What's 'Peer'ing Around the Corner

by Melinda Gipson

September 26, 2008
What if you could apply the same principles of "search" to contextual display ad serving without the platform? For efficiency's sake, you probably need both, but an Israeli-based company called Peer39 is pushing at the edges of contextual advertising and re-opening the discussion about whether you really need to track user behavior to know what interests them and put relevant ads in close proximity.

President Amiad Solomon told me recently that the better the semantic understanding of both content and advertising targets, the better the match can be, even without the back-channel confirmation of user behavior. Solomon gives all kudos to Tacoda for its efforts in the realm of BT -- and has certainly hired enough former Tacoda employees to prove his admiration. Listen to some of his thoughts about the future of display advertising -- which were NOT shared in any context with the Yahoo! APT platform, btw; this is my own reflection -- and see if you don't spot some key differences.

Solomon: We believe that, using our technology which understands both the meaning of concepts on a Web page and whether the advertising is a good fit "we'll be able to track extremely precise matches." Tacoda's BT aproach was laudable, but "we see a huge opportunity to grow advertising targeted in a different way." The company already has several years of work under the hood and some 50 employees, 44 of them in R&D.

The technology has grown out of eBay, shopping.com, academia, and the seemingly ubiquitous intelligence research of many Israeli companies. The company's chief scientist hails from a background from Princeton's Natural Language Processing labs on text analysis; "one thing we're expert in is that we can take any text on the fly and understand the meaning of the page," i.e. when the words "trip" + "camera" mean something other than travel. Imagine advertising linked to a real semantic cookie that can targete readers at the page level. The way it works is, "we segment to the page and match supply and demand at the page level on the Internet at large." And the machine learning built into the system automatically updates it's understanding of how to make those matches. "We don't believe in building rules [for content categorization] -- the taxonomy isn't scaleable that way."

As an example of what Solomon means, think about the words "Barack Obama" -- these words wouldn't have meant anything a couple of years ago. Peer39's categorization engine crawls the whole Internet -- "the Internet is our training set" -- and it built its taxonomy on what's there. Using four characteristics of the net -- Users + Interests + Volume + Words -- Peer39 could, in effect, build a "network" of content with no single ad server at all, and without cookies. "We're bringing in revenue, and we help the user experience too, since most ads now aren't necessarily relevant," he explained.

The company has partnered with some publishers it can't yet reveal, and is actively soliciting more partners. It's not yet clear whether brokering sites' inventory to advertisers or revenue sharing with publishers themselves will be the best business model, but it's sure that agencies in New York, San Francisco and Chicago will be interested in what it finds. And the reason they will be interested, Solomon says, almost casually, is because the company is on track to have a reach of more than 1 billion impressions by year's end. "For the first time, advertisers can have access to a user generated content site," without worrying about their branding exposure, because they'll know not just what the site's content is about, but whether the sentiment expressed on those subjects is positive or negative. The same technology can open up more business-to-business inventory, because its natural language processing can be publisher agnostic. "We can know who is reading what content, and target advertising with a very good result," Solomon asserts.

"Fine-tuning semantic advertising will lead tot he next generation of advertising," he added. It's much more than search, "you can't just plug in a credit card and get these types of results; even search campaigns can't segregate what kinds of users you're targeting," but Peer39 can, he said.

Which is "better" or more productive for advertisers, only time will tell; they'll probably try both. Hopefully publishers will too, if only to hedge their bets on who's got the better mousetrap.