In responding to customers' quests for individual goods and services, the World Wide Web often seems not only a stateless environment, but a mindless one as well.
Analysts agree it remains a struggle to make the Web more responsive to the individual visiting a given Web site and less inclined to assume one size fits all. The ability of automated systems to recognize and respond to us as individuals is still a far cry from that of the lowly shoe clerk, industry representatives say.
E-commerce vendors have taken a few baby steps toward personalization, but they must struggle with how well their systems can identify and research what they know about a visitor, and still react in real-time while the visitor is on the site.
The first groping attempts to react in a personalized way are being implemented at book and CD seller Amazon.com, Levi Straus, Yahoo!'s portal site and other sites. All allow users to enter personal data so information about products can be served up to individuals. But real personalization, where the system knows who you are and what your preferences are as you move toward a buying decision, still seems a long way off.
"Overall, it's very early in the implementation of personalization systems," says Steve Larsen, senior vice president at NetPerceptions, maker of NetPerceptions for Ecommerce, a collaborative filtering e-commerce system.
The dilemma: Online retailers want to solicit customer preferences early in a site visit, but customers are turned off when confronted with a lengthy questionnaire as soon as they hit a site.
Another way of divining customer interest is a quick analysis of clickstream data, as a visitor navigates a site. A system can generate a profile of the user based on other users who previously navigated the site in a similar pattern. Through the technique of collaborative filtering - getting a quick sketch of an individual based on the pattern of a like-minded group - companies such as NetPerceptions and Andromedia can anticipate what a visitor will be interested in and try to offer an appropriate promotion or package of goods. Andromedia earlier this year acquired the collaborative filtering technology of LikeMinds and combined it with its site management and reporting product line.
Another way to learn about the customer is mining data and analyzing it from a variety of sources, including the customer's record of previous purchases. But data mining is typically done behind the scenes through lengthy queries to a data warehouse. The site visitor could come and go before the typical data mining system could come up with a profile of who that user is. In addition, says Steve Kanzler, vice president of marketing at Andromedia and former chief executive at LikeMinds, a purchase history analysis does not necessarily reflect the customer's present interest the way clickstream analysis does.
"It's not just purchase history that we want. It's purchase history in the present context," he says.
Right now it's painful to integrate all these data sources," says Kanzler, whose firm is using the eXtensible Markup Language Metadata Interchange standard as a means of exchanging data between legacy systems and its LikeMinds analysis engines.
Another company, DataSage, connects legacy data sources to its Customer Knowledge Hub, a database that keeps its focus on the individual's purchase history, customer service and other real data. Much of the data, drawn from outside sources, gets batch processed before being called up to be combined with click stream analysis and expressed customer preferences.
For those who wish to do business on the Web, personalization will quickly move from an expensive frill - systems cost $50,000 to $75,000 - to a necessity.
When personalization works, "only the things you find interesting will be put in front of you," DataSage CEO Dave Blundin says. If the seller's context is not relevant, that seller "will be weeded out. It will be a matter of Darwinian survival," he says.