I recently mentioned the importance of scale to advertisers. But it’s even more important (and more underrated) for publishers.
In the decade and a half since the birth of the commercial Internet, only two third-party advertising models have proved resilient. Search-engine marketing, as Dan has pointed out, has succeeded because of its effectiveness. Goto.com (which later became Overture, was bought by Yahoo, and whose model was adopted by Google for its more-famous AdWords) figured out that turning search results over to the free market was an efficient way of producing relevant results that benefitted both the user and the advertiser (see, more stakeholder theory). Pay-per-click search advertising has almost single-handedly made Google the powerhouse it is today, though it is interesting to note that whereas AdWords (which places ads next to search results) has proved a successful revenue generator for Google, AdSense (which runs those same ads alongside targeted content on third-party sites) has been far less successful for publishers.
The second model that has withstood the test of time has lasted not because of its effectiveness but in spite of it. For years, we’ve been hearing about declining click-through rates on online banner ads, and also about how click-through is a poor metric on which to focus. In other words, banner ads are delivering the wrong thing, and they’re not even very good at that.
So why does the model persist? Because it’s scalable. Banner ads are a commodity, and commodity businesses require volume. There is simply no form of online advertising easier to set up than an untargeted banner campaign. Holy grails such as behavioral targeting or customized ad campaigns have never managed to break through; because they are complex and difficult to implement, they require committed, “smart” advertisers. But there aren’t enough smart advertisers to ensure a workable revenue model. You need the dumb ones as well.
The day may eventually come when banner advertising finally runs its course. But it won’t happen until a more effective, but equally scalable model is developed to supplant it.
