Part 1 of 2: The Future of Artificial Intelligence in App Monetization — Q&A with Changsu Lee
Previously the co-founder and CEO of Korea-based 5Rocks, a mobile analytics and automation platform that Tapjoy acquired in August of 2014, Changsu Lee is now the SVP of Platforms here at Tapjoy and is responsible for keeping our platform technology on the cutting edge.
We sat down with Changsu to talk about Tapjoy’s use of Artificial Intelligence and how he sees AI affecting the app developer community over the next several years. Here’s what he had to say.
How is Tapjoy using Artificial Intelligence and Machine Learning to help app developers today?
There are two main ways we use Artificial Intelligence today. The first is by making predictions about whether a player is likely to to become a paying customer or remain a non-payer, and how much revenue a developer stands to earn from each. By analyzing user behavior and looking at certain variables, such as how often the user opens the game, what levels they completed, whether their session intervals are increasing or decreasing, and so on, we can use machine learning to train our entire network to make predictions. For instance, we can predict to a fairly high degree of certainty the amount that a particular user is likely to spend over the next 30 days, or how many ads they are likely to engage with. This is the basis of our Future Value Map™.
The second way we are using Artificial Intelligence is to predict churn. Using machine learning and regression analysis, and again by studying the behaviors of similar users and devices, we can make accurate predictions about who will stay with an app and who is likely to leave, as well as when they are likely to leave. That way we can make recommendations to our developer partners about what to do in order to retain the user.
What makes Tapjoy’s use of AI different from other companies in the space?
At Tapjoy, our use of AI is focused on turning intelligence into actionable insights. For instance, we are able to say with a high degree of confidence that a specific user is likely to make an in-app payment, or to engage with a rewarded advertisement, or to churn, and then we can make specific recommendations on how to convert those users. And then there’s the fact that we have so much data to look at and base our predictions on — we are in a unique position because our network reaches more than 500 million monthly active users and therefore have access to more data, which leads to better accuracy. We have been using AI since 2015, and at this point have reached about 90% accuracy with our projections.
What are the biggest challenges that companies face when implementing AI solutions?
The single biggest challenge that companies usually face is getting enough meaningful data to make accurate predictions. Fortunately, that hasn’t been a problem for us here at Tapjoy because of our vast network.
Another problem is regarding privacy. I can’t stress enough how important it is to ensure that everything is tracked anonymously so that individual users’ rights are respected. We take extra steps to ensure all data is anonymized, but not all companies do this. When AI violates privacy policies and makes predictions based on personal data, that’s when it becomes creepy.
What is Tapjoy working on next? How else will you apply AI to app monetization?
We are also using AI to understand something we call “IAP-to-Ad Path”. We want to understand the path that users take from when they first download the app to when they either make an in-app payment, engage with an ad, or both. There’s an almost infinite number of paths they can take in order to get there, and we are training our neural network to understand these patterns so that we can predict which users are likely to follow which path. This is very important for developers to understand in order to form a more cogent and targeted monetization strategy.
There’s still a lot of fields where we are just now beginning to apply AI. Take fraud prediction, for instance: soon our risk prevention team will be able to identify potentially fraudulent users even before they make a fraudulent transaction. Then there’s the field of mediation and using AI to increase revenue for our publishers in real-time. In the meantime, we are continuing to train our neural networks and increase the accuracy of our predictions by better understanding things like IAP-to-Ad Path and other key behaviors of mobile app users.
Thanks Changsu! Stay tuned next time for when we interview Tapjoy’s Yohan Chin to explore how Tapjoy uses AI to benefit mobile advertisers.