Twenty-nine-year-old Jennifer Zweben has a weakness: she loves to buy CDs. Her studio apartment in San Jose, California, is littered with new purchases, everything from hip-hop to alternative, classical to techno. She's also a frequent online shopper, usually ringing up an order for CDs or books at least once a month. Indeed, just the other day, Zweben dropped by www. cdnow.com, an online music store, to check the price for her latest must-have, the movie soundtrack from "The Big Lebowski." While reading about the album, she noticed that the Web site had generated a list of other CDs for her to consider. One stood out: "Mermaid Avenue" by Billy Bragg and Wilco. Zweben, Webmaster at IBM Research, had heard a few songs from the album on the radio and liked them. "I'm always scouting for new music," she says, "but I wouldn't have remembered to look for this title on my own." She decided to pass on the soundtrack (too expensive, she says)-but plunked "Mermaid Avenue" into her virtual shopping cart.
It's no coincidence that "Mermaid Avenue" popped up on Zweben's screen. As the online shopping mall continues to add new stores, many of the Web's top retailers are banking on real-time personal recommendations to convert lookers into loyal repeat buyers. Amazon.com, Barnes & Noble, Moviefinder, CDnow, and others tailor targeted suggestions to their customers. In a recent study of 25 online merchants by Jupiter Communications, 40 percent said they already use recommendation technology at their Web sites; 93 percent of those that currently don't plan to add the application within the next year. They know there's a lot at stake: According to Jupiter, online sales could top $40.8 billion by 2002. Suggestive selling, the research firm posits, could contribute 34 percent of total sales revenues within the first year of implementation. "It's direct marketing to the nth degree," says Ken Cassar, an analyst at Jupiter in New York City.
Pumping out these recommendations is a technology called collaborative filtering. Unlike mass-marketing tools, collaborative filtering doesn't rely on demographic or psychographic profiles. Instead, it looks at individual consumer's behavioral data-such as purchasing history and stated preferences-to predict the future behavior of like-minded people. In Zweben's case, the technology de-tected her interest in "The Big Lebowski" and scoured the Web site's database to determine which other albums were bought by people who also either purchased or said they liked the "Lebowski" soundtrack. "Mermaid Avenue" turned out to be a match. Who knows whether Zweben's tastes matched those of another white Gen Xer like herself, or a middle-aged African-American male? Who cares? "Collaborative filtering allows businesses on the Internet to show a complete and unique store to each customer," says Steve Larsen, vice president of marketing and business development at Minneapolis-based Net Perceptions, a leading provider of collaborative filtering products for clients like Amazon.com and CDnow. "They're not limited by looking at you as a baby boomer, a soccer mom, or a dink [double income, no kids]."
Like a good salesclerk, collaborative filtering sharpens its suggestions-and increases the likelihood of a sale-as it learns more about the customer.
Take Tower Records, for example. At the beginning of this year, the West Sacramento, California-based retailer started to send a monthly e-mail newsletter to customers who signed up for the service. Each newsletter contained eight to ten album offerings produced through collaborative filtering and targeted especially to the recipient. Hyperlinks in the e-mail connected the customer to each album on the retailer's Web site. Tower's first mailing had a 4.4 percent response rate, says Ray Kaupp, vice president of marketing at Digital Impact, which implemented the campaign for Tower Records. In subsequent mailings, collaborative filtering honed its recommendations even further based on which albums the customers checked out. In three months, the response rate rose to nearly 15 percent.
Fans of collaborative filtering also praise the technology's impact on sales and repeat traffic. Sixty-three percent of Amazon.com customers are repeat buyers, says Net Perceptions' Larsen, while a typical commerce site averages 35 percent to 40 percent. Amazon. com is one of Net Perceptions' oldest clients, having used its GroupLens recommendation engine for the past year. "How often do you go back to a restaurant where the maitre d' remembers your favorite table and the fish is always fresh?" says Larsen. In fact, Jupiter forecasts that merchants using personalization tools can break even on their investment within 12 months. Clients of Net Perceptions' GroupLens recommendation software pay between $45,000 to $250,000.
Still, use of the technology needs fine-tuning. To get recommendations at most online stores today, customers must stop what they're looking for and jump to another area of the Web site to access them. That takes time-and motivation. Jupiter's Cassar expects implementation to improve within the next few months as retailers better grasp the cross-selling potential of the application. "Product suggestions will pop up on a customer's screen based on what they're buying at that moment," he predicts. "Customers won't notice it, but it will improve their shopping experience."
There's another drawback though: Online retailers don't know whether their customers are buying for themselves or someone else. As collaborative filtering works now, if you buy a book on gardening at Amazon.com for your friend's birthday, the retailer will suggest similar books to you from then on in-even though you have no real interest in pruning.
"As we come into the holiday season, many people will be buying gifts," Cassar says. "That information shouldn't be part of their purchasing behavior because it doesn't relate to them."
Despite these limitations, online merchants still consider collaborative filtering a valuable tool. Ticketmaster Multimedia, for example, just launched a personalization site called my.ticketmaster.com that relies on the technology to target advertising and content to consumers. Users of the site fill out a survey, including their home address and entertainment interests. Andover, Massachusetts-based Engage Technologies captures and updates these profiles; Net Perceptions' GroupLens filters the information to gather like-minded people together. For instance, say a Chicago user mentions she likes jazz music. When a jazz concert is announced in the Chicago area, Ticketmaster immediately serves up information about it to her the next time she visits the site. It may also suggest other events to her, based on what people who have expressed similar tastes have liked. Ticketmaster is banking that personalization will boost online sales, which already top $10 million per month.
Perhaps the most intriguing potential for collaborative filtering lies off the Web, not on. Net Perceptions recently met with a major U.S. retailer that's evaluating the potential impact of the technology in its stores. The chain's average customer, according to Larsen, drops by every 21 days, purchases an average of 13 or 14 items per trip, and spends approximately $75. If GroupLens were at work, it would detect items as they were scanned at the cash register-shampoo, nail polish remover, and deodorant, for example- and instantly determine which other products were bought with these items in the past by other consumers. Maybe toothpaste and chocolate chip cookies. Printed on the back of the customer's receipt would be coupons for toothpaste and cookies, valid for the next 14 days. If 1 percent of customers returned to the store in 14 days instead of 21, Larsen says, the retailer could boost its bottom line by $10 million.
Here's another scenario to contemplate: Picture standing in the checkout line at a grocery store, your cart filled with Thanksgiving standards: a bird, stuffing mix, veggies, potatoes, cans of mincemeat and pumpkin pie mix. The cashier rings up your purchases-instantly fed through a collaborative filter-and then asks, "Did you need cranberry sauce today, sir?" You gulp, wondering how the clerk knew you forgot the sauce, then nod heartily. She must have read your mind.