Fake News: a Technical Review of Modeled IP Targeting

Imagine a mail vendor offering zip code targeting or a phone vendor promising area code-level precision. They’d be laughed out of the room twenty years ago. Yet today we see several IP targeting vendors and their resellers offering just that very capability. The best-informed players in our market know it’s BS. The others need a little help.

Let’s be more specific. Modeled IP targeting without data from Internet Service Providers lacks reliability, consistency and accuracy, which translates to unnecessary wasted impressions to the wrong geo or desired audience and a missed opportunity to reach and persuade your intended audience. Dozens of low-information consultants and candidates around the US are wasting campaign funds on digital targeting that is about as accurate as a water balloon dropped from thirty thousand feet.


The best source for data on IP privacy is Maxmind, a company dedicated to detecting online fraud and locating online visitors. Maxmind does good work. Their research says IP targeting is rarely accurate at the household level, and only 58% accurate at reaching the right city.

The Electronic Frontier Foundation, a nonprofit focusing on privacy in the digital world, investigated accuracy of modeled IP targeting and concluded it should never be used alone to determine location:

“There is no central map or phonebook that connects IP addresses to particular locations, particularly given that IP addresses are often reassigned to different Internet users over time. There’s also no uniform way to systematically map physical locations based on IP addresses; although some mapping techniques may be extremely accurate for some addresses, the resulting maps are not “official” and will not be completely comprehensive.”


Overblown claims are what you get when there is a mad rush to capture part of the $2B+ political digital market next year. Another vendor claims to target 200M households. The number of records is immaterial without understanding the data source and quality. Having millions of records does not mean correct linkages between IP and household. Another false claim is 85%+ match rates by loosening the criteria by calling it a match. There is no such thing. There were 244M people online in the US in 2017. These vendors are claiming they’ve matched 275M of the 244M people online in the US. And finally, there are claims on HH targeting except that you as the client have to provide an address list and are then subject to the varying degrees of quality of the data onboard partner they use to match those addresses and activate using a daisy chain off-line to online techniques. When a4 onboards a marketer’s list, we do it using our onboard tech and match that list in a deterministic way with our known verified, authenticated, address universe.


Here is what is happening: Our competitors serve an ad to a mobile device that reported its location at a longitude and latitude and then observe that the mobile device’s IP address is ‘xxx.xxx.xxx.xxx’ The challenge is that mobile apps often report imprecise or very inaccurate longitude and latitude positions or simply default the to the center point of the US; having a location on their ad call makes it more valuable so they send one; app publishers get paid for having the location, not for having it be accurate; they are incentivized to ‘estimate’. And we have not touched on the fact that geo collection typically requires the app to be in use meaning that geo location and targeting is not on default an always on state for marketers to reach that audience.

Based on a potentially accurate geo-coding database our competitors buy they believe that longitude and latitude to the closest match at “123 Walnut Place 19075”. From there, they associate IP address xxx.xxx.xxx.xxx with and save that correspondence for future campaign targeting. Even if the IP address was accurate (it’s not), home IP addresses are dynamic and change frequency. So, without a constant feed of fresh data, low precision goes to no precision. Check modeled IP precision here for a map of where modeled IP vendors think you are: http://checkip.org/


Let’s suspend disbelief for a moment. According to Maxmind, IP data is only 58% accurate to the city level but let’s be charitable. Even if the mobile device properly reports location (which it usually won’t if it is being sent with an ad), and even if your geo-coding database is perfect (which it won’t be) your circle of confidence will intersect about three houses, depending on housing density. So, in the absolute best-case scenario is that your postal to IP mapping will be accurate one third of the time. Meanwhile our authenticated household mapping is accurate basically 100% of the time because of the daily feed of fresh authenticated IP data coming from our partners, backed up with a cookie match for anyone not matched and verified.

Even if these vendors are doing what they’re claiming to do, they’re relying on 3rd party data from app makers layered on top of 3rd party, publicly available IP addresses, layered on top of latitude / longitude data from someone, all to get a “match” that best case scenario, this all fall apart when the ISP refreshes their IP address.

Without the data from the organization providing home Internet service, solutions that leverage traditional IP targeting are the key that only sometimes opens the lock. Those solutions are not real. Made up. IP Zones, which some vendors refer to, rely on predictive modeling. Choose probabilistic data from other modeled IP vendors at your own risk; it doesn’t work with tight targeting. Just go ahead and light your cash on fire- it’s just as effective.

Jordan Lieberman
General Manager, Politics and Public Affairs

Sahil Gambhir
Sr. Director, Product Management

Learn More