Last week, we talked about the ways that advertising sites collect information that goes into developing profiles. This week, we want to focus on the profiles themselves.
The power of profiles is that they enable personalization. (Yikes! Another word that requires a definition.) Simply put, personalization is the ability of a site to change what a user sees based on previous information about the interests and preferences of that user. For example, when you go to Amazon and are greeted with a book or music recommendation, this is personalized web content created especially for you.
On advertising-supported sites, the promise of profiling is the ability to serve ads on a more targeted basis. In the ideal world (at least from a publisher and advertiser’s perspective), when I arrive at a site, my profile would be so robust that the ad server would know exactly what ad to put in front of me.
This, of course, would translate into fantastic results for the advertiser thus justifying higher revenues for publishers. And theoretically, the consumer would like it as well because only ads of relevance to his or her interests would show up, decreasing the perception of junk and increasing the perception of value. (Yes, don’t worry, we’re going to get to the privacy issues pretty soon.)
That’s the theory of why profiling should be so great in advertising. Now for the reality.
In truth, personalization in advertising is not directly analogous to our Amazon example. Amazon can use my actual buying behavior plus a technique called collaborative filtering to create content that is personalized only for me. In advertising, it is not yet feasible to create truly individualized ads on the fly for each new user that comes along. So, instead, most profiling in advertising works in the sense that it helps to put people in different buckets so that the chances of serving them a relevant ad increases.
For example, if my hypothetical profile says that I have visited etoys, parentsoup, and babycenter.com in the last week, chances are pretty good that I have children, and I get put in the “parent” bucket. Therefore, it would be a good bet to serve me an ad with a child-related offer. In this situation, we’re taking an educated guess that past web activity is an effective surrogate for interest much as we assume that the keywords used in a search indicate subject-matter interest. Essentially, it’s a way to improve the odds that your ad will be seen by an interested viewer.
The longer term promise goes beyond making assumptions based on past behavior and moves into the realm of predictive modeling. In theory, if you were able to observe enough users and their web-surfing behaviors, you could discern patterns about not only where they have been, but where they are likely to go next.
So, if you were tracking a user that checked out a gaming site, collegeclub.com, and suck.com, the model might be able to predict that he or she is likely to visit an MP3 site in the next 15 minutes and automatically serve an ad for the site. And, again, in theory, this would provide a big boost in response for the advertiser and revenue for publishers.
Offensive from a privacy point of view to many? You bet. When the line between helpful and intrusive is crossed now that’s fodder for multiple columns in and of itself there will be increasing warranted pressure (in our opionion) to limit this type of cross-site tracking. But data mining in one form or another is here to stay, so it’s best to be aware of the possibilities.
Does profile-based advertising work? Good question. We’ve heard anecdotal case studies all over the board from wildly successful to a total flop. What are you hearing? Send us your experiences, and we’ll share them with our readers next week.