• by Olly Downs , May 20, 2008

Can we agree that there is nothing more frustrating than discovering something better, cheaper or more attractive than the item you just bought? Thanks to the power of behavioral targeting, I had this experience recently after buying a car. The ad engines rapidly caught on to the fact that, over a period of about two weeks, I was in the market for a new family car.

During each online session, I was presented with ads from various car manufacturers and even ads from dealers in my area, which, at the time, seemed quite helpful. But, nearly six months later, I'm still being hit by the same kind of ads. Of course now they're heavily featuring a model that was not available when I was shopping, which trips my buyer's remorse trigger every time I see it! Will someone please get the hint that I'm not in the market for a car anymore?

Clearly my experience isn't unique, and as a guy with a bent for analytics, it got me thinking: How might you identify transient targets more effectively? Behavioral targeters look for a given user's consumption in specific categories that is "over-indexed" relative to the overall Internet population - like automotive classifieds, in my case. However, they clearly need to improve their ability to determine which behaviors indicate constant, rather than transient, traits of the user, so that the targeting criteria are relevant to both the user and the advertiser.

Using data from a recent Nielsen Online study, I spent some time developing targeting criteria using observed Internet consumption across a taxonomy of categories commonly used by behavioral targeters. I was looking specifically to understand how transient specific behavioral trends are.

Looking at the aggregated weekly surfing behavior of several thousand individuals by category over a three-month period, I found it easy to identify users with transient, atypical bursts of activity in certain categories. For example, one consumer (we'll call him "User X") exhibits a surge in the "Automotive" category starting in Week 3, culminating in a major burst in Week 5, followed by a spell of no activity in the category for the subsequent seven weeks. The burst in activity is significantly overindexed relative to the baseline Internet consumption by category of that particular user, and relative to population-wide consumption in the category, suggesting a major atypical episode, which behavioral targeters love.

So is User X, who was pretty clearly in the market for a car in weeks 3 through 5, still in the market for a car in Week 12? I doubt it. Is User X still an "auto enthusiast," as his burst of automotive content consumption might have led a behavioral targeter to classify him, or was his enthusiasm limited to the three-week window in which he had to choose and buy a car? The answer seems obvious, and yet a behavioral targeting engine might well have drawn entirely different (and incorrect) conclusions about the nature of User X by observing his transient behavior out of context.

Behavioral trends vary rapidly, whereas other user attributes, like attitudes or psychographics, change slowly. The challenge of behavioral advertising is to try to address both fast-changing and enduring attributes. What can we learn in the short term that allows us to deliver "right-now relevant" advertising to the user, and what can we learn in the long term that allows us to deliver messages of ongoing broad relevance to the user, regardless of what the user is doing right now?

As advertisers look for deeper insight into behavioral targeting methodologies, and as choice proliferates, the winners will be those who can communicate how the data, measurement and metrics they are employing trade off the impact of transient behavior and long-term attributes of individuals. Insights like that will generate always-relevant and hence high-value targeting.

Olly Downs is chief scientist at Mindset Media. (odowns@mindset-media.com)