Volume, uniqueness of users, journey stage, and recovery rate form a workable score. None alone is enough: a rare crash in payment can outrank a frequent crash on a rarely used settings screen.
Write the score so another engineer can rebuild it. Name the events, the lookback window, and the exclusions for staff and bot traffic.
Once the score is shared, pair each top cluster with an owner and a next checkpoint. Impact work dies when lists have no names attached.