| Personalization is an essential tactic in our IEBI/CRM strategy. As we discussed in Chapter 12, the goal of CRM is to create a mutually beneficial relationship with our customers. It is a symbiotic relationship in which we provide to the customer products and services that fulfill his or her needs and wants. The customer in turn provides a continuous stream of revenue to the company. We establish this relationship through one-to-one marketing, dealing with each customer as an individual. In the brick-and-mortar world, we have certain tools at our disposal that facilitate the creation of this dynamic, such as customer card programs. While many of these technologies are very useful, the data available to the organization limits how far we can take our analysis. We simply cannot answer some questions. The Internet takes us to the next level of customer interaction. When a customer comes to our Web site, every action that he or she makes on that Web site is recorded. This is the clickstream. We have also seen that while the clickstream is helpful in some respects, it cannot provide us with any degree of confidence an understanding of individual customer behavior. To develop this understanding, we must add cookie data to this stream of information. Cookies, while not foolproof, increase the system's ability to identify an individual customer's activity on our site. With this information, we can create a one-to-one interaction with our customers and a personalized experience when they visit our site. While we can certainly do many things to create a more one-to-one environment for our customers, not all of them should be considered personalization. In some environments, we can create customized Web pages. Customers who are primarily interested in books and music can customize their home page to display the most recent releases in their areas of interest. Customers who are primarily interested in quilting and power tools can customize their home page to provide links to companies that provide such products. It may allow them to create portlets . A portlet is a portal or Web page embedded within a Web page. Portlets can be used to display the headlines from the latest quilting or power tool magazines. While it does create for the customer a page that is tailored to their specific needs, it does not achieve our ultimate objective. It does not create a relationship with the customer in which we are consulting on his or her needs. Customization does not convey the feeling that we understand the needs and wants of the customer. Consider a wedding registry. When a couple is about to get married, they register with various stores, listing the products they would like to receive as gifts. These lists reflect the needs and wants of the couple starting a new life together. One would expect to see things like china and silver patterns or sheets and pillowcases. Typically, the bride (any former groom knows he has very little say in the matter) selects something that matches the decor of their new home. Then, when the father of the bride's golfing buddies or business associates need to come up with a gift for the couple, they simply select an item from this registry. While it is certainly generous of them to do so, this doesn't demonstrate any real knowledge of the bride and groom. They told them what they wanted. It isn't as if Rocco the mechanic , who plays poker with the groom's father every third Thursday, noticed a particularly lovely tea set while he was out shopping and thought the bride would just love it. The closest Rocco ever gets to a tea set is a plaid thermos. The same is true of customization. While it is certainly a good thing for the company to provide, it does not develop a sense that the company has any insights into the individual customer. When customers set up a personal page, they are telling the company what they want. Customization is something that the customer, not the company, does. Another means of creating a one-to-one environment on our Web site is through the use of business rules. Business rules are simplistic in their approach. If I happen to purchase Nostromo and Heart of Darkness by Conrad, business rules would recommend Secret Agent . They might even go recommend Apocalypse Now on DVD. This isn't really any more insightful than customization. Isn't it obvious that anyone who is reading through a particular author's work would also be interested in all of the author's work? If I buy a digital camera, is it especially insightful for the vendor to offer me additional lenses or a memory card? Note that none of these things should be looked at as wrong. We most definitely should provide to our customers the ability to customize their personal home page. We should develop business rules that enable a user to easily find the obvious. While I may be a huge fan of Kurt Vonnegut, I may not be aware of everything he has published. I may discover a new author like Phillip Yancy and want to read all of his books. Sure, I could go look for them on the Web site. It is easy enough to do, but it demonstrates a willingness on the part of the company to assist in the buying process. It is the equivalent in the brick-and-mortar world of a salesperson seeing a customer reaching for an object on a high shelf and saying, "Here, let me get that for you." It is a courtesy that will be appreciated by the customer. It is not, however, personalization. Personalization goes beyond the obvious. Personalization must be bold and take the consumer to places he or she might not otherwise go. Let's say that like Barney in our earlier example, I buy a copy of Object-Oriented Data Warehouse Design: Building the Star Schema . We would expect the recommendation engine to suggest books written by Bill Inmon or Ralph Kimball. We might even expect to see something from Jacobson on object-oriented design. These are all data warehousing- and software-related works. What would be interesting would be books on related topics in other disciplines. Why not recommend Edwin Abbott's Flatland or Stephen Hawking's A Brief History of Time ? Surely, anyone interested in multidimensionality would also find these other topics of interest. In addition, the consumer may not be aware of these other titles. These recommendations both enrich the customer's experience when they visit our site and provide us with a means to market additional products. It demonstrates to the customers that we understand their needs and wants. Personalization is driven by more than just the current activity reflected in the clickstream. Personalization brings together customers' current and past behaviors to tailor their experiences to meet their individual needs. Figure 14.1 presents the personalization process. The activities of the customer on the Web server are fed into the data mining engine. These activities are defined in the clickstream data. In the event that this is a returning customer, these activities are considered in the context of that customer's demographics. The data mining engine mines similar activities of other customers with similar demographics . Based on this mining process, the data mining engine makes recommendations to the application server. The application server then dynamically composes the Web pages in real time based on these recommendations. The dynamic content of a page can include any number or type of items. Products, banner ads, navigational links, product recommendations, and even product descriptions can be dynamically created. Figure 14.1. The personalization process.  Personalization is not limited to known customers, but is applicable to anonymous customers, customers whose identity, past behaviors, and demographics are unknown to the server. With personalization, our site can observe the behavior of totally anonymous users and, based on the behavior of other customers, provide recommendations. In our example from Chapter 13, when Barney first came onto our site, he followed a specific path. Other customers, who ultimately purchased products, followed a similar, if not the same, path through our site. The recommendation compares Barney's path through our site with the paths of other customers and recommends the products that were of most interest to these customers. One method of providing a personalized environment is collaborative filtering. We must be careful, however, not to confuse this with data mining. Collaborative filtering creates groups and subgroups of customers and prospects with similar profiles. Recommendations are made on how well the customer fits within a particular group . The difference between collaborative filtering and clustering may not be clear, so let's examine collaborative filtering in a bit more detail. We can see this strategy at work in Figure 14.2. In the figure, we have represented the customer as a simple square. Each side of this square represents a characteristic, or feature, of the customer. Each feature can have one of three possible values. The smaller squares made up of dotted lines represent a possible value. A customer profile, as represented by the polygon on the left, is created when we fill in a value for each of the four features. We then compare this polygon with the polygons on the left that are the profiles of the groups formed by previous customers. When we locate a group whose profile matches that of the prospective customer, we can provide advice based on the actions of those previous customers. Products and services that were purchased by those customers are recommended to the new prospective customer on the assumption that customers of similar profiles will have similar interest. While the basic reasoning behind collaborative filtering is sound, there are some drawbacks to the method. Figure 14.2. Collaborative filtering.  In our simple example, the customers have only four characteristics, each with three possible values. The number of different profiles shown to the right is a small subset of actual possible combinations. In this particular example, there are 81 different possible combinations, or 3 to the 4 power. The number of possible combinations increases exponentially. If we were to add just one more possible value to each characteristic, we would increase the number of possible combinations to 256. If we had as few as 20 different characteristics with 10 different possible values, we would have 10 trillion possible combinations! This limits the number of different characteristics that can be matched in real time, thus limiting the depth with which we can analyze customer profiles. As one can well imagine from such an example, even when the possible combination sets are within an acceptable range, collaborative filtering requires a large volume of data. In addition, actually filtering the results is computationally expensive. Just think of the last time you filled out a questionnaire for a free subscription to a trade magazine. There are sometimes 30 to 40 questions on one of those things, with dozens of different answers. Imagine the profile matching task for that many variables ! While customization and business rules are helpful to the customer, they are only part of the story. To develop a mutually beneficial relationship in which we strive to become a trusted advisor to our customers, we need to be proactive, to take the initiative. This initiative is in the form of personalization. While collaborative filtering may provide us with some capabilities in recognizing customer groups, due to its inherent limitations, it is not appropriate for environments in which customer profiles and behaviors are of even a moderate complexity. The only adequate solution for personalization is data mining. |
