Behavioral targeting has been described most definitively as moving the needle in advertising from the page to the user. In the discussion below Julian Steinberg, vice president of operations and strategy at Inform Technology, cautions against setting behavioral and contextual apart from one another. In fact, he suggests, deeper understanding of consumer behavior begins with deeper understanding of the page.
Behavioral Insider: How do you see the work you're doing with news aggregation and search relating to advertising and targeting?
Julian Steinberg: What we're doing is structuring currently unstructured data in such a way as to create more inventory by helping publishers, especially news organizations and newspapers, to more deeply engage their readers. What we do is automate links from within news articles, allowing the reader to get further information on as many facets of the subject they're interested in.
We've been working with many well known newspapers such as The Washington Post, as well as newer and smaller publishers. For the New York Sun, for instance, we embedded links and extract topics and related articles to give readers a means of penetrating as deeply as they need or want on a topic within the article they're reading WITHOUT having to leave the site. The Sun increased page visits and time spent on the site by 15% to 20%.
The interesting thing is that expanding page visits means not only a quantitative jump in the amount of targetable inventory, but a major qualitative leap forward, potentially, in the understanding of each particular reader's interests.
BI: Can you explain the difference between advanced semantic analysis and keyword links as we know them?
Steinberg: The way I'd illustrate it might be to take the example of a passage in an article about an electric car seat. A blind CPM ad might run a banner for an electric wheel chair for a nursing home, just completely misunderstanding the context and what's relevant about the material from a reader's point of view. The next generation beyond that was the keyword, which of course is a massive improvement. There you'd have the intelligence to associate the article with such ad topics as 'cars' or maybe even refined enough to pick up on particular brand names like 'Toyota' or 'Ford,' but it remains pretty generic.
Where you go from there is to refine systems able to tag content to get a wider understanding of how keywords relate to what the major themes of the article really are. This means the system will understand that we're not only talking about cars or a particular brand, but about electric or hybrid engine cars. By providing what we call a fingerprint analysis of the article it will also understand that we're talking about a luxury vehicle even though the phrase luxury vehicle may not be in the text itself.
BI: I can see how that enriches content targeting, but what about the behavioral component?
Steinberg: Once you've got a stronger handle on what your content is, it opens up new dimensions for more precise contextual targeting. But a step beyond that is being taken -- which is to personalize the delivery of content by reader interest.
BI: How does RSS work into this?
Steinberg: RSS is enormously important, but it's used today mostly to deliver entire publications or sections of publications (e.g. sports).
The technology can be far more customized than that, however. A reader can say they are interested in hybrid vehicles, and the reader will scan all news articles related to that subject or a combination of subjects of interest.
Once you really understand the page, you can finally begin to understand who the user is. Most of what goes under the name behavioral targeting doesn't attempt to understand, either. It will take a reader who's read an article on hybrid cars and automatically assume that they are a car buyer. But that isn't necessarily the case. It might mean they are in the market for a car, but it's just as possible they are an environmentalist whose interests are in eco-friendly products. It might mean they are looking at luxury items. All of which suggest different behavioral profiles and strategies. If you don't understand the content on a deeper level you're not even going to begin to know how to ask, much less answer, these questions.
BI: Any thoughts about how your approach relates to video content?
Steinberg: We're approaching video with a similar sort of semantic framework. The challenge with delivering video ads is quite similar to that for text. It's first and foremost to understand what the content is really all about. The technology we use is designed to extract the maximum amount of understanding of both semantic meaning and user interest. For video content, audio and blogs, we can do that by associating text with the video files. There are technologies in development which actually translate video into text. As long as you can tag video information with an accurate textual description, you can target it as ad inventory.
BI: Where do you see semantic analysis being deployed over the next six months or so?
Steinberg: Our premise over the next six to 12 months is that online publishers are on the cusp of an important learning curve. In so many cases content owners today don't understand the unrealized value of their own content. A key role for technology right now is to unlock that value.