33Across: The next generation of behavioral ad targeting

Like its competitors, the start-up is trying to help social networks such as Meebo goose their ad sales by identifying (and studying) members that are so-called influencers.

Stefanie Olsen Staff writer, CNET News
Stefanie Olsen covers technology and science.
Stefanie Olsen
5 min read

Despite having access to a treasure trove of data about people's hobbies, demographics, and friends, social networks can find it tough to sell banners ads for more than the price of a stick of gum.

So ad start-ups such as 33Across are trying to take advantage of profile data on social-network members to create the next generation of behavioral ad targeting. Or behavioral ads 2.0. (For a portrait of another start-up targeting the intersection of social networks and advertising, with an announcement due Monday, see "SocialMedia to unveil 'friendship ranks'.")

New York-based 33Across has developed an analytical engine that can look at behavior patterns of members of a social site in order to track who, in the so-called social graph of friends, is most influential to others. It examines things like how many messages a person sends or receives, how many people he or she has befriended, or whether that person tags photos, blogs, or forwards links to friends.

With that information, it can pinpoint who are so-called viral propagators, or the people most likely to "start a viral cascade" about a product or service, according to CEO and founder Eric Wheeler, a former ad executive from Ogilvy.

"We understand where a person sits among their friends and friends of friends...and the likelihood of how viral they would be," Wheeler said.

Meebo partnership
In recent weeks, 33Across signed its first major customer, Meebo, a social site that synthesizes multiple chat applications into a single browser interface.

Martin Green, Meebo's vice president of business, said the company is using 33Across to provide better analysis and research for advertising clients such as Universal Pictures. It is not using the 33Across self-service ad technology to target ads yet, though it is talking to 33Across about adding that functionality.

In a nod to potential privacy concerns, Green said Meebo is prepared to disclose its practices to members and give them a way to opt out.

Under the deal, Meebo gives 33Across information about its roughly 36 million monthly active users, excluding names, phone numbers, and addresses. With that data, Meebo and 33Across can see how a marketing promotion spreads throughout the members of its service by looking at which members are sending links to friends on the The Incredible Hulk, for example.

Meebo can study those users' demographics, how many "buddies" they have, whether they're influential among a group, and whether they typically shy away from sending links but are apparently inspired by a certain campaign.

"We're giving advertisers more information on how their ads are pervading the network," Green said. "We're looking at the nature of the people who shared that studio's trailer or content with friends. Are they (the) same 'connectors' in the network? Or did you appeal to people who don't normally act on promotions?"

Whereas most advertisers pay as low as five cents to reach a thousand viewers on an ad network, Meebo is charging a list price of $12.50 per thousand impressions using 33Across' technology, Green said.

Tracking social patterns
Technologists have long talked up the ability to track people's behavior across the Web, create individual profiles, and then better target ads for higher rates. And the trend took off with behavioral-ad start-ups like Revenue Science and Tacoda Systems, the latter of which has since been bought by AOL. Now, with the rise of social networks, a new crop of companies is trying to mine even more profile data about people by analyzing their social patterns.

33Across, which has raised more than $1 million from backers such as First Round Capital and former Tacoda CEO Dave Morgan, eventually plans to use data on people's relationships and influence in social networks to better target ads to them on mainstream sites.

33Across influencer map
33Across' Influence Map shows more influential people in warmer colors (red, orange, yellow) and less influential in darker colors (blue). 33Across

Wheeler said 33Across cross-site tracking would work like behavioral ad targeting, in which partners of 33Across would place a tracking pixel on their page and that piece of technology would call its servers to place a cookie. The cookie would identify that person without linking his or her name, address, or phone number. That way, 33Across could target an ad to a person on a hypothetical partner site such as eBay, based on the behaviors at Meebo, for example.

Wheeler said today, people don't need to opt out of 33Across' site-wide ad program because it's not in operation yet. But the company plans to announce new partners in the coming weeks, and that could prompt the start of its cross-site tracking and ad delivery system. When that happens, he said, 33Across will give people the ability to opt out of the service at its Web site. (He did not say that program would be imminent, however.)

So how does all this friend and influencer targeting work? Much of it is based on machine-learning algorithms for social networks that have yet to be proven.

One issue with this technology is that it can be hard to track down who's most influential in a group. That person may be influential among friends, when it comes to autos, but he or she might not hold sway, when it comes to travel. Pinpointing expertise can be tough, and influential people might be the same for each category.

And then there are the privacy qualms.

"The issue with social networking and advertising is largely not a technical problem; it's a cultural one. When you're out there with your friends and interacting, people are somewhat resistant to ads. It's a different context, and people can get pissed off," said Jeffrey Davitz, program manager and director of the social-computing group in SRI International's Artificial Intelligence Center.

At SRI, Davitz has researched machine learning in social networks as part of a multimillion-dollar project funded by the Defense Advanced Research Projects Agency, or DARPA. Specifically, he developed an application that could automatically monitor people's interests and influence in military communities such as Company Command (for captains).

The idea was to identify influencers who care about specific topics, such as attacks involving improvised explosive devices in Iraq, and then ensure that they see relevant information in a news feed to that topic, such as an officer posting a document to the site pertaining to explosive devices.

That's what's called "smart push," he said. The technology is currently deployed on three military sites, but SRI is looking at commercial applications for it, not related to advertising.

"You clearly can learn more about people from MySpace and Facebook," said Davitz. "The question is whether or not people will accept that kind of advertising. People feel it's kind of creepy," he said.