What is cohort analysis?

Grouping records by a shared start date and tracking each group over time, rather than aggregating everything in a reporting period. It separates changes in behaviour from changes in mix and volume.

Why use cohorts instead of period totals?

Because a period total mixes populations that started at different times under different conditions. Retention, conversion and cycle length all move on volume and timing alone when measured by period.

What should you cohort by?

The event that starts the clock for the question you are asking — signup date for retention, opportunity creation for win rate, first purchase for expansion. The wrong start date produces a confident, meaningless chart.

How long do you have to wait?

At least one full cycle of whatever you are measuring, and report the share of the cohort still unresolved. A cohort read too early looks better or worse than it is depending on which cases resolve first.

Cohort Analysis

Grouping customers by a shared starting characteristic — usually the period they were acquired — and following each group over time.

The problem it solves

Period metrics divide one population by another. Conversions this month came from entries made weeks or months ago, so the rate moves whenever volume or timing changes — with no change in how anything actually performs.

The fix, stated simply

Group by when things started, then follow each group forward. Slower to read, because a cohort must mature before its numbers are final, and the only version that describes a process rather than an accident of timing.

Where it changes the answer

Period versus cohort

Metric	Period version misleads because	Cohort version
Retention	Rapid growth adds young customers who cannot yet churn	Follow each signup month forward
Win rate	Deals closing now were created under old conditions	Group by opportunity creation date
Conversion	Numerator and denominator are different populations	Follow each entry month to outcome
Payback	Spend and customers are from different periods	Match spend to the cohort it produced

Source: Framing stated here

Running one

Choose the start event deliberately — it defines what the analysis can answer.
Fix the cohort at the start and never add to it. Adding later arrivals is what turns a cohort back into a period.
Report the share still unresolved alongside every figure, so an immature cohort is not read as final.
Compare cohorts against each other, not against a blended average. The point is the difference between them.

Where it gets hard

Small cohorts are noisy. A month with twelve customers moves several points on one departure.
Long cycles mean the most recent cohorts are always the least informative — and the most asked about.
A change in the start-event definition breaks comparability silently, so record any change to it.

RELATED TERMS

Logo Retention

The proportion of customers retained by count rather than by value.

Net Revenue Retention (NRR)

Revenue retained from an existing cohort over a period including expansion, expressed as a percentage of where that cohort started. The clearest single indicator of whether the base grows without new customers.

Customer Acquisition Cost (CAC)

The fully loaded cost of acquiring one new customer: sales and marketing spend for a period divided by new customers acquired from it.

Churn

Customers or revenue lost over a period. The leak determining whether acquisition compounds or merely replaces.

COMMON QUESTIONS

What is cohort analysis?: Grouping records by a shared start date and tracking each group over time, rather than aggregating everything in a reporting period. It separates changes in behaviour from changes in mix and volume.
Why use cohorts instead of period totals?: Because a period total mixes populations that started at different times under different conditions. Retention, conversion and cycle length all move on volume and timing alone when measured by period.
What should you cohort by?: The event that starts the clock for the question you are asking — signup date for retention, opportunity creation for win rate, first purchase for expansion. The wrong start date produces a confident, meaningless chart.
How long do you have to wait?: At least one full cycle of whatever you are measuring, and report the share of the cohort still unresolved. A cohort read too early looks better or worse than it is depending on which cases resolve first.