ELI PORTNOY: It depends. On one hand, there are campaigns where good location requires 100-meter granularity precision. You need to know that this impression is within 100 meters of where you think it is. That type of good location data is basically anything that helps you understand more about the person. There are other campaigns where if you’re trying to reach someone within 3 miles of a location, the granularity doesn’t have to be that specific. So for those campaigns, good location data is 500 meters or less. Good location data means it has the right accuracy to do what you’re trying to accomplish.
How do you locate people? Are you relying on ad requests and combining that information with other data points?
One of the things we noticed in our experiments is the location data in an ad request wasn’t always accurate or true. The reason for that is publishers have many different ways of pulling location data. They can pull it straight off of the device’s GPS, they can triangulate it from Wi-Fi signals, they can get it through IP triangulations — and each one has a different accuracy level.
The publisher would get location data however they were getting it and translate it into a lat/long and then send it as part of an ad request. From the ad network, ad exchange or DSP [demand-side platform] side, [the ad requests] all looked the same and there was no way to determine which was “good” or “bad” location data. So we came up with a bunch of cool algorithms to distinguish between good location and bad location data. While we were building out our capabilities to do that, we also started discussing the acquisition with Telenav.
What attracted you to Telenav?
What was super exciting about Telenav is that it has an incredible first-party set of location data because they power the GPS solutions for some of the biggest mobile carriers and automakers of the world. Getting access to all that data and being able to find correlations and interesting insights to make ad requests even better was very exciting to us.
How specifically can you improve your ads?
In terms of attribution, there are a couple pieces that are interesting in the combination of the first-party data and our expertise. Telenav has developed a Web-based HTML5 turn-by-turn navigation system that we can embed in our ads and so when someone clicks on one of our ads, we can show that person how to get to that store or address. That’s a great proxy for attribution. We can tell people that we can literally help customers get to their store.
Where do the ads show up on Telenav’s GPS systems?
We run ads across thousands of premium mobile apps and mobile websites. In addition, we run ads on the Telenav GPS apps (AT&T Navigator, Scout, Telenav GPS) and these primarily show up on the search results. So, if you go to AT&T Navigator for instance, and do a search for food, we might show you an ad for Dunkin’ Donuts or if you’re looking for the nearest airport, we’ll show you hotels. This is a powerful mobile ad because it’s got two things. The first is it’s intent-driven. You’re actually searching for a certain type of business. The second thing is that you’re searching on a GPS on a mobile device, so you can quickly go to your destination. The search results are sponsored, very much like Google’s search results.
Do the ads become targeted over time?
They’re targeted over a number of dimensions. The first is distance. So, we’ll only show you stuff that’s close to you. We’ll also target based on keyword and category. And then there’s time of day and a whole bunch of other factors. We recently did this study where we looked at driving patterns across the GPS solutions and found that people tend to drive different distances based on what region they’re in, where they’re going and the time of day. For instance, people in Houston tend to drive 4.5 miles, whereas people in New York drive under 2 miles, which is interesting.
When people think about location-based advertising, many people are just creating broad geofences, but we’re saying you have to be more nuanced about it.
How do you see location targeting developing further?
I think there have been two phases so far and we’re heading into the third phase. The first one was all about proximity. It was about show me an ad because I’m close to something and tell me to go there. Location 2.0 is all about audiences — understanding who you are based on where you are.
If you’re at a car dealership, for example, you’re very likely an auto intender, and if I’m in a sports stadium, I’m probably a sports fan. The next phase is going to take this idea of a location profile to another level. It’s going to be a bunch of interesting audience profiling based on census data and other third-party data sets that will help advertisers get a more holistic view of the person they’re reaching.
Thinknear was originally a yield-management tool. What led you to pivot?
The pivot happened because we were trying to build this yield-management solution for SMBs that was a very complicated product. At the heart of the product were algorithms that would predict how busy or slow a business was and we would provide mobile ads during those slow periods. We started doing deals with pretty much all the mobile ad networks to buy mobile ads.
We encountered two big problems when we tried to do that. One was that we couldn’t buy enough inventory. Every time we’d try to do a deal, we’d want to do X number of impressions and they’d be able to deliver only a tenth of that, because everything we wanted to do was hyper local. The second issue was not being able to verify where the ads ran.
We couldn’t build a business on top of [the mobile ad networks’] solutions because they didn’t have enough scale and they didn’t have the targeting we needed to do a good job. So instead of waiting for someone to figure it out, we said, "Let’s ditch our business model and focus on solving hyperlocal [advertising]."
How do you “solve” hyperlocal advertising?
Solving hyperlocal for us meant making sure we have enough scale to run big campaigns and good location targeting. The first thing we realized is to get scale we couldn’t be an ad network and we couldn’t go directly to publishers exclusively, because when it comes to location, it’s important that you only buy the impressions that make sense.
Why does this preclude the publishers?
Most publishers think of monetizing their app on an eCPM basis. The obvious calculation is how much of the inventory gets sold times what price it gets sold at. And if you’re only buying 1% of traffic because you only care about impressions that are in a car dealership within a certain distance, you’re basically not buying the other 99%. But to do location really well, you need to be able to buy a small fraction [of inventory] and still have massive scale. That doesn’t really happen in an ad network model.
So this was late 2011 and at that point there wasn’t much happening on the mobile RTB front. Nexage was just starting their RTB efforts and there were no other exchanges out there. We had faith that RTB would become important though and so we started working wth Nexage, then MoPub came along, then Smaato and we started to see scale.