← People analytics

People analytics · technique

Cohort analysis: compare like starting points

A cohort is a group defined by a shared starting event or period, such as joining in a particular quarter.

Statistical and evaluation practice; ADM HR worked application · 2026 teaching guide · Original editorial explanation
Cohort analysis: compare like starting points — simplified model sketchJoining cohort · Same follow-up · Comparable outcome. Editorial interpretation, after Statistical and evaluation practice; ADM HR worked application (2026 teaching guide).90-day retention · invented cohortsJanuary: 80 / 100April: 88 / 10080%88%ADM HR · ORIGINAL INTERPRETATIONStatistical and evaluation practice; ADM HR worked application · 2026 teaching guide
Original simplified sketch after Statistical and evaluation practice; ADM HR worked application (2026 teaching guide). Read the explanation for assumptions and limitations. Download SVG ↗

The core idea

A cohort is a group defined by a shared starting event or period, such as joining in a particular quarter. Cohort analysis follows comparable groups over equivalent periods. It can reveal early-tenure problems that a company-wide snapshot hides. A recent cohort is not automatically comparable with an older one because its members may not yet have reached the outcome window. The grouping and outcome definition should be chosen to answer a clear question.

Source and attribution [1]

Using it in practice

Define the cohort, starting date, outcome and observation window. Include only fully observed cases in a simple fixed-window comparison, or use an appropriate method for incomplete follow-up. Show exclusions and counts. Check whether the cohorts differ in role, site, season or recruitment route. A change across cohorts is a finding to investigate, not proof that the latest onboarding initiative caused it.

An example, not a reported case

Worked example · illustrative

An invented January cohort has 100 starters, with 80 still employed at day 90: 80% retention. An April cohort has 100 starters and 88 retained at day 90: 88%. The difference is eight percentage points, or a 10% relative increase from 80%. If a July cohort has only 30 days of follow-up, its current 95% retention cannot be compared as a 90-day result.

What to watch for

Define how rehires, transfers, contract endings and missing dates are handled. Excluding incomplete cases can itself select a different population. If follow-up varies substantially, consider survival analysis. Avoid revealing small cohorts or treating a cohort label as a judgement about its members.

Build a cohort table

Use rows for joining periods and columns for elapsed time, such as 30, 60 and 90 days. Leave unobserved cells blank and label them clearly. Do not fill them with zero or carry forward a short-term result as if it were complete.

Interpret the difference

Report the starting number, retained number, rate and time window. Investigate changes in work availability or role mix. If the decision matters, include uncertainty and a suitable comparison design rather than relying on two percentages alone.

Define entry and observation consistently

State what makes someone part of a cohort: joining month, completion of a programme or another event. Then specify the follow-up period and outcome. For ninety-day retention, only compare cohorts whose members have had the relevant opportunity for observation, or use an appropriate time-to-event method. Keep late data corrections visible. A recent cohort cannot be called successful simply because there has not yet been enough time for its members to leave.

Work through a like-for-like comparison

Imagine 50 January starters with 40 still employed at day ninety, and 60 April starters with 51 still employed at day ninety. The observed rates are 80% and 85%, a five-percentage-point difference. That is not automatically the effect of a new induction. Check role mix, recruitment source, site, season and any other change. The example demonstrates a comparable denominator and follow-up, while leaving causal interpretation open.

Examine the shape as well as the final rate

Look at when exits occur within the period. Two cohorts can share the same ninety-day retention but differ in whether people leave during the first week or later. That distinction may suggest different questions about job expectations, access readiness or supervision. Avoid overinterpreting very small weekly counts. Use the pattern to guide interviews or process checks, then test the explanation against additional evidence.

Keep the next action specific

If the first-week experience appears problematic, choose an intervention connected to that stage and define how it will be evaluated. Retain a comparable reference where feasible and record implementation differences. Do not simply rename the next intake a 'pilot cohort' and assume comparison establishes impact. Report uncertainty and any remaining differences between groups. Cohort analysis is a strong descriptive tool when the definitions are clear; its causal strength depends on the design.

Align the starting point and exposure time

Compare people who joined in different periods at the same tenure point. A cohort that joined last month cannot fairly be compared with one observed for a full year using total departures. Define how transfers, rehires and incomplete records are handled. For an onboarding question, a 30-, 60- or 90-day view may be useful, but retain the denominator and the number of people actually observed at each point.

Separate a cohort effect from a period effect

A cohort recruited during a busy season may face different managers, demand or shift availability. If retention improves, ask whether onboarding changed or whether the operating environment became easier. In an invented example, 8 of 40 starters leave within 90 days compared with 4 of 40 later starters; the difference is worth investigating but does not prove the new induction caused it. Compare relevant conditions and use employee evidence before recommending a wider rollout.

Take it into your next conversation

Three useful questions.

  1. What question can this method answer, and what can it not establish?
  2. Are the comparison, observation period and assumptions defensible?
  3. What decision follows, and how will its consequences be reviewed?

Related terms

Go to the evidence

Sources & attribution

Further reading: Nadeem Khan and Dave Millner (2023), Introduction to People Analytics: A practical guide to data-driven HR, 2nd edition, PDF pp. 227, 244 and 294–295. Edition & reading notes ↗

[1] CIPD: Employee turnover and retention ↗

The core idea is an original summary of the cited work. Application notes, examples and sketches are our interpretations, not quotations or reproductions of the authors’ figures. Publisher records may require access to read the full original work.

Explore People Analytics →

Application guidance expanded on 26 September 2026. These practical suggestions and invented examples are editorial explanations; further-reading sources are not the original authors of the historical models unless expressly identified. Original overview review dates are retained.

Published 2026-09-20 · Reviewed 2026-09-22. Editorial approach

Keep exploring

A little more context.

People analytics · technique02

Survival analysis for retention

Survival analysis studies time until a defined event.

Statistical and evaluation practice5 min
People analytics · technique03

A/B tests and pilot evaluation in HR

A pilot explores feasibility; an impact evaluation asks what changed because of an intervention.

Statistical and evaluation practice5 min