweeness Centrality, which is a measure that identifies key linkages between nodes of a network. What do you think about this one? Again, both Ethan and Frank occupy critical positions in the network, as they link the two nodes together. The final measure we'll look at is called Clustering Coefficient Centrality which is also known as the all my friends know each other measure. It's the degree to which someone's direct connections are connected to each other. Who do you think scores highly in our example? Because all of Chris and Brad's friends are also connected to each other they have the highest degree of Clustering Coefficient Centrality. In fact when we have a set of people who are all connected to each other we called this is Clique it turns out the Brad, Chris, Dave and Amy form a Clique in an narrow since they were all connected directly to each other. So how might we use this information? One of the things we might do is try to identify influencers within a social network and engage them in some way. Our hope is that we can ultimately reach more people with less effort by taking advantage of word of mouth or other social cues that influencers might propagate on our behalf. We might also make sure that influencers have great costumer experiences. Since the impact of poor costumer experience can likewise be amplified far beyond one individual. But who exactly is in influence? It's not always the case that people at the center of the network are the most influential. Usually what we need to do is a little testing against our network to see how messages propagate. For example we can send information about an offer to a set of potential influencers then look to see who else seeks out the offer. This would provide information about who the real influencers are. Our centrality measures help start the process by identifying the best candidates for a potential influencer pool. One thing you may be noticing, is that while all these areas, the web, mobile and social media are relatively new in the broad history of business and commerce. The basic business issues associated with them are not really all that different than those organizations have faced for decades. Businesses have always needed to understand how their sales and communication channels are functioning. Understanding location and where people move have long been a part of retail strategy. And the impact of community and word of mouth is as old as commerce itself. What is different is how broad a reach newer technologies have and the amount of data we are able to capture about how they work. This opens up all sorts of new possibilities, and of course, opportunities for analytics. Keep this in mind as we continue to explore how we turn our analyses into strategy and tactics for action in the market.
