Guaranteed Audience Buys: What's Promised Isn't Always What's Delivered
One of the promises of the Internet is the ability to target an exact individual. After all, unlike radio or TV, where messages are simultaneously broadcast to a broad audience, online advertising is served one impression at a time. Theoretically, if you know who the user is, you can perfectly target your message, or choose not to deliver it at all. A trend has recently emerged, whereby agencies can buy a ‘guaranteed’ online audience, in the sense that the people exposed fit specific audience criteria, such as male teenagers, females 18-34, or auto intenders, etc. The sellers are responsible for the guarantee, which they base on what they “know” about their users or visitors. It is however often the case that the level of knowledge about the individual user is less than perfect, and there is a substantial discrepancy between promise and reality. This naturally results in a lot of consternation and finger pointing that is ultimately bad for the industry.
To understand the source of discrepancy, we need to remember that user identification on the Internet is ultimately based on a cookie. If the cookie can be associated with some level of information about the visitor, it opens the door for targeting. The problem, however, is that a cookie is associated with a visitor at a particular point in time, such as a registration or a transaction, but may point to a different person at a later time. For example, you may register as a user of the Washington Post on your home computer, an event captured by a Washington Post cookie set at the time you registered. However, any user that visits the Washington Post from your computer at a later point, whether a spouse, a child or a friend, will be represented to the Washington Post by your registration cookie, which naturally leads the Washington Post to think it is actually you. If a campaign on the Washington Post seeks to target an adult like you, it may be actually reaching a different user on the same machine, which could be a person with very different characteristics – such as a child or teenager.
This situation happens much more frequently than many people realize. To quantify the incidence, we conducted a study on a number of popular sites to assess how frequently a cookie on the site is associated with multiple users. (Note: Comscore identifies users in its panel based on a series of unique biometric identifiers including mouseclick and keystroke patterns and does not rely on cookies.) We found, as the table below illustrates, a site-average of 44% of cookies corresponding to a single user. In other words, a cookie points to a given user with certainty only 44% of the time. An average of 56% of cookies point to multiple users, which means they will point to a different user than the one who originally registered at least 50% of the time. That indicates an overall probability of at least 28% that a cookie will point to the wrong user.
Even if a cookie points to the right user, there is no guarantee that the self-reported demographics of the user are accurate. The degree of accuracy of registration information varies depending on the site and demographic variable. Many people misrepresent their age and income. Some teenagers represent themselves as adults to avoid access restrictions. Many people overstate their income and understate their age on a dating site. Others just enter the wrong information out of privacy concerns. The following example, based on Comscore panel data, shows a handful of individual teenagers, each with conflicting self-reported age across a handful of popular social networking sites. The uncertainty about the accuracy of self-reported information compounds the errors caused by the cookie issues discussed above.
Where this leaves us is that it is virtually impossible to “guarantee” an audience 100%. Based on our experience, a delivery of 80% of the target is best-of-class. More often than not, the real delivery is typically around 50%, depending on the campaign, the quality of registration data, and whether login cookies are used or not.
Rather than setting unrealistic expectations, the industry might be better served with a measure of targeting lift – such as has been used in the direct mail industry for decades. A campaign that promises a 300% target lift means that the targeted audience has 300% the incidence of the target that would be achieved without targeting. These kinds of impressive target lifts are unique advantages for the Internet, and yet they represent goals that can be realistically met.
We strongly urge ad buyers and sellers to start a productive dialogue about this issue. The absence of a common understanding only serves to undermine the confidence traditional advertisers have in using online media -- which is obviously something none of us desire.