Cohort Analysis with SeekTable

Cohort analysis is used for getting insights about user behaviour when they are grouped by some criteria. This kind of analytics is performed with dataset that represents a history of some events; this might be page views statistics or log of user actions (say, purchases).

On input you need to have tabular data with the following columns:

• unique user identifier, like user_id
• properties needed to determine user cohort, like create_date or registration_date
• date of the event performed by the user, like event_date

This article describes how to use SeekTable for cohort analysis and create user retention report (like in Google Analytics).
Online demo for this kind of report: user retention report sample.

CSV data source

Lets assume that our history data is represented by user_events.csv file. After file upload SeekTable automatically suggests "year" and "month" dimensions for date columns:

• "Reg Date (Year)" and "Reg Date (Month)" for the user's registration date
• "Event Date (Year)" and "Event Date (Month)" for the event date

To perform cohort analysis we need to add one custom dimension to calculate number of months (this may be days or weeks) between event data and registration date (Cube → Edit Configuration). For this purpose "Expression" type of dimension can be used:

In dimension with type="Expression" 1-st parameter is a formula expression, and next N parameters are names of dimensions that are used as arguments of the formula. See also reference on calculated fields for more details on this topic.

Then let's add a measure that counts number of unique users inside groups:

Now we can configure user retention by month report:

1. select "Reg Date (Year)" and "Reg Date (Month)" for rows
2. select "Months from Reg Date" for columns
3. select "Count Unique Users" for values
4. click on "Apply", you should see a report like this.

You can use options from "Format" tab to make your user retention report like in Google Analytics:

• Value calculations: select % of row
• Heatmap: select By Row
• Totals: uncheck Grand Total and Column Totals

Now you have basic cohort report; you can use another dimensions on columns to change cohort criteria; also you can change the formula and user another period instead of month - say, a day or quarter.

SQL data source

In case of SQL database all steps are the same as for CSV, only difference is that instead of CountUnique measure type (which is available only for CSV cubes) you need to configure custom measure based on SQL COUNT(DISTINCT):

Also you can calculate number of months (weeks, days) between event and registration dates on SQL level, without usage of "Expression"-type dimension for month_from_reg. For example if you use MySql you can define it in the following way: