Underscore Marketing's president talks about the difference between declared and observed data, and how that can affect your geotargeting.

As the internet economy heats up again to the approximate temperature of the surface of the sun, quite a few aspiring publishers and newer content sites have asked me about how to target their advertising geographically.

Geotargeting is one of the most commonly used targeting options, so regardless of whether you're an ad seller or an ad buyer, you should be familiar with geotargeting and how it works. There are a variety of methods used to determine the location of a web page viewer, and not all methods are created equal. If you're a seller, you want to offer the most accurate targeting possible to your advertising clients. If you're a buyer, you want to ensure that the targeted ads you buy are consistent with your client's expectations for spill (ads served into markets other than the ones you're targeting).

Over my career, I've worked with a number of clients who are not very spill-tolerant. That is, the overwhelming majority of their ads needed to be served within target geographies due to government regulation, limited geographic penetration or other important business factors. For example, insurance companies may not advertise in certain states because state laws make it too restrictive to do business there profitably, or telecoms might need to advertise only in certain area codes where they can offer service. In cases like these, accuracy is key.

So let's talk methodology. The first thing to understand is that most content sites and networks offer geotargeting through multiple methods. For instance, a content site that needs to serve an ad to someone in ZIP code 11792 will first check the user's registration profile for a ZIP code. If no ZIP code is available, the site's ad server may try to check the user's IP address against a database that correlates IP addresses and ZIP codes. This approach strikes a balance between accuracy and abundance. The website might not have registration profiles from everyone who visits their site, so for everyone else, they use a large IP database. Registration data is, by its nature, more accurate than IP targeting (I'll explain why later), but IP data is more easily obtained.

To gauge the accuracy of your geotargeted campaigns, you first must get a handle on which method(s) you hope to use. The methods fall into two buckets, declared data and observed data.

Declared data refers to any data a user volunteers about himself. A registration profile where the user fills out his home address is a good example. Declared data is highly accurate in most cases, particularly when the user has a compelling reason to avoid fabricating data during the registration process. If a user is providing a home ZIP code to customize his news or get local information, you can be pretty sure that the data is accurate.

If you're a buyer, always ask how publishers get their declared data and how often it is updated. If you're a publisher, remind your users to update their profiles from time to time, so that your data doesn't grow stale.

Observed data is just that, data gathered from observing a user or their surfing behavior. An example would be the IP address a user's browser gives to your web server as part of an HTTP request, or a user propensity for viewing the weather forecast in Austin, Texas. Since it's difficult to ascertain a geographic location from observed behavior, many companies that geotarget advertising simply use a geographic database keyed to IP address. But know that there are other methods-- some ISPs can target geographically based on the physical location of the Point of Presence (POP) that its users dial into.

One of the least accurate ways to geotarget is by content. You might make the assumption that many of the people hanging out at the Miami Herald's website are from the South Florida area. In actuality, the paper gets plenty of traffic from out of market.

So if accuracy is of critical importance for a campaign you're planning, you're going to want some quality declared data-- registration profiles from sites that remind users to update their profiles frequently. If you can tolerate some spill, targeting on observed data may be an option for you, particularly with the help of an IP database.

A word of warning on IP databases-- while some of the vendors offering these database services claim up to 95 percent accuracy, that doesn't mean you'll be able to target every ad with a confidence level of 95 percent. Not every IP address is able to be correlated to a location. After all, there are ISPs that mask that data through proxy servers. There are also IP blocks assigned to corporations and other entities that may straddle multiple geographies. So you will have plenty of users that can't be pinned down to any specific geography.

To see this in practice, try visiting DNSStuff.com and entering your IP address into the box marked "City from IP." (Don't know your IP? Visit WhatIsMyIP.com for a hint.) This geolocation database gets my city right, but only because the ISP we use for our connection at Underscore Central Command is based in the same city my office is in (New York). But if I were to dial in to the office while at an iMedia Summit and perform this exercise again, the server would still think I was in New York. In actuality, I'd be in one of the exotic locations Rick Parkhill picks for the events.

Additionally, an IP address can correlate to more than one location. When that happens, the good ad servers will go back to the ad queue for a non-geotargeted ad. The bad ad servers will consider that impression eligible for ads targeted to either geography. Obviously, this can significantly decrease accuracy. Keep all of this in mind when someone tells you that IP geotargeting is 95 percent accurate.

So there you have it. Geotargeting is only as accurate as the data and its implementation. If you're a buyer, be sure you know how all the sites on your plan are geotargeting. If you're a seller, remember that the best way to get accurate targeting information is to ask your users (and extend some value to them for being truthful with their answers).