cohort

Introduction

Cohort analysis is a common method for tracking groups of users, customers, or entities over time to understand retention, behavior, or trends.

The cohort() function in fpaR provides a database-friendly implementation of cohort analysis that is compatible with duckdb and snowflake backends. It combines the functionality of multiple traditional cohort tables into a single streamlined workflow.

Function Purpose

cohort():

• Assigns each member in your dataset to a time-based cohort based on the first occurrence in the .date column.
• Aggregates distinct counts of cohort members over time.
• Supports multiple calendar types ("standard" or "554").
• Can summarize cohorts by various time_units (day, week, month, quarter, year).
• Returns a segment object, which can be processed using calculate() to generate the cohort table.

Arguments

Argument	Description
.data	A tibble or DBI object containing your dataset.
.date	The date column to base the cohort assignment on.
.value	The column representing members to track (e.g., customer ID).
calendar_type	Type of calendar to use for aggregation: "standard" or "554".
time_unit	Time unit to summarize the cohort table by (day, week, month, quarter, year). Default is "month".
period_label	Whether to use numeric period labels (TRUE) or actual dates (FALSE). Default is FALSE.

How It Works

1. Assign Cohorts
   Each member in .value is assigned to a cohort date, which is the first date they appear in .date.

2. Aggregate Over Time
   The function counts distinct members per cohort across time periods defined by time_unit.

3. Database Efficiency
   The function is optimized for DBI objects, generating SQL-compatible operations using dbplyr.

4. Internal Execution
   The internal function cohort_fn():

   • Floors dates to the chosen time_unit.
   • Groups members by cohort and date.
   • Computes distinct counts and period IDs.
   • Reshapes the table into a wide format for analysis.

Output

The function returns a segment object. Use calculate() to retrieve the cohort table:

#|eval: false
library(dplyr)

Attaching package: 'dplyr'

The following objects are masked from 'package:stats':

    filter, lag

The following objects are masked from 'package:base':

    intersect, setdiff, setequal, union

# Example cohort analysis
fpaR::cohort(.data = contoso::sales,
                  .date = order_date,
                  .value = customer_key,
                  calendar_type = "standard",
                  time_unit = "month",
                  period_label = FALSE
             )

── Time Based Cohort ───────────────────────────────────────────────────────────

Function: `cohort` was executed

── Description: ──

This segments groups based on a shared time related dimension so you can track
a cohort's trend over time

── Category Information ──

• The data set is grouped by the customer_key and segments each group member by their first order_date entry to define their cohort customer_key
• This creates cohort ID that each member is assigned to eg; January 2020, February 2020, etc
• The distinct count of each customer_key member in the cohort is then tracked over time

── Calendar: ──

• The calendar aggregated order_date to the month time unit
• A standard calendar is created with 0 groups
• Calendar ranges from 2021-05-18 to 2024-04-20
• 222 days were missing and replaced with 0
• New date column cohort_date was created from order_date

── Actions: ──

✖Aggregate

✖Shift

✖Compare

✖Proportion Of Total

✔Count Distinct

── Next Steps: ──

• Use `calculate()` to return the results

────────────────────────────────────────────────────────────────────────────────