Multi-Dimensional Targeting: Merging Contextual And Behavioral

40 or so years ago, two mostly distinct groups of marketers stood at the cutting edges of an advertising world. One looked to systematically segment the U.S. market by the then- exotic notion of population demographics. The other looked to exploit the zip codes the postal service had introduced in 1963 to develop a new kind of targeting, direct mail, based on geography. Few at the time were ready to see the possibility of merging or even meaningfully relating these two methodologies. Eventually, of course, the notion of a new synthesis using zip codes to target unique demographic "lifestyle" clusters would become self-evidently obvious to another generation of marketers. Peter Bordes, CEO of ad network AdValiant, and AdValiant's managing director, Jivan Manhas, believe a similar evolutionary synthesis, this time between contextual and behavioral online targeting, may be in the offing,     

Behavioral Insider: Proponents of online targeting seem to be dividing into either the contextual or behavioral camp. Where does your ad network come down in this debate?

Peter Bordes: We think the whole debate is a straitjacket we, and our advertisers and publishers, want to get out of. One side is thinking of consumers in terms of where they happen to be at a given moment online. The other, in terms of where they've touched. But both [strategies] are one-dimensional. Neither approach by itself is getting to the key question of who consumers are.

BI: What are the key limitations of online targeting as it's conventionally presented?

Bordes: From a pure click analysis, imagine a man who visits fishing and finance sites, but also buys his wife's feminine products online. It makes perfect sense from a pure behavioral perspective to start serving him tampon ads everywhere he goes. That's basing advertising on what he's touched online, but not really who he is. Then again, if he visits fishing sites, it makes perfect sense from a contextual point of view to deliver him ads for fishing equipment, but if you bombard him with nothing but outdoor ads you're limiting yourself to just one dimension, which equals endless repetition and clutter. So pure behavioral and pure contextual are limited models.

BI: How can analytics be reframed to avoid these cul de sacs?

Bordes: The question is, how to use the data about where you go to address the more important fact of who you are. The key is really to learn to model based on how people interact with contextual ads in unique patterns.

BI: Doesn't that involve capture of personalized data most people rightly consider off-limits?

Bordes: No. It's more of a shift in the way you look at anonymous data. Our goal is to look at things more sociologically, to profile in terms of similarities in patterns of interactions with content to identify and group together specific types of consumers based on anonymous behavioral and contextual data. To hark back to the analogy of zip code clustering, if I know someone lives in Princeton, N.J. and likes to go to Nantucket in August I have a good idea of the kinds of places they might like to go in the winter, based on their similarities with others who come from where they do, or places that fit the same profile as theirs.

BI: Can you provide a more detailed sketch of how anonymous ID can work? And any examples of how an advertiser might deploy it in a promotion or campaign?

Jivan Manhas: Anonymous ID profiling works as a means to connect the current Web site visitor to anonymous profiles that we have collected of other visitors to the network of sites that host the AdVario embedded content and traditional banner ad serving technology. [It works] by tracking users' surfing behavior, the traditionally served banner ads they clicked, the articles/content they read, the content-embedded ads they viewed and clicked on, the embedded content ads they viewed but did not click on, etc. All this information allows our technology to connect the site visitor's behavioral patterns and anonymous profile (which we build of individuals through their surfing patterns/behaviors) to anonymous profiles of other previous visitors to our network of sites. We do not continue to track users once they leave a Web site that is not powered by us.

We are constantly correlating points between various profiles--or possibly just a single profile--to the current site visitor. This allows us to better target advertisements to the visitor by various means, i.e. we noticed they clicked on more content-embedded automotive advertisements than any other type of ad. The next site the user goes to within our network, we can serve complementary or similar advertisements as regular/traditional banner ads, even if the content of the site is not automotive related. The content itself can serve the embedded ads that are relevant to the content of the site. If we find the profile/behavior of the surfer is very similar to an anonymous profile we have in our system, we can serve advertisements similar to the ones we served to the anonymous user on file that had the top click-through rates or top impressions. Since we are taking a more individual approach as opposed to an aggregate "profile/behavioral" approach when it comes to ad serving relevance, we feel it allows for better ad targeting, higher click-through ratios and a higher-quality user.