The core idea
Segmentation groups observations to answer a question: for example by role, location, contract arrangement, tenure or relevant experience. It can reveal a pattern hidden by a company-wide average. The grouping is an analytical choice, not a statement that all people in a group are alike. Define segments before looking for a desired result where possible, and distinguish legitimate analysis of unequal outcomes from using sensitive characteristics to determine individual opportunities.
Source and attribution [1]Using it in practice
Start with the decision and select a small number of plausible groupings. Show group sizes, numerator and denominator alongside rates. Check whether differences reflect job mix, time at risk or data quality. Avoid publishing small or easily identifiable groups; a single fixed minimum group size is not a complete disclosure-risk assessment. Review the purpose, access and lawful handling of personal data with appropriate specialists.
An example, not a reported case
Worked example · illustrative
In an invented organisation, Site A has eight voluntary leavers from an average workforce of 80 and Site B has six from 30: 10% and 20% respectively for the same period and definition. Raw counts alone would point to Site A. HR then checks whether contract mix and tenure differ before blaming Site B’s manager. Interviews and scheduling evidence help identify which explanation deserves testing.
What to watch for
Many comparisons increase the chance of finding an apparently unusual group by accident. Segments can act as proxies for protected characteristics. Performance ratings may themselves contain bias. Do not label a small team a hotspot on the basis of one unstable percentage or publish combinations of filters that reveal individual records.
Choose groups for a reason
Write a sentence explaining why each grouping could help answer the question. Role and tenure may be relevant to onboarding; they may be irrelevant to another problem. Resist adding every available field. A narrower analysis is easier to interpret and easier to govern.
Check the denominator
A group with more people usually has more events. Compare rates over equivalent periods and report counts. If exposure differs substantially, consider an appropriate person-time measure or a time-to-event approach. Explain the choice in language the decision-maker can understand.
Investigate composition
A site’s overall rate may change because its mix of roles or tenure changed. Compare like groups and inspect whether a company-wide pattern reverses within groups. Do not assume that adjusting for a few recorded variables removes every alternative explanation.
Turn the finding into a fair action
Use a pattern to improve a process or investigate a hypothesis, not to attach a negative label to everyone in a segment. Agree how the intervention will be evaluated and check whether it creates unequal access or unintended burdens elsewhere.
Choose groups that answer the question
If investigating early turnover, tenure, joining cohort, role and site may be useful. If examining development access, working pattern and eligibility may matter more. Define groups before searching for an interesting difference where possible. Avoid slicing data into so many combinations that every apparent pattern rests on a handful of people. The segmentation should make an operational question clearer, not create labels that follow employees indefinitely.
Show the denominator and uncertainty
In an invented example, two departures among ten people is 20%, while ten among one hundred is 10%. The smaller group's rate is higher, but its evidence is less stable and the larger group accounts for more exits. Show counts and rates together. Consider whether the periods and opportunities to leave are comparable. A useful discussion asks both where risk appears elevated and where an intervention could make a meaningful difference.
Protect people in small groups
Decide what can safely be reported, to whom and at what level of detail. Removing names may not prevent identification when a combination of role, site and tenure points to one person. Consider aggregation, suppression and complementary disclosures across several reports. A minimum group size is only one control and should not be treated as a universal guarantee of anonymity. Limit access to the underlying records and explain the purpose of the analysis.
Test whether the difference survives context
A site-level gap may reflect role mix, seasonal hiring or different follow-up rather than a manager effect. Compare relevant like-for-like groups and seek operational explanations. Avoid using a segment as a proxy for a person's likely behaviour. The aim is to identify work conditions or processes worth improving. If the analysis cannot support a fair action, reconsider whether the segmentation is necessary or whether a broader service improvement would be more appropriate.
Choose a grouping that answers the question
For an absence question, role, shift or working environment may be more informative than an organisation-wide average. Define the groups before inspecting every possible combination and explain why the distinction matters. Small cells can be unstable and identifying, especially when combined with location or sensitive information. Aggregate or suppress appropriately and restrict access. A segment is an analytical view, not a permanent label attached to a person.
Check whether composition explains the difference
Suppose one site has more early exits. It may also have recruited a much larger proportion of new employees into difficult shifts. Compare like periods and relevant conditions before attributing the difference to the manager. Explore the result with local knowledge and employee evidence. Segmentation reveals where to ask a better question; it does not establish why a pattern exists or justify treating everyone in a group as if they had the same needs, risk or performance.
Take it into your next conversation
Three useful questions.
- Which decision will this analysis change?
- Are the population, denominator, time period and assumptions explicit?
- What alternative explanation or unintended effect must we check?
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.
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