The Accuracy of IP Targeting

Last week I started this series on online ad targeting with a discussion about targeting on http request information. We know that the IP address of the requesting machine is one of the pieces of information that is conveyed whenever an ad request is made. So today we’re going to focus on IP targeting.

IP targeting has been around since the early days of ad serving. It’s not very hard to write code that will strip the IP address from a request, compare it to a database, and deliver an ad accordingly. The true difficulty, as we shall see, is building and maintaining an IP database.

One of the first applications of information in an IP database was targeting to specific geographic regions. Most commercial ad management systems have IP databases that can make geographic targeting possible. However, there are a couple weaknesses in this method. The first (and biggest) problem is that, for various reasons, not all IPs can be mapped to an accurate location.

Take all the IPs associated with AOL users, for instance. Anybody who has seen a WebTrends report knows that all AOL users appear to be coming from somewhere in Virginia. This is caused by AOL’s use of proxy servers to handle their web requests.

In the interest of saving space, we won’t get into the reasons why AOL makes use of proxy servers. The important thing is that AOL does use them, and as a result, all its users appear to be accessing the web from Virginia. Thus, it is impossible to attach meaningful geographic location data to an AOL IP, and those IPs must be discarded from any database that wants to maintain a reasonable degree of accuracy.

Other ISPs and networks may use a method known as dynamic IP allocation for its users. In other words, a user might have a different IP address every time he visits the Internet. You can see how this might affect the accuracy of a database.

But the real difficulty in discerning geography from an IP address has to do with the level of specificity that a media planner might expect from this targeting method. The first few geo-targeted campaigns that I put together early in my career had to be accurate to the ZIP code level. This level of specificity is not practical via IP targeting. Here’s why…

Let’s say your next campaign needs to be targeted to New York City ZIP codes only. You direct the sites in your campaign to use their IP targeting databases to target these ZIP codes. Let’s say that someone who works at IBM then visits one of the sites on your campaign. The ad server gets his request, and figures out he’s accessing the web from IBM’s domain. Great… What does it do now? How does the database distinguish between a New York-based IBM worker and a New Jersey-based one?

This is where the issue of the quality of the database comes into the picture. Whoever maintains the ad management system can contend with this issue in one of three ways:

  1. Chuck the IP out of the database or disregard it

  2. Attempt to ascertain how IBM has set up its network and act appropriately
  3. Fudge it by guessing what percentage of the IP block is in New York, using that percentage as the degree of accuracy.

Thus, as the geographic footprint gets smaller, you can see how either accuracy or targetability can be compromised.

Still, geographic targeting by IP is possible for larger footprints. If you’re looking to target to a specific country, you’ll be able to achieve a reasonably high degree of accuracy greater than 98 percent, according to Mark van der Linden, who works for a company called RealMapping in Amsterdam. His company’s mission is to provide the most accurate and comprehensive information on IP addresses. Mark was kind enough to write me after my last column appeared to talk to me about the pros and cons of IP targeting, as well as point out some non-advertising-related applications of an IP database (the serving of web content in the appropriate language to users, based on their IP addresses, for instance).

While I agree that an accuracy score of over 98 percent is possible with a meticulously maintained database, that accuracy is based on only the IPs for which you can get a discernable location. Remember all those AOL users we were talking about earlier? They’re not present in the RealMapping database. For some applications, it may be okay to skip over these users. For other applications, AOL users make up too high a percentage of web traffic to ignore.

Still, the IP method may be useful outside geographic targeting. Believe it or not, some of the commercial ad serving solutions out there still don’t have the ability to target ads to a specific domain. I dunno about you, but I kinda like the idea that I can call DoubleClick at any time and run a campaign across its network, targeted to, with messaging like “Sorry about that one-day loss of $11 billion, Mr. Gates” and know that Microsoft employees are the only ones that will enjoy it.

So now that we’ve seen the strengths and weaknesses of IP targeting, next week we’ll get into some of the more sophisticated methods of targeting advertising online.

Related reading