When “Unknown” Becomes Your Largest Source
Why unclassified installs pile up, and how to shrink them without inventing fake precision.
An “unknown” bucket is not a moral failure—it is a signal that source rules lag how people actually arrive. Organic search, shared links, and delayed deep links often land without the parameters your model expects.
Treat unknown as a work queue. Sample recent installs, map the real entry paths, then decide which deserve named sources and which should stay unclassified honestly.
Shrinking unknown by inventing default labels is worse than leaving it visible. Stakeholders trust numbers they can challenge.