[kpsychwave]

GeoIP is pretty accurate at the state/country level for most users, but you will run into precision issues at the city level.

A bigger problem seems to be that many forget to continuously sync their IP DB with their provider. Your targeting is only as good as your IP -> Geo map.

My team built a tool for testing GeoIP implementations here: https://www.geoscreenshot.com to get around the issue of testing if it works.

[kjs3]

That depends on your use case. Huge numbers of people (e.g. people at work) use VPNs, and their 'geolocation' could be wildly different than their actual location. If you're an IBM employee (what...quarter of a million people) on the VPN, you look like you're in New York someplace. At my current employer (80k), most of us look like we're in Minneapolis, even though I'm half a continent away. If you're, say, targeting ads based on city/state level GeoIP, that's a lot of misdirected ads.

[kpsychwave]

VPN users are an exception. I would think it would be best to use a proximate node to reduce latency.

For corporation, there is another form of targeting (account based targeting) that relies on IP ranges. I believe DemandBase covers this specific use case.