Econometa

I just read an interesting story about technology from a startup called Invidi that delivers targeted ads to cable TV viewers. The targeting is based upon user behavior, and like Tivo, the system maintains privacy by never allowing behavioral data to leave the digital set top box (DSTB). This means that any personalization has to take place “locally” (in the DSTB).

Tivo uses behavioral data locally to make recommendations against the program guide, which is also locally stored. In contrast, Invidi wants to use this data to select ads, which are *not* locally stored. Their clever solution, which also fits in well with the cable architecture, is to add “channels” of ads designed for specific user segments, then locally decide on which segment best fits the user’s behavior. As I understand it, the integration happens essentially by covertly switching channels during ads to the appropriate “ad channel.”

Following up on my previous post, this idea of “local personalization” is interesting because, being both locally stored and remotely inaccessible, my data is presumably more under my control than if it was either remotely stored (as with Amazon) or remotely accessible (as with cookies).

The problem, of course, is that the system only works if the personalization is very coarse; for ads, this means a small number of user segments, since every ad on the net can’t be sent to each user. Something closer to Google AdSense, which essentially segments users according to all possible search keywords, wouldn’t work: if ads appeared matching the arbitrary terms you searched for in the program guide, it would be clear that the cable operator “knew what you watched,” and this raises privacy concerns because they also “know who you are.”

It’s interesting to compare this situation to the web. Web ads can also be targeted according to user behavior; the big difference is that behavioral data on the web is carried in a cookie, which is remotely accessible to the publisher but anonymous; DSTB data is deliberately not remotely accessible, because it is necessarily not anonymous — the cable company knows your name, address, phone number, etc. and the ID of the cable box you have. This problem is avoided on the web because your ISP and your content provider are two different entities, and the ISP has ended up with the implicit duty to not sell your account data to content providers. So an ad network might “know what you surf,” but that’s OK because they don’t “know who you are” (and you can change what they know by deleting your cookies).

So it turns out that by controlling the entire system and thus not being able to safeguard the anonymity of users like ISPs do, cable operators are forced into a model that is less powerful, since it can only personalize locally. This becomes even clearer if we consider personalization beyond ads. For example, the new Google Personalized Search, like A9, stores your search history and then personalizes new results (and presumably ads) based upon this history. There’s no way I can see for this to be done locally; the search algorithm must have direct access to your history.

On the other hand, ad targeting on both the web and Invidi also uses available demographic information. On the web this is limited to things like connection, browser, and geographic location; but for cable TV user data is much richer, since they have exact address, account info, and full U.S. census data to draw upon. So as with direct mail, cable operators have the advantage of publicly available data about you, while web publishers have the advantage of interactivity and behavioral data.

It’s interesting to consider the ways this situation could change:

– Storage and bandwidth could increase to the point that it *was* possible to send every ad to every DSTB or browser. This would allow local personalization of ads (but not other services such as search, unless the entire Google index could be locally stored!)

– Cable operators could imitate ISPs, and allow open but anonymous access to DSTB data to third parties for personalized services and/or ad targeting. This would be a huge change, and one that seems easiest to implement by just making the TV a web device, like WebTV.

– ISPs, who know your IP address, could imitate the cable model, and sell access to your demographic information for targeting purposes. I’m actually surprised this isn’t already happening (maybe it is? hope not!).

The privacy that we expect seems to be largely driven by past history. Credit card companies and the bank know what we buy, but we expect them not to share it with others. Cable TV evolved from broadcast TV, and so we still expect not to have our viewing monitored. We’re used to the Internet being anonymous and user-controlled and don’t want that to change. Many of these issues were “decided” by technological limitations, but it seems to me that as the dreaded “big database in the sky” becomes increasingly possible (and profitable), a coherent and consistent public policy on personal data becomes increasingly important.

This entry was posted on Wednesday, June 29th, 2005 at 9:39 pm and is filed under Media, Personal data. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.