On Behavioral Targeting

Standard

On Wednesday, ClickZ reported on Return Path’s (relatively) new behavioral targeting functionality. Inasmuch as I work at Return Path and am one of the people who worked on designing this tool, I’m really happy to see it out there in the wild and getting some attention; as regular readers have come to expect, however, there are a few additional thoughts that I want to toss out there.

The first is an observation that I would have though was obvious: behavioral targeting is a technique that can be applied to many advertising products. The title of the ClickZ article is “Behavioral Targeting Applied to E-Mail,” which seems to imply that this is a weird, radical idea; given that marketers have been applying behavioral targeting to every ad mechanism that they possibly can for decades, adding email to that list doesn’t seem like a great leap forward. Not to say that it’s easy to do, of course, but the idea of applying behavioral to email doesn’t strike me as revolutionary.

There probably isn’t much reason to be surprised by this view, though, when you consider a report from eMarketer (found via Fred Wilson’s post) that shows “behavioral targeting” being compared to things like “email house lists,” “text link ads,” and “popups.” Behavioral targeting is…well, targeting, which must be applied to something; that something may be email house lists, text link ads, or, um, popups …sigh…

The second item on our menu for the day is a couple of thoughts on behavioral targeting itself. As the attention given to behavioral targeting increases, we can expect to see lots and lots of people offering it in one flavor or another. The question then becomes: what exactly do you mean when you’re talking about behavioral targeting?

The most obvious approach to BT is also the weakest: somebody has responded to a travel ad, so you flag them as a “travel responder.” You then throw travel ads at them wherever they go, hoping that they’re actually planning another trip soon. Not so good. Maybe they clicked on that travel ad because their grandmother was looking for a cheap flight, maybe they’ve burned through all their vacation time for the year and won’t be travelling again until 2007…you just don’t know.

That’s the weird, counter-intuitive thing about behavioral targeting: in order to target more effectively to individual people, you have to look at groups of people. That’s why god invented statistics. If you’re looking at a single individual it’s really hard to accurately predict what they’re going to do until you have a huge amount of data about what they’ve already done.

On the other hand, it takes somewhat less data to decide what groups that individual belongs to; if you get your groups correct (which, again, isn’t necessarily an easy thing to do), you can then predict with a mathematically defined degree of confidence how I might behave when presented with a travel advertisement, even if you have no information on how I personally have reacted to travel ads before.

As a final note, it’s critical to remember that none of this means that the people who are collecting the data know better than that actual people on the receiving end what is appropriate and interesting. Ideally (as in the case of PMD/RP’s behavioral targeting), BT is a technique that supplements — not replaces — targeting based on people’s explicit requests for information.