This is a variation of a (common) reactivation-retention-churn analysis.
To put a customer into a particular cohort bin for a particular period,
one could lookup both previous and next period for a customer activity.
For example, if a customer placed an order in the current month
and also placed an order (or two :-) in the previous month
but no more orders in the next month, then the cohort is Retain,Churn.
The naming of cohort bins are somewhat arbitrary (hope you've got the idea).
Screenshot (of Churn & Reactivation Rates):
There are two main approaches to this kind of analysis in Tableau, mainly,
via Table Calculations or via LOD expressions (the latter requires v9+, of course).
Though nothing special comes with a Table Calc approach,
it is included in the attached workbook as reference
(on the worksheets Customers : TC and Cohorts : TC).
New to me is was :-) applying FIXED LOD calculations,
mainly exploring the idea of N distinct aggregation domains
shifted by a period to each other. In this particular case N=2.
So we have two domains (Period Frame 1 and Period Frame 2)
with the bin size of 2 periods each (Period is set by the Parameter).
For a particular Period (say month of January 2012)
the calculation looks up the aggregation in the respective bin,
such as Jan-Feb 2012 for the Churn calculation,
or Dec 2011-Jan 2012 for the Reactivation calculation.
Both results then combined to form a Cohort bin.
Since all Cohort calculations use fixed LOD as a base,
their result (Cohort bin) is a Dimension, so one could use it
at its full value to visualize any specific analysis scenario.
I've intentionally put the First and the Last Period into their own bins.
Since there are no Periods prior to the First (and next to the Last),
Reactivation couldn't be calculated in the First (and Churn in the Last).
All the necessary calculations are grouped in two Folders:
-- Cohort LOD (LOD Calculations hence found in Dimensions)
-- Cohort TC (Table Calculations hence in Measures)
Feel free to use / modify / enhance.
Hope it could help a bit.
PS I've previously explored the idea of using fixed LOD
to calculate moving weekly count distinct (as seen here):
Tableau Version: 10.0
Original Author: Yuri Fal