A major entertainment platform wanted detailed insight into their users behavior, but due to system limitations was unable to collect user behavior data at scale. Valkyrie developed models to classify and label user behavior data based on the limited data available, extracting value from a previously opaque data set. Our objective in deploying these classification models was to equip our client with the customer information necessary in order to reduce churn and increase new customer capture.

Science applied to music and entertainment

The Challenge

Our client needed a way to fill in major gaps in user behavior in order to measure quality and engagement with a new product offering. Collecting the user behavior data at scale was too expensive, but our team was able to leverage the data they did have. Valkyrie was tasked with using the available data to build models to retroactively construct user behavior.

The Solution

We developed a set of statistical models that use downstream side effects (changes in a user’s consumption, use of other features, etc.) of user behavior to fill in the missing data gaps. Our scientists were able to navigate a complex, messy, and incomplete data set to construct data models of user behavior based on the observed secondary effects of the unobserved user behavior. The predictions of our models provide a scalable way to label previously opaque user behaviors.

The Outcome

Our data labeling models allow the client to better understand how users are interacting with their products. This actionable insight has wide reaching effects on a range of business functions including product developers, marketing, content curators, and sales. The client is now able to identify and therefore mitigate everything from software bugs and implementations that cause costly data traffic to product quality gaps.

Events Classified

2.5 Million

Per Day


1.2 Million



515 Congress Ave. Austin, TX 78701 Suite #1425




(512) 947 – 6472

Stay in touch