Web Analytics · Product & Customer

Cohort, Retention & LTV Analysis

A young cohort is measured against another cohort's first months, never against a finished one.

A recent cohort may look unprofitable simply because it has not reached its second renewal. We align cohorts by maturity rather than calendar date, creating a like-for-like comparison before you draw conclusions about customer quality.

Comparable retention and LTV curves with maturity and data limitations stated clearly.

Plan cohort analysisAll Product & Customer Analytics tasks
A Zeo researcher lining up cohort jars on shelves by month, with an LTV curve above

Some of the 500+ brands we've worked with

See all references
  • BMW
  • Gedik Yatırım
  • Hotiç
  • DYO
  • Elle
  • Tatilsepeti

How we run it

Project censored cohorts without presenting projections as facts.

Fair cohort comparisons account for the time each group has had to develop. Four steps produce a maturity-aligned comparison with its uncertainty visible.

How we hold ourselves to it

  • Apply one customer definition to every cohort — We fix what counts as a customer and when the cohort clock starts, then use that rule throughout the comparison.
  • Let maturity set the comparison window — We fix the comparison window to the youngest cohort's age, so a 6-month-old cohort only ever meets another cohort's first six months, whatever that cohort's total lifetime turns out to be.
  • Label unfinished cohorts clearly — When a cohort has not fully matured, we project cautiously and identify the result as a projection rather than an observed fact.
  • Add cost to turn retention into LTV — We connect retention curves with actual revenue and service cost, producing a usable LTV figure rather than a percentage alone.

  1. Define the cohort model

    We agree on cohort entry, customer grain, and the value and cost definitions that go into LTV.

    Cohort definition

    AI assist
    Drafts cohort-entry and grain definitions from your data.
    Human gate
    Your analytics owner confirms the customer definition.
    Owners
    Cohort Analyst, Data Owner
    Illustrated figure sketching plans at a drafting table

  2. Build the curves

    We calculate retention and value curves for each cohort, handling censored (not-yet-mature) cohorts explicitly.

    Cohort curves

    AI assist
    Calculates retention and value curves per cohort.
    Human gate
    Analyst flags any cohort with too little data.
    Owners
    Cohort Analyst
    Illustrated figure stacking patterned building blocks

  3. Align and compare

    We compare cohorts at matched maturity points and flag where a difference is real versus an artifact of timing.

    Maturity-aligned comparison

    AI assist
    Matches cohorts at the same maturity point.
    Human gate
    Analyst rules out timing as the explanation before calling a gap real.
    Owners
    Cohort Analyst
    Illustrated figure reading an oversized measurement dial

  4. Walk through the findings

    We present the findings with their uncertainty and discuss what decision they can support.

    Cohort readout

    AI assist
    Drafts the readout from the aligned comparison.
    Human gate
    You confirm the decision the findings support.
    Owners
    Finance Stakeholder, Product Owner
    Illustrated figure presenting a bar chart on an easel

Every cohort gets compared at the same age

Automation calculates the curves, matches cohorts at the same maturity point, and drafts the readout from that aligned comparison. A person owns the definitions and the verdicts: what counts as a customer, which cohort has too little data to speak, and whether a gap is real or just timing.

What you get

Cohort curves accompanied by a decision memo.

You receive comparable curves and their limitations, each cohort measured at the same maturity point rather than collapsed into one average that hides the differences.

  • Working document

    Cohort and LTV model

    The reproducible model defining cohort entry, retention, value, and cost calculations.

    Accepted when

    Cohort entry and customer grain are defined once and applied identically to every cohort.

    Cadence: Refreshed as cohorts mature

  • Comparison report

    Maturity-aligned comparison

    Cohorts compared fairly at matched points in their lifecycle, with projections clearly labeled.

    Accepted when

    No cohort is compared on a longer window than the youngest cohort in the chart.

    Cadence: Redrawn at each refresh

  • Decision memo

    Readout and limitations

    What the curves actually show, what remains uncertain for less mature cohorts, and what decision they support.

    Accepted when

    Each projected figure is marked as a projection and names the method behind it.

    Cadence: At the readout

We call it done when: every comparison in the readout runs on matched time windows and each projected point is labeled as a projection, not an observation.

Fit and readiness

A three-month cohort may simply need more time to mature.

These signs help show whether a cohort analysis fits the decision ahead.

A good fit when

  • You need to see whether customer quality is improving or declining from one acquisition cohort to the next.
  • Recent cohorts appear weak, but you do not know whether the difference is real or simply reflects their younger age.
  • You need a defensible payback period or LTV figure, and the readout must separate projected cohort values from observed ones.

Better handled as other work when

  • You need to identify which marketing channel drove an acquisition rather than how customers behave once they're in. That is Marketing Attribution Modeling.
  • You need a controlled test of a specific product change rather than an observational comparison across cohorts. That is Product Experimentation Measurement.

If one of these is closer to your situation, start here instead: All Product & Customer Analytics tasks

We call it done when: the customer definition, the cohort entry point, and the value grain are agreed with the data owner before a single curve is drawn.

People who build your measurement system

Zeo designs measurement systems that connect a business decision to governed collection and reporting you can check. The people shown here work on the part of that system this page covers.

Tools we use

Tools behind this work

  • Jupyter

    runs the LTV projection so a forecast for an active cohort is a shown calculation, not an opaque platform estimate

  • Amplitude

    builds the maturity-aligned curves so a young cohort isn't compared against an old one at the same calendar date

Next step

Compare cohorts fairly before making the decision

Bring us your customer and revenue data. We will align cohorts by maturity and explain what the differences can support.
Plan cohort analysis

Before we start

Questions teams ask before booking

It may not yet have reached the renewal or repeat-purchase points that create later value. Aligning cohorts by maturity prevents that timing difference from distorting the comparison.