« More data usually beats better algorithms, Part 2 | Main | The story behind Google's crawler upgrade »


Feed You can follow this conversation by subscribing to the comment feed for this post.

Peter Turney

"This image shows the landscape of Affinity versus Targetability for several forms of social media."

This graph is very interesting, but, since there are no numbers on the graph, I guess it represents your subjective opinion. I think it would be worthwhile to try to get some objective numbers. Reality has a fascinating way of biting back when you poke it.


In cases where affinity is high, frequency is also likely to be high. That might well compensate for low targetability. Also, as you said, it would make sense for social medai to have some topical content, whose viewing is not a function of affinity. Finally, it's important or social networks to realize that friends only exist to bring stickiness to the site. Friends cannot and most likely, will not create good content that can support advertising. Editorially created content might just help.

abhishek gattani

I feel the act of viewing a facebook profile could give a lot of information about the viewer assuming the person viewing is also a facebook user. A facebook user profile exposes tons of information (for instance her interests, friends, background, location, etc.). Won’t this data lead to better targeting? Maybe this data used more effectively might improve targeting.

Does nature of intent (i.e. transient vs. persistent) have a role to play in all this? Social media sites lack transient intent so although they may know a lot about their visitor they don’t really know what she is precisely interested in at a given point in time. They don’t have even have content in the first place to satisfy the different user needs. Instead search engines and transient intent sites such as wikipedia, travelocity, kosmix, etc. meet that need but even those thrive to learn more about their visitors. As you already alluded to (by topic pages) if social media sites start providing content that start addressing transient intent they could improve targeting.

Maybe social media sites are not so great for highly targeted ads but for ads that build brands (e.g. banner ads). Like television channels they could become great platforms for building brand awareness. So targeting may not even be a strong measure to understand value of advertisement on social media sites.

Anand Rajaraman

Peter Turney: I've updated the post with some details of my methodology, and added the numbers to the picture.

Abhishek and Bharath: I don't entirely buy that reach and frequency (i.e, sheer volume) makes up for targetability. In the long run, there will be a huge aggregate number of untargeted page views across the web, and these will all sink to roughly the same CPM levels i.e, close to their serving costs. It cannot be otherwise if CPM rates are rationally determined based on supply and demand, which they will be in the long run, even if they are not in the short run. Therefore, targetability is key to building a sustainable ad-supported business.


Great post. Facebook applications themselves may get a higher CPM (iLike has around $3) than the social network that hosts them, presumably because they are better targeted than ads on the network itself


The analysis is spot on. FaceBook and MySpace are like virtual worlds in themselves and as such should take cues from SecondLife or WoW and start renting 'digital space' to advertisers rather than pushing ads to users. Charge for participation in the network rather than annoying users with annoying display/text ads. Let the 'topical' pages you cite attract users rather than the other way round .. After all people don't live in vacuum .. they are as vocal about brands they like and don't like and products they consume. Its so much part of their identity. Exploit that!!

Peter Turney

"Peter Turney: I've updated the post with some details of my methodology, and added the numbers to the picture."

What I meant was that there are no numbers on the X and Y axes. I know its not easy to do, but "targetability" and "affinity" should be quantified. You say CPM rates are a proxy for targetability, which sounds like a reasonable approximation to me, so the Y axis should be labeled with CPM rates in dollars. Now what can be a proxy for affinity? How can we label the X axis? Maybe affinity could be estimated by the geographic diversity of the visitors to a page? The comments on TripAdvisor are labeled with the city where the commenter lives. I just now looked at some comments on Toronto hotels, and the commenters are all from Canada or the US. Maybe TripAdvisor has a higher affinity than you assume? Now I took a look at Facebook, where "friends" are tagged by the "networks" that they belong to, and these "networks" are mainly geographical. Some of my "friends" are in networks called "India" and "China", although I haven't been to either of these countries. Maybe Facebook has a lower affinity than you assume?

This could be made more rigorous with just a little bit of work. Randomly sample N hotels in TripAdvisor and calculate the average number of different countries of the commenters, per hotel. Randomly sample N people in Facebook and calculate the average number of different countries of the friend networks, per Facebook user.

Anand Rajaraman

Peter T: Great suggestions. Will get on it and report back.


YouTube is being extremely precise and careful about the implementation of the ads. Ads will show up no earlier than 15 seconds into a video, will be overlayed on the bottom 20 percent of a video with 80 percent transparency.

Cliff Lyon

I notice to Colorado tourism folks took down the rate card (PDF) you linked to.

I saved it here

The comments to this entry are closed.