Yo.com Tries Targeting En Masse
Collaborative filtering, which draws anonymous profiles of users based on their clickstream and transactional behavior and makes recommendations based on the actions of similar users, has often been criticized for serving up obvious suggestions. For example, if a user buys a sprinkler, it might suggest a new garden hose.
To combat this problem, Yo.com is attempting to build a network of sites that share data on user activity in an anonymous fashion. The idea is that the more data the company collects, the more fine-tuned the offers will be.
"For collaborative filtering recommendations to work, you really need a whole lot of data," said Lixuan An, president of Yo.com, New York. "If I don't know you at all, then I don't know a lot about what you would like. If I know you for a month, I'll know the book you liked or the movie you hated. Then I can recommend something."
Yo.com has the right idea, said Sam Clark, senior research analyst at the Meta Group, Stamford, CT. "You need a decent amount of volume for the data to be worthwhile," Clark said. "It improves the chances of making better offers."
In a short time, the company has built its network, which includes small sites such as Allaboutjazz.com, Wizshop.com and Lanaconline.com. However, it has failed to pull in some of the bigger-name sites.
The sheer variety of sites it is attracting is one potential stumbling block for Yo.com, as collaborative filtering is not great for connecting disparate data, Clark said.
"It works best when you ask people their opinions about a set of similar things. If you give all of your favorite drama movies, it can tell you what other movies you might like," Clark said. "It's not good connecting products that are different and distinct. Trying to establish a connection between people who like jazz and Lands' End knit polo shirts isn't going to work."
This strategy also raises privacy concerns, said Steve Larsen, senior vice president at Net Perceptions, Minneapolis, a provider of advanced marketing solutions. Larsen knows a thing or two about collaborative filtering, as companies such as Amazon.com, CDNow.com and eToys.com use his company's technology to target offers on their sites.
"The reason we haven't done it is concerns over privacy issues," Larsen said. "Most customers accept the idea of a company like CDNow understanding their preferences and using that to help enhance the experience that they have on their site.
"It's a different thing to take that preference information that's gained on one site and use it somewhere else unless it's done with the full consent and knowledge of the customers. Our experience is that customers aren't particularly keen on the idea of cross-site profiling."
However, since all the information is anonymous, "we don't have this problem," An said. "It's implicit behavioral data, not personally identifying information."
An believes the company's model is not its only competitive advantage; its pricing also serves as a differentiator, she said.
"Collaborative filtering is expensive. It can cost $140,000 and up and take three to six months to implement," she said. "Only the big boys can afford the time and the price tag. It's an elite product for an elite few."
Yo's setup fee ranges from $5,000 to $10,000. The company earns commission on the revenue it generates for its client sites. It can be implemented in days for merchants and in minutes for affiliates.
Sites that belong to the Yo.com network offer a Yo Box where product and content recommendations are served to Web site visitors. The box can be expanded to serve as the entire page of a site or catalog. The page and boxes have the same look and feel of the client sites.
Yo.com encourages clients to offer the Yo Box to their affiliate sites for free. Instead of the standard, stagnant affiliate banner ad, the box will serve up targeted offers designed to draw consumers to the parent site.
Yo.com also is looking to partner with an e-mail company this month. The agreement will have Yo offers incorporated into 23 million HTML e-mail messages.