Customer retention is the name of the game, and "Web mining" your customer interactions can help you win it
By Jesus Mena
With every click of the mouse, visitors to your Web site are telling you what products and services they want. But more important, they are also telling you who they are; these visitor and customer interactions reveal important trends and patterns that can help you design a site that effectively markets your products and services.
However, like most other firms, your organization is probably unprepared to decipher this avalanche of data. A recent survey by Forrester Research of 50 of the largest U.S. corporations found that three out of four were not using their Web data at all. (Admittedly, all the surveyed firms vow that they plan to use this information to drive their marketing, sales, and customer service by next year.) And if large corporations are not using their Web data, small-to-medium sized companies are even less likely to leverage this asset to uncover core customers.
Your firm is probably already using some sort of log analyzer and may subscribe to an ad network such as Flycast. However, log analyzers, packet sniffers, plug-ins, ad networks, and ad managers report mainly on TCP/IP activity and not consumer demographics, lifestyle, values, behavior, and attributes. In essence, they report on browser activity, not individuals. Some of these tools do provide valuable information about how visitors found a site, what keywords and search engines they used, how long they stayed, and so on, and banner ad networks and ad managers can provide some valuable insight into clickstream behavior. However, the bottom line is that these types of tools provide little insight into consumer habits and life- styles, which are what e-businesses need to know in order to manage customer relationships.
In retailing and other verticals, whoever has the best knowledge of their clients’ preferences, rate of consumption, and overall behavior will get a bigger share of business in the long run. In addition to click-through behavior, your Web data can also reveal certain lifestyle information about your customers, which in the end you can use to strengthen your relationship with them. Failing to leverage this valuable insight about your current clients will only damage how you deal with them — eventually driving them to your nearest competitor.
As I’ll explain here, the most effective e-businesses will be those that use the Web data they generate daily to make ongoing strategic business and marketing decisions. In addition, successful long-term organizations will also be those that mine their Web data and combine it with external demographics to better understand who their core customers are.
Collecting Web Data
Meeting this goal comprises several fundamental steps: including collecting online data, enhancing it via offline demographics, and then mining it to drive customer relationships. But one of the problems in mining Web data is its diversity; a single customer interaction may be captured in log files, email servers, or databases created from site forms.
Every visit to a retailing site generates important consumer behavioral data in these places, regardless of a sale. Thus, one of the challenges in mining Web data is organizing it into a cohesive view of visitors and customers. A good strategy is to quickly interact with your visitors and customers via registration and purchase forms, enticing them with contests, free offers, and discounts. These forms are by far the most effective method of gathering Web site visitor and customer information.
Forms are a simple browser-to-server mechanism that can lead to a complex array of customer interaction from which lasting customer relationships can evolve. This relationship can turn into a direct feedback system through which a consumer can communicate with your company; helping you gather important consumer information that you can further enhance with offline demographics. For example, forms let “e-tailers” interact with customers and collect intelligence about their needs, values, choices and preferences.
However, you have to be careful to ask for only the most essential information; nobody likes lengthy and intrusive questionnaires, especially those dealing with personal finances. Various optional methods and sources for gathering demographic information — rather than asking for it directly — are available.
For example, in order to benefit from third-party data providers, you need to capture both email and physical addresses from your visitors and customers. To ensure you are capturing the correct information, give them something in exchange — anything from a free calendar, to coupons, a white paper, or a mouse pad. Then you can use the addresses to match and append additional demographics as well as deliver a targeted offer or incentive. A demographic provider can use the physical address to match important household information, while an e-tailer can use an e-mail address to deliver a targeted offer or incentive to that same customer.
Enhancing Web Data
The success of any Web-mining project largely depends on the quality and depth of its data. A common methodology in data warehousing is to leverage the value of internal customer information by appending the external demographic and behavioral data. Similarly, you can append a variety of demographics to the information you capture from your registration and purchase forms. This demographic and household data can range from your customers’ projected income to the presence of a pool at their home. You can then link this external information via a physical address to your Web site database, enabling you to gain an additional insight into the identity, attributes, lifestyle, and behavior of your visitors and customers.
This demographic and household information is available from various vendors, including Acxiom Corp., CACI, Experian, and Polk. In fact, an entire industry is now devoted to segmenting, classifying, and reselling consumer behavior information. Obtaining a cohesive and comprehensive view of customers involves not only using powerful data mining technologies; it also requires enhancing internal transactional data with this external consumer information, which describes the tendencies and values of customers in detail.
For example, new products from several large data depositories — such as Acxiom and Experian, which have consumer demographics on more than 95 percent of U.S. households — make it possible to match and retrieve consumer and household information in real time over the Web. These products include household information such as age, education, occupation, marital status, presence of children, household size, income, and net worth.
You can couple this household information with online transactional data, mine it, and then develop a profile of a your site’s most profitable and loyal customers. Not only can e-tailers benefit from this type of analysis, but so can content providers — by identifying their core audience and adjusting their marketing efforts accordingly. Access to these types of consumer demographics, coupled with data mining technology, can significantly boost customer relationship management in ways that directly affect online visitor acquisition and retention.
Mining Web Data
Let’s take a look at an example. Using the Clementine data mining tool from SPSS Inc., my company performed several analyses on a data set from an e-commerce site that sells clothing for men and women. The transactional data from the site included the number of purchases and type of items bought by customers, to which we appended household demographic information such as gender, age, income, and other consumer attributes. (See Figure 1) Then, using Clementine’s visualization component, we found some interesting associations among gender, age, and number of purchases made over time. In summary, the link analysis found strong associations between older male customers: They tend to make a high number of multiple purchases.
FIGURE 1 Purchase and product information appended with demographics.
We next performed a segmentation analysis using the tool’s rule generator component. Segmentation is the process of dividing your customer base into smaller markets based on different needs, preferences, behavior, and attributes — the idea being that as you segment your customers into smaller and smaller sectors, you will be able to interact with them in different ways. For example, when you segment your customer database so that you can find out who your most profitable clients are, you may want to orchestrate a marketing campaign to reward them for retention purposes.
The segmentation analysis, like the prior association analysis, found age to be a key factor affecting the number of sales a customer was likely to make. One of the most important market segments discovered by this analysis is simply that when the age of a customer is equal to or greater than 45, the number of multiple purchases tends to be quite high (nine). Other key attributes discovered in this analysis were the importance of a customer’s income and projected worth.
One of the benefits to this type of Web mining is that it lets the e-tailer target specific offers and incentives to a smaller segment of all its customers based on historical purchase patterns. In this case, the company in question can use these type of rules to drive its email marketing campaign more intelligently: Rather than marketing to all customers in the same way, it can focus on rewarding those clients who are over 45 because they are by far its most profitable ones.
Finally, we constructed a predictive model from our Web mining analysis using Clementine’s neural-network component. First we “trained” the neural network on the historical Web data in order to predict the number of purchases new visitors are likely to make. After we trained the model, we viewed its overall design to find out what attributes are the most important inputs for predicting the output of the total Projected Number of Purchases. A sensitivity report showed that the network achieved a predicted accuracy of 94.67 percent, and that these input values were the most important:
Customer Age .55976
Customer Income .31237
Marital Status .17328
Projected Worth .03903
Gender .03834
Presence of Children .01982
As you can see, the most important value for predicting the number of sales at this site is the customer’s age, with income, marital status, worth, gender, and presence of children following in priority. Furthermore, Clementine, like most other data mining tools, not only helps you construct predictive models but also generates C code that you can compile and incorporate into a production system, such as a marketing or email manager. This system can then use either the rules or formulas from a neural network for targeting potential new sales prospects.
In this Web mining analysis case study, it became clear that certain customer features are the driving forces in predicting online sales for the company. Characteristics such as age, household projected income, and marital status are the key factors in determining the number of purchases customers would make. For this company, it is clear that its most profitable customers are mature males over 45 years in age with a high income.
Interestingly, through the processes of segmentation and classification, this analysis uncovered hidden patterns and structures in this data set — such as the fact that although this particular retailer is very popular with young people and kids, its most profitable online customers tend to be older consumers. (Perhaps the youngsters are doing the buying using their parents’ credit cards.) Furthermore, we discovered that this e-tailer’s prime customers are smack in the middle of baby-boomer country, which according to Media Metrix, an Internet and digital- media measurement service, represents 35 percent of the Internet’s 63 million users. These consumers are aged 35 to 49 and have household incomes averaging $75,000 and above, compared to only $58,000 among the overall online population. Plentiful, wired, and wealthy, these consumers are an e-tailer’s marketing dream. Of further interest to this particular company is the fact that these consumers are very interested in recreational travel (given their high disposable income), which it may leverage in its marketing promotions at its site.
Leveraging Web Data
During the past five years, the number of U.S. households with online access has grown from 6 million to 37 million; with 10 million of those households coming online just within the last year. Projections for online sales vary; however, they all agree that this marketplace is exploding. (Forrester Research predicts online buying will hit $185 billion by 2004.) Although online sales still account for much less than 1 percent of total retail spending, the possibility that they might soon account for much more has produced a high number of dot-com retailing ventures.
Furthermore, not only can most items be bought online; they can be bought online at dozens of different places. This amount of choice translates to increased competition among e-commerce sites for the attraction of new and existing customers for everything from cars, books, and insurance to pet supplies. As consumer options increase, retailers will face pressure to improve customer service, broaden product offerings, and reduce prices. As competition increases, so will the need to attract and retain online customers, which is where offline demographics and data mining come into play — the Web offers an incredible channel for understanding customer behavior and expectations.
Indeed, customer retention will become the metric by which e-retailers will measure themselves. (For example, the Yankee Group estimates that new customer acquisition costs for Amazon.com increased from $24.89 in 1998 to $37.37 in 1999.) Web data mining is ideally positioned to give e-tailers a methodology for acquiring and retaining customers. As the marketplace grows and consumers become more sophisticated, Web mining will be a key to attracting and retaining them.
One caveat about mining your Web data: Tell your visitors why you are doing it and give them a choice of opting-out from the process. Explain your practices clearly and let your customers have access to their own profiles. In the end, they will realize that your objective is to service them better than your competitors.
Jesus Mena (jmena@webminer.com) is the CEO of WebMiner, a Web-mining consultancy, and the author of Data Mining Your Web Site (Digital Press, 2000).
