Great post here by one of the more vocal and well-read venture capitalists in the blog community, Fred Wilson. He does a good job of starting to explain why it makes sense for an ad network to roll-up various web sites and parse out their inventory, as compared to a more traditional in-house sales model. As behavioral, contextual, profile, and related data contribute more and more to advertisers' campaigns it makes more sense that the company selling and empowering those ads would sell across and have access to and data from a number of sites. Significant chunks of the major players (or even smaller sites) inventory which is not inherently very attractive or targeted to advertisers that is sold for relatively low CPMs can often be parsed out by a 3rd party which has user data the primary site does not, at a significant premium.
Consider an ad that appears on news.yahoo.com. The average user visiting this URL doesn't skew in any demographic, psycographic, or related direction widely, and you would be hard pressed to sell this ad spot at a premium. If a 3rd party however knew that a particular user on this page had visited, say, edmunds.com recently to view reviews of a new SUV, that 3rd part could potentially sell that ad spot to an auto manufacturer at a significant premium to the standard rate.
Yahoo could indeed potentially bring this type of targeting in-house to their sales force, but it just makes more economic sense that the sales organization would work across multiple sources of inventory given where the variable costs are in this model.
