Push developers are trying to reduce this churn by championing improvements in personalizing content for individual users.
Vendors of personalization software for push applications, such as Autonomy and Net Perceptions, say their applications are ready solutions to help users get only the information they're interested in. Internet analysts, however, caution that personalization technology is still quite immature.
"The state of personalization today is extremely raw," said John McCarthy, group director of new media research at Forrester Research. "We're at the point with push and personalization of laying the 20-inch water main down the middle of the street. Water is being delivered to people, but if you try to drink out of the 20-inch main, you'll drown."
Most Internet publishers offering push services currently try to customize information with broad strokes such as topic-specific channels or keyword filtering. The Sporting News (www.sportingnews.com), for example, offers 10 SportsMail channels divided along the lines of NFL, NBA, NHL and so on. But fans who want to follow only one particular team must still receive all the stories in a category.
Inquisit (www.inquisit.com), meanwhile, lets users subscribe to news from dozens of sources delivered by e-mail according to keywords. A combination of keywords "Web" and "advertising," however, delivers about 20 stories a day, of which maybe only one or two focus on the online ad industry.
USING ARTIFICIAL INTELLIGENCE
Software developers Net Perceptions and Autonomy use artificial intelligence technology to help publishers develop systems that learn from users' behavior to better target content to individuals.
Minneapolis-based Net Perceptions (www.netperceptions.com) has announced partnerships in recent months to integrate its GroupLens personalization product into leading push platforms including Marimba's Castanet, Netscape's Netcaster and Microsoft Corp.'s Active Channels. GroupLens is based on a technique called collaborative filtering, which analyzes the preferences of various users to make recommendations to individuals.
Steve Larsen, Net Perceptions' VP-marketing and business development, said publishers using GroupLens in push applications would be able to track users' interactions with information. For example, the system would be able to track whether users print or forward articles, and then recommend to them other content based on what other users with similar tastes were reading.
While Mr. Larsen said projects are in development, he would not name any customers using the product for push channels.
Customers using GroupLens for standard Web sites include Amazon.com, E!Online's MovieFinder.com and the music site N2K.
Forrester's Mr. McCarthy is skeptical that collaborative filtering holds much promise for push.
"The notion that they're going to figure out who I am from a bunch of mouse clicks is nuts," he said.
His colleague, Eric Brown, senior analyst at Forrester, agreed, saying the effectiveness of the technology is limited to products that are only easily compared to each other.
Collaborative filtering "works really well with the following four categories: movies, books, CDs and Web sites," Mr. Brown said. "After that, I believe the value of collaborative filtering falls off pretty quickly."
Autonomy (www.agentware.com), based in Cambridge, England, uses a pattern-recognition "neural network" technology to understand the gist of written text and thereby target content more effectively than keywords.
"Say you had an investment banker interested in Malaysian teak. A good push system may have categorized its content down to the level of Malaysian stories," explained Mike Lynch, Autonomy's president. "But the banker doesn't care about the country's hundreds of other industries. Not even lumber. Just teak."
NOT YET CLAIRVOYANT
Sites that use Autonomy's software, called Agentware i3, allow users to feed a few articles that interest them into the system, after which the software should understand the kind of specific content the user seeks.
Companies using Agentware i3 for push and Web implementations include News Corp., Barclays Bank and British Telecom. News Corp.'s site, LineOne.net, even uses the technology to serve banner ads according to users' interests.
Promising as the technology sounds, analysts nonetheless warn publishers about believing the vendor hype at face value.
"I don't think the technology has quite reached the level where the application can be clairvoyant about the user's needs," said Forrester's Mr. Brown. "It took IBM 20 years to figure out how to launch a speech-recognition product. They