How OneSignal will use data science to optimize per-user delivery

Following up on our guest post on when to send notifications for each application genre, in addition to researching the optimization of notification timing, Kevin Kahn also worked on modeling individual user targeting for OneSignal. Kevin writes:

I wanted to push my AUC metric higher and one topic I was particularly excited to tackle was User Personalization. One Signal has a lot of information of how each user interacts with their notifications, so why not leverage this? To accomplish this, I use Bayesian Learning. When a user downloads a new social app, I have no information about their activity. I start with the prior assuming they act like the average social app user does.

Here you can see that most social users have the highest conversion chances in the morning and evening, which makes sense for a typical working adult. However, as a specific user receives more notifications, I slowly shift away from this prior towards a more personalized model. Take this one social user:

You can see that unlike the average user, this specific user is most likely to interact with their social app around 3pm. This is most likely since they are not a working adult, but instead a student getting out from school. With this user personalization, model performance (AUC) is pushed from 0.76 to 0.86.

We are currently working on including this model as a notification scheduling option on our platform. This scheduling option will use historical user behavior to send notifications to each user at their own best possible time.

This is only one of the many ways in which OneSignal is working to improve the effectiveness of push notifications for our clients.

~The OneSignal Team