Here at Tapjoy, we use Artificial Intelligence and Machine Learning to help make mobile advertising more rewarding and impactful for all parties involved. We recently sat down with Changsu Lee, our SVP of Platforms, to talk about how AI affects app monetization. You can read the results of that interview here.
For part 2 of this series, we spoke with Yohan Chin, our Vice President of Data Science, to learn how AI is being used to benefit mobile advertisers. Sporting a tee-shirt proudly proclaiming that “Data is the new bacon,” Yohan helped shed light on AI’s role in the mobile advertising ecosystem. Read on for his thoughts.
Can you explain how Tapjoy is using Artificial Intelligence and Machine Learning to help advertisers get better campaign results?
Absolutely. There are currently two main ways that we use AI. The first involves Audience Segmentation. We use AI to to build a profile on each device ID, based on which apps they use, when they use them, for how long, and so on, and from there we use graph belief propagation modeling to decide which category or genre a specific device ID belongs to as aggregation. For example, we can assign some devices as being into Fashion, or Sports, or Role Playing Games. We then layer on top of this other data from surveys that we run in order to get demographic details such as a user’s gender, age, education, etc. This gives us millions of datapoints about demographic data that we use to confirm our models, giving us extreme confidence in the accuracy of our conclusions. We can predict with a high degree of accuracy the demographics of a device’s user based on their app usage. Advertisers use this data to target the most effective audiences for their campaigns.
Another area where we are applying AI is in our Video-to-Install product. This is where AI goes beyond branding campaigns and begins to deliver true, measurable performance. The goal of any Video-to-Install campaign is to drive the highest quality or highest value users possible, so we don’t present the video ads to just anybody. Instead, we use AI to predict the probability that a user will install an app after viewing the trailer video for it. Then we can make a recommendation regarding how much it’s worth for the advertiser to bid on a specific audience. This lets them maximize their bid rates and hit their goals for app installs and user quality.
What’s the biggest challenge that ad-tech/mar-tech companies face regarding AI?
One of the biggest challenges when using AI is around scale and getting access to a sizable enough set of datapoints to provide meaningful analysis and make accurate predictions. Fortunately, that’s not a problem at Tapjoy, as we see more than 500 million monthly visitors and have access to more than 1.5 billion total devices.
Another big challenge involves privacy and ensuring that no personal data is collected while gathering all the datapoints that you need. Tapjoy has several measures in place to respect the privacy of our users and sensitivity of their data, but not all companies are so careful about it. So this is an industry-wide issue that must be addressed collectively.
Just having a massive amount of data is not solving any business problem, however. Companies need to figure out how we can get high quality of data and design AI system smartly to figure out the underlining meaning of data and take the right actions to drive successful outcomes.
What are you working on next as it relates to AI?
Tapjoy is using AI in our future Event Goal Based Bidding solution. This goes beyond just the install to make predictions about how likely a user is to complete a certain event within the app after they install it, such as watching the tutorial or completing level one. App marketers who want only the highest quality installs can use this feature to predict the likelihood of a user completing such an event, which helps them know how much it’s worth to bid on that user. It helps them make the economics of their campaign work at an effective rate. This product is still in beta but it is showing real promise!
How will advertisers use AI a year from now, or five years from now?
One area where AI stands to really make a difference is in regards to ad creative and not only optimizing its performance, but recommending how to actually design new creative that it knows will make an impact. You are starting to see some of this happening already, such as with IBM’s Watson and what they are doing. Already we are seeing how AI is able to build custom creative based on user preferences and behaviors, but that is just the tip of the iceberg. Pretty soon we’ll see entire ad campaigns using creative built by computers.
There’s also a lot of effort going into deep learning, which is a subset of machine learning. Basically, deep learning uses neural networks and intelligent algorithms to uncover insights that we humans would probably never discover on our own. A lot of advances are happening in the deep learning space right now and they will really push the boundaries of what is possible — in advertising and in many other fields. This deep learning is really useful to figure out hidden structural features from massive amount of data with greatly advanced computational power.