Interview: IPinfo’s Ben Dowling on the Geomob podcast (part 2)

IPinfo founder and CEO Ben Dowling sat down recently with the Geomob podcast, where he and host Ed Freyfogle discussed the many uses of IP geolocation and other IP address data, and how Ben was able to grow what started as a weekend project into a leading global provider of IP address datasets. The transcript is below, edited for readability. Listen to the full episode here and read the first part here.

New Frontiers in IP Geolocation

Ed Freyfogle: So, one of the big joys of geo is the crazy edge cases. When I told someone that I was going to be talking with you, he had an immediate question. What about the computers on the space station? Presumably they have IP addresses. How do you tag those?

Ben Dowling: We’ve had this question as well about Starlink HQ. The ground space stations are an obvious example. They’re not on earth. What do you say for that? There are other examples that are interesting. Traditionally, people think of an IP address as having one location or mapping to one device. But with anycast, one IP address can map to multiple devices. Google's 8.8.8.8 public DNS. That isn't one server, it’s 20 load balancers, each of which goes to hundreds of devices probably. There are more and more of these, these types of things. We have anycast detection now. In our API, for popular ones like 8.8.8.8 or1.1.1.1,  but there are also a bunch of not well-known anycast IP addresses. In those cases, we return the API, and say, we've detected this as anycast. We detect that through our probe network. If we're seeing ping results from 10 places around the planet that are faster than the speed of light, we know that it can't be in one location. It has to be an anycast result. We flag that, but then we still give you a city. It simplifies the process for users who may want to do some website customization, but may not want to get into the granular details. They just want to show something so we return the address of the owner of that IP space.

Ed Freyfogle: I use your site quite a lot whenever I need to find out my IP address, because it detects my own location. The last time I went, a little pop-up came up and said, is the state correct? And I could confirm it. So are you also crowdsourcing from the users and the customers?

Ben Dowling: We don't use the results of that directly in our geolocation data set. We started asking for that to come up with some internal accuracy metrics and track how we’re improving over time. If you go to one of our IP pages through Google, we’ll ask if you are where we say you are. If not, then we'd try to get a location fix and ask if you want to share that. That goes into a data accuracy pipeline for us, where we're measuring how many people are saying yes, how many people say no, and for those that said no, where we have the location, how far off are we, and track that over time. This is really interesting, because we're getting to the point where the issues that we're fixing tend to be more geo issues than IP geolocation issues. It may look like someone’s in Philadelphia on Google maps, but they’ll say, no, I'm in this neighborhood. That’s definitely not Philadelphia. It gets into the whole geo-hierarchy thing. Is this a suburb of Philly? Is this really a separate town?

Ed Freyfogle: That's a level of pain I know. I saw you did a 2020 summary Tweet that you had half a trillion API requests in 2020, which was phenomenal. Who is doing so many API requests, and what's the business model?

From the Basement to Billions of API Requests

Ben Dowling: That's a pretty mind-blowing stat. We track our volumes daily and we have internal reporting, we only realized at the end of the year that we almost broke a record for capturing an API request, which is crazy. We are consistently doing over a billion API requests a day now, which is pretty incredible. The caveat on that is that we offer a freemium API, so you can call the API without being a paying customer. So we’re not getting paid for all of those half a trillion requests.

We have a couple of offerings. We have an unauthenticated API, which is what our API was initially. It was only a free plan. You could pinpoint and it would return results, no API tokens. We still offer this and it’s feature-frozen. As we add new features and new data fields, they go into our authenticated API. The unauthenticated API has a limit of a thousand free requests a day. If you're on a low volume website, you can do that. Then we have an authentication API where you sign up and get an API token. The current limit on that is 50,000 requests a month. The experience is better because we’ll email you when you approach your monthly limit and offer an upgraded subscription for additional requests.

The bulk of our requests are from free users. We have a really long tail of websites and mobile apps. But we’re approaching 1,500 paying customers. Our two biggest API consumers may consume as much as the next 10 or 20 paying customers combined. We have some high-volume ISPs and some mobile carriers. We have ad networks, cybersecurity companies. There are various use cases. We've talked a lot about geolocation, and that’s the most common. Someone hits your website, you call the API and then do something about that result. It may be a trivial personalization that says, “good evening, how's the weather in Barcelona?”.  Or it could be a change of the content that’s not user visible, like taking the user to Amazon.es instead of Amazon.com. We also have carrier detection. We’ll look at an IP address and whether it comes from a mobile carrier, and which one? Or on a VPN or proxy? Some of these other datasets have more specific use cases around cybersecurity, around payments, fraud, and spam.

Ed Freyfogle: I remember when IPinfo first started, you got a lot of press because you said you whipped it up in a weekend. How big is your team now?

Bed Dowling: We've come a long way. I literally whipped it up over the weekend 7 or 8 years ago. I just downloaded some publicly available datasets, wrapped in a small API, used it myself, and posted to Stack Overflow in the hub. Very quickly people started using it a lot. That's when I added paid plans and it started becoming a business. And then we started investing really heavily on building our own datasets and improving the data that we have.

Initially we’d get feedback saying, your API is great, it's really responsive, but you’ve got my location wrong. We’d say, it's not our data, go complain to the data provider. Then we realized there was an opportunity there, so we learned to build our own high-quality datasets from scratch, and we work to improve them all the time. That's been a multi-year journey and we've learned a lot. Now we're a team of 15 people, and a lot of that is data engineering work on the geolocation datasets and our data quality. There's a web team that keeps the API up and builds out all the web tools. In the very early days, if you needed to make a change, you'd have to email me. Now there's a whole web portal and a support team that responds quickly to people interested in buying our data, or need service. It's come a long way from me sitting in the basement.

Find a Need and Build a Solution

Ed Freyfogle: It's a fantastic success story. Any advice for any members of the Geomob community thinking of starting a business and the learnings from the journey?

Ben Dowling: In some ways, no, right. I mean, I stumbled on to this success. It was a need that had helped me and helped a bunch of other people and then very organically grew into business. What I will say is I think that the great thing about geo is that it's a great way to build a very specific product that is useful for other people. The deeper you get into it, the more there is. The more edge cases, the more domain specific stuff. For most products, geo is not something that people would get deep into. They just want a map or an address or another location. Geo is a great area to build a business around specific pieces of functionality. We've seen that with lots of our customers who’ve said, we were doing this stuff in-house before, but it sucks and it’s difficult. We just want to pay someone to do this for us. My team loves working on this stuff. We love to iron out all these edge cases and get deep on it. Most companies just want to solve their problem and outsource layers of the geo stack. I think there are lots of businesses that could be successful in this space.

Ed Freyfogle: Very true. Our best customers are the ones who first tried to do it themselves and then realized how hard it is. Geo is a huge ongoing operational challenge. Congrats to you and your team that you're able to stay on top of that, because it’s very dynamic. So what does 2021 hold for IPinfo?

Ben Dowling: We’re always working on improving the accuracy of our datasets. We're pretty happy about the accuracy of our data is an ongoing issue. We're constantly striving to improve and really be world-class at geolocation. Privacy protection API is a product we launched last year. We've had a bunch of success with that, and we're investing further in improving our VPN detection. And a new product we launched not too long ago was Host.io, which is IP info for domain data. We'd already been collecting domain data and processing it internally for some of our IP datasets. We had a bunch of customers asking for it, so we built out a proof-of-concept API and launched it to a bunch of happy customers. So we’ll invest in further there as well.

Enjoyed the talk? Follow Ben on Twitter or get in touch with him and the team here.