[druid-user] Funnel analysis where order of events is preserved

one approach would be to use a virtual column with users and time part (hour of day, day week…). Compute a theta sketch on the virtual column with event filter of buy say. Then get another virtual column with users and timepart of time shifted by -1 (hour, day…) and compute a theta sketch on this virtual column with event of the earlier event in the sequence. Now the intersect should give you what you want. The issue with this approach would be that a user may not follow through with the entire sequence in a specific interval