Power of Suggestions Forges New Web Model

Aggregate Knowledge Recommends Products Based on User Behavior

By Published on .

NEW YORK (AdAge.com) -- A few years ago, Paul Martino had a problem. Tribe.net, the social-networking website he co-founded before MySpace made the term a household word, was struggling to help its users find dating partners. Mr. Martino and his partners tested a few different solutions -- looking at the profiles members surfed and the people in their own or adjoining networks -- before arriving at a decidedly unsexy conclusion: The best way to hook up Tribe members was to match them based on the news stories and message boards they read.
Paul Martino: 'We're not trying to match ads to people; we're trying to match content to people, content that they'll care about, engage with and love.'
Paul Martino: 'We're not trying to match ads to people; we're trying to match content to people, content that they'll care about, engage with and love.'

The realization that serendipity can be a guiding principle for how people choose their mates is now at the heart of one the most buzzed-about companies in Silicon Valley, a firm that's going to shake up how retailers do business, how web publishers share content and, quite possibly, how consumers use the web.

Aggregate Knowledge, started two years ago and backed by legendary venture firm Kleiner Perkins, is founded on the notion that what you read on a newspaper's website can actually inform what you're likely to buy from a retailer somewhere across the internet -- provided, of course, there's enough data compiled to bear out the recommendations. It sounds simple, even obvious, but until now it hasn't been done, at least not on the scale Aggregate Knowledge is working on.

Twelve companies -- including retailers such as Overstock.com, web publishers such as WashingtonPost.com and some as-yet-unidentified social networks -- are now piping vast amounts of anonymous buying and browsing behavior into what Mr. Martino, in technophobe-friendly fashion, calls "a pot." In coming months, that pot will allow a publisher to recommend to a reader of a story on, say, ergonomic chairs, both articles that other readers of the same story looked at and products that they bought. It'll even allow some media partners to stray into the until-now-virgin territory, at least for old-line content players, of recommending news content from competing sites. The idea is not to create another ad network but to use crowd behavior to structure the web experience.

Matching content to people
"Users don't like ads much, but if they view a recommendation box as something that's been helpful, they'll come back to a site," said Mr. Martino. "We're not trying to match ads to people; we're trying to match content to people, content that they'll care about, engage with and love."

Aggregate Knowledge is one of a diverse score of companies including the likes of Digg, Reddit and StumbleUpon that serve what media expert Tim Hanlon calls the need for "guidance and navigation" in a media world characterized by enormous, underorganized piles of audio, video and text content. Until now, Google and its ilk have been the alpha and omega of navigating those piles, but the old search box only gets you so far if you're not sure what keyword to punch in.

"Search implies you know what you're looking for," said Mr. Hanlon, senior VP-ventures at Publicis Groupe's Denuo.

Fortuitous relevance
Discovery, moments of fortuitous relevance -- not search and not ads -- are what Aggregate Knowledge is delivering. And not so serendipitously, it's helping retailers add incremental revenue and content sites to keep visitors around. On Overstock's site, it operates a recommendation engine that's already helped the retailer with its cross-selling efforts. The simple "People who bought this also bought" window that pops up on product pages was responsible for 20% of the company's holiday sales, results that earned Mr. Martino's company a shout-out from CEO Patrick Byrne on earnings calls.

The Washington Post's website has a similar feature that recommends relevant stories. Caroline Little, publisher-CEO of Washington Post.Newsweek Interactive, said it's too soon to tell how its recommendation window will affect page views or time spent on the site. "It does look very promising," she said. "We say we're Amazon-ing our site."

At this point, neither Overstock nor the work done for The Washington Post truly capitalizes on the scale of computing that Aggregate Knowledge brings to bear -- countless "predicates" that relay simple bits of behavior such as "User 27 added Item 83 to his cart at 4 o'clock" or "User 28 looked at news story 47 at 6 o'clock" collated into a recommendation system based on the behavior of large groups of people. Pick your metaphor to describe it: "Wisdom of Crowds" thinking meets the algorithm or an Amazon.com-style recommendation engine on steroids, blown out across a network of sites that gather vast amounts of browsing and buying information.

"That network is really what the magic is," Mr. Martino said in an interview. "If you just have a recommendation box on your site, you're missing out on the magic of everyone's collective experience."

The supercomputing pot
In the days since it came out at a January tech conference, Aggregate Knowledge has aggressively publicized the Overstock case study, but Mr. Martino is more cautious in talking about how his service will work when the collective discovery network -- his supercomputing pot -- emerges from testing in coming months. Most disruptively, it could push media companies into recommending competitors -- perhaps trading the sacrosanct integrity of a website's boundaries for more trust and loyalty from consumers who, as Mr. Hanlon put it, "don't care about walled gardens."

He's cagey on how this will take shape, but said, "You're going to see that with a whole lot of media partners over the next several months." Ms. Little said her outfit is considering expanding to recommend products and content from other sites, provided the context is right and the relevance is there. "You're not going to talk to just one person at a party," she said. "If you think of the internet as one big party, we want to have some role in introducing them to people and helping them sift through all the information out there."

Of the dozen companies he's working with, Mr. Martino has revealed only a few -- in addition to WashingtonPost.com and Overstock, ABC and NBC stores managed by Delivery Agent. He said the rest are companies that rank among the 200 most popular websites as ranked by Alexa. In addition to drafting other online stores, publishers and social networks, Mr. Martino's firm is doing deals with web-service providers to make the sites more engaging. Today, Aggregate Knowledge is announcing a partnership with Bazaarvoice, a company that manages online customer reviews for massive retailers such as Sears and Dell.
Most Popular
In this article: