STEVE GLANZ: We gather massive amounts of data, all not personally identifiable, related to things like location, IP address, interests, the apps people use, etc. About 30 data points in all. Then we look for patterns in that data to indicate that this PC and this phone must be the same person based on looking at the behaviors on those devices over time.
How do you do that without PII?
We look at the data over enough time to come to a highly probabilistic conclusion that those devices must be the same person. The test is taking deterministic data, where we do know for sure, and using it to test our statistical accuracy. It’s based on machine learning and statistics, so we don’t know for sure. But if we see 10 devices coming out of the same IP address every night in Long Island, it’s pretty easy to figure out that’s probably a home. And then if we see two of those 10 devices traveling together every day into Manhattan, we know they probably belong to the same person. And then we look at the types of apps that person has and the domains that user visits. Of course, that’s a somewhat simple example. Something like the dorms at NYU would be much more complicated. But that’s why we have data scientists.
Are you SaaS?
We license our technology to other ad tech companies like ad servers, DSPs, exchanges, etc. We provide those companies with an identity solution and they can then go out and use that data as they see fit in their business. It’s completely self-service. The product is just data. Our customers call our API with a user and we get back to them with that user’s other devices. So, I guess you could say that we’re more platform-as-a-service.
Do you have access to any first-party data?
None of it really gets to the level of first-party. We’re making inferences on different data points. But sometimes something totally unexpected ends up being a key to identity. Maybe, I don’t know, people who visit cnn.com on the web also like playing Angry Birds. That’s something you’d never assume to be true, but the data tells us everything. But you need a large set of validation data and learnings to make these kinds of connections. It’s all about the data.