COHORT ANALYSIS WORKSHEET — ORIGINAL SYNTHETIC EXERCISE
ShowMeYourSaaS — 2026-10-27
Guide: https://showmeyoursaas.com/articles/cohort-analysis/
Companion data: cohort-retention-workbook.json

MEASUREMENT CONTRACT
Unit: one account, deduplicated within each interval.
Start: signup. Every account in a listed cohort has the same synthetic signup
instant, 00:00 UTC on its cohort date; this simplifies the teaching dataset.
Return: at least one completed project brief in the specified interval.
Week 0: elapsed days [0,7), not displayed as a return column here.
Week 1: elapsed days [7,14).
Week 2: elapsed days [14,21).
Week 3: elapsed days [21,28).
Cutoff: 2026-10-27T00:00:00Z.
Return-in-window retention permits an account to miss one week and return
later. This is not permanent cancellation or recurring-revenue retention.

EXPECTED SYNTHETIC TABLE
Signup 2026-09-29: 8 accounts. Week 1: 5/8=62.5%; Week 2: 4/8=50%;
Week 3: 4/8=50%.
Signup 2026-10-06: 6 accounts. Week 1: 4/6=66.67%; Week 2: 3/6=50%;
Week 3: incomplete (null, not zero).
Signup 2026-10-13: 4 accounts. Week 1: 3/4=75%; Weeks 2 and 3 incomplete.
The JSON contains all 18 invented accounts and their return intervals.
No real customer records, company results, or benchmark claims are included.

AGGREGATION
Week 1: (5+4+3)/(8+6+4) = 12/18 = 66.67%, rounded.
The unweighted mean of row rates is about 68.06%, a different weighting.
Week 2: (4+3)/(8+6) = 7/14 = 50%.
Week 3: 4/8 = 50%, using only the complete cohort.
Using all 18 for Week 3 produces 4/18=22.22% by counting unfinished
observation opportunities as failures. Do not do this.
Keep original starting accounts in each complete-cell denominator.

YOUR ANALYSIS
Question and decision:
Account/person unit and deduplication rule:
Starting event, date range, and cohort membership rule:
Qualifying return event and natural repeat cadence:
Elapsed or calendar boundaries; timezone:
Cutoff and incomplete-interval rule:
Source data and missing-event checks:
Counts, eligible denominators, and interval-level rates:
Weighted aggregate and exact contributing cohorts:
Customer mix or product/instrumentation changes:
Observed facts:
Interpretations and alternative explanations:
One next investigation, owner, and review date:

INTERPRETATION LIMITS
A four-account cohort moves by 25 percentage points when one account changes.
A higher cell is not proof that a feature caused better retention. Check
identity, maturity, source changes, customer jobs, and instrumentation first.
Product return, paid-customer retention, and revenue retention are different
outcomes. Do not silently substitute one for another.

Sources checked 2026-10-27:
https://stripe.com/resources/more/saas-cohort-analysis
https://amplitude.com/docs/analytics/charts/retention-analysis/retention-analysis-calculation
https://amplitude.com/docs/apis/analytics/dashboard-rest
https://mixpanel.com/blog/what-is-retention-analysis/