Online sellers may soon offer
personalized service not available in any store.
As e-commerce matures, merchandisers are coming to appreciate the difficulty of building customer loyalty in cyberspace. A familiar store name can only draw customers; it can't keep them. Look-alike Web sites tend to merge in shopper's minds into one giant, ramshackle discount store, high on convenience but low on amenities. To win customers over, retailers have to offer new services that are unique to online shopping.
IBM is developing several solutions and has begun testing them with retail customers. One novel concept, for example, is the "personal book," an online organizer that tracks a consumer's dealings with one or more e-business vendors and is carried from Web site to Web site.
"Imagine a chart with a few lines running across it, each representing a different variable," explains Clare-Marie Karat, a user-interface researcher at
IBM's Thomas J. Watson
Research Center. "One line could be purchases for yourself, another could be purchases for your spouse, another could represent birthdays and holidays. Along each line, icons represent purchases. If my husband likes a particular kind of shirt, I can pull up that icon to see where I bought it, how much it costs and how many I've already bought him."
For customers, the book provides a way to jog their memories or make sure they're not spending twice as much on Dad as on Mom. In short, "it gives you tremendous control over your shopping," says Karat. Her team has filed for a patent.
For online merchants, meanwhile, the book offers a chance to personalize the shopping experience. Loyal customers might be notified of new merchandise in their favorite lines. Or the merchant could tailor customers' view of the merchandise to reflect their answers to questions about their lifestyle and preferences. In making purchases, customers could even arrange to have certain consumables replenished regularly. These and other forms of personalization can save customers time and energy. "The personal book lets merchants coordinate their customer service and customer relationship management activities through one interface," says Karat. "Its data visualization capabilities enable merchants to see and respond to customer activities with greater understanding and insight."
Moreover, sellers will be able to adjust the level of personalization according to the amount of business a customer provides. Karat calls this concept "affordable intimacy." The best customers, for example, might be offered the services of a personal shopper who has been able to review, with the customer's consent, his or her preferences, lifestyle information, and past purchases. As they shopped together on the merchant's site, the personal shopper would offer information and suggestions.
To display transaction history, the shopping organizer uses a system called LifeLines, a data visualization approach that depicts how personal information changes and grows over time. LifeLines was developed by Ben Shneiderman and Catherine Plaisant, of the Human-Computer Interaction Laboratory at the University of Maryland, to track juvenile offenders through the justice system. IBM then partnered with Shneiderman's team to adapt the system for use in patient medical records. A program that can represent a lifetime's worth of medical events helps doctors spot the information they need and gives patients ready access to their medical history, Shneiderman says. He envisions other uses as well, such as tracking pension information.
IBM Research received positive feedback when it demonstrated prototypes of the shopping organizer to U.S. retailers, and is now incorporating their suggestions.
Another approach to personalization is the "recommendation algorithm," a program that anticipates customers' desires based on past purchases. Amazon.com uses one to recommend books.
Most recommendation engines now used in Web retailing simply tote up all the buyers of a particular item -- for example, if you purchase a compact disc, the retailer lumps you in with all other buyers of that title. It might then recommend another artist's music that was purchased by at least 60 percent of the people in this group.
However, buying behavior is not so simple. For example, the CD might have been a gift for someone whose taste differs from your own. So retailers need an engine that bases its recommendations on your entire buying history, not just a single variable like "last CD purchased," says Philip S. Yu, a researcher at Watson.
SMART SUGGESTIONS
Yu and his colleagues have developed a prototype called the Intelligent Recommendation Analyzer, which can use a variety of personalization techniques simultaneously. After all, he says, different considerations go into the purchase of a house, a car, a book or a CD. With its ability to incorporate different algorithms, the analyzer is not limited to one kind of product. Rather than simply aggregating past buying behavior, says Yu, the system "takes advantage of the context of the shopping session." For example, he explains, "since a purchase can be either for oneself or for somebody else, the analyzer can take into account the current browsing pattern."
Recommendations can be useful, says Yu, "because often people have only a fuzzy idea of what they want. They may want to find a book on a certain topic, but do not know the title or author. They may know what they want a printer to do but not have a specific model in mind. An intelligent recommendation system can steer people toward appropriate products on the basis of their buying history or browsing and searching patterns."
Such a system would be a boon to advertisers, who could show customers items likely to be of interest that day. It could also help the growing number of portal sites (like AOL or Yahoo!) that offer personalized Web pages; more precise information about preferences can be used to create a Web page that is "truly relevant to your experience," Yu says.
The analyzer rapidly processes information -- which might come from past purchases on a Web site, or from a survey filled out by the arriving customer -- and finds the "peer group" most relevant to that particular shopping expedition. "Part of what we have done is develop the mathematical techniques needed to do that clustering fast enough," says Yu.
TRACKING PROGRESS
The cultural shift that personalizing tools require will not come easily, though. Merchandisers know the 3-D shopping experience well, but may not be experts on the Web. Meanwhile, Web-savvy technicians don't think the way retailers do.
To bridge the cultural gap, Watson researchers have developed a prototype solution called E-Commerce Intelligence. "It's a tool for linking the kind of information a Web page generates with the kind of information a retailer wants," says researcher Edith Schonberg. Retailers want to know, for instance, what products do better if sold together. "For each product, we keep track of how often it's put before the customer on the site, how often it's clicked on, and how often each appearance leads to a purchase," Schonberg says. She and her team are working with a major retailer to incorporate the solution into its system, and they are talking with other IBM customers.
David Berreby is a freelance writer who lives in New York City.
