microsoft / Kusto-Query-Language

Kusto Query Language is a simple and productive language for querying Big Data.
Apache License 2.0
510 stars 97 forks source link

funnel_sequence_completion step is not sufficiently documented #52

Closed keenbrowne closed 1 year ago

keenbrowne commented 2 years ago

Not to be confused with "funnel step" or "sequence step", phrases used elsewhere in the same page of documentation, "Step" or "window" has a large impact on the results returned by the function. Under the hood it appears that the function may be binning the TimelineColumn but exactly how and why that bin impacts which records are counted is unclear. For example does this mean that records are only counted when they map to the same bin regardless of the array passed as MaxSequenceStepWindows? Should I just always select the largest bin to get one funnel per sequence step window? Why is the sum of funnel sequence step counts binned with a window of 1d so different than that of a single large window of say 365d?

royoMS commented 2 years ago

Thanks for the comment: the docs have been improved - please let us know if you have further feedback.

The measure for the funnel analysis is a distinct count (dcount). Since dcount is not Associative (dcount(day1) + dcount(day2) is greater than dcount(day1+day2) in most cases) you can get very different results when doing a funnel analysis on each day separately (large window) compared to doing the analysis on the entire period.