Predicting that next purchase

You’ve just purchased a Bonnie Raitt CD from online music seller CDnow. Immediately, the Website analyzes your needs, using information about other customers’ past purchases within the same music genre, and recommends the latest Eric Clapton CD, which flashes on your screen. Sound impossible? Not with collaborative filtering technology software.

Collaborative filtering software uses customer data analysis, statistics, and pattern matching to compare one online customer’s responses or purchases to those of a universe of online customers. The software then immediately predicts and recommends additional products that might interest the customer. As a result, a cataloger can create customized Websites for customers, offering in real time the products most relevant to their needs. The more customers buy, the more information is stored, thereby increasing the number of possible recommendations. Collaborative filtering enables catalogers to upsell and cross-sell; i.e., automatically prompting a customer who buys a clock to buy a battery.

Collaborative filtering is “a powerful predictor of customer behavior,” says Steve Larsen, vice president of marketing and business development for software developer Netperceptions. The software can cost $50,000-$250,000, depending on your size and needs. Catalogers should also purchase a separate server to house the data to avoid overextending existing servers.

Rivertown Trading Corp., a St. Paul, MN-based multititle gift cataloger, began testing collaborative filtering in October; Minnetonka, MN-based general merchandise cataloger Fingerhut plans to start testing the technology by spring. Both companies say that the opportunity to upsell and cross-sell products is the driving force behind the tests. “With collaborative filtering, we can go beyond traditional datamining techniques to entice customers to buy more products online,” says Jane Westlind, Fingerhut’s director of electronic commerce.-SO