How to build conversion tables from event logs
Understanding time-to-conversion is important for many applications. In medicine, researchers ask which treatment most increases patients' survival time. For predictive maintenance, we want to know which pieces of hardware have the shortest time until failure. For customer care, how long until support tickets are resolved?
Modeling time-to-conversion, however, remains rare in industry data science. One possible reason is that setting up the data can be tricky. In this article, I'll show how to do it starting from an event log, using the example of web browser events.
Event logs are a type of transactional fact table. They have a long-form schema where each entry represents an event, defined by the involved unit (e.g. user), timestamp, and type of event. Typically, each entry also contains fields for application-specific details.
Mixpanel's documentation shows an example log of user actions in a web browser. The key columns are USERID (the unit of interest), TIMESTAMP, and EVENT.
