Turnover Analytics: Metrics, Cohorts and a Review Template

Short answer

Turnover analytics examines who leaves a defined workforce, when departures occur and which working conditions deserve investigation. A useful analysis combines consistent headcount and departure records with employee context, compares equivalent groups and tests specific actions. It distinguishes a measured pattern from a possible explanation, then records what evidence would change the decision.

Start with a decision, then build the dashboard

A dashboard becomes useful when someone can act on its findings. Define a question before adding measures: should a regional team change new-starter support, investigate a shift pattern or allocate recruitment capacity differently? Name the person who can make that decision and the evidence they need.

The CIPD people analytics factsheet frames people analytics around solving business problems using people data. Applied to turnover, that means keeping descriptive findings, explanations and decisions visibly separate.

For example, an operations director might ask why early departures increased among warehouse starters. The first deliverable is a checked cohort comparison and a shortlist of explanations to investigate. An individual departure score would not answer which part of onboarding to change.

Use this guide to organise the analysis. For the arithmetic behind the headline measure, see the employee turnover rate guide.

Define the population and counting rules

Write a measurement note that another analyst could reproduce. Include:

• The reporting period and extraction date.
• The locations and worker types included in both departures and headcount.
• The departure event date used, such as the last employment date.
• The treatment of resignations, dismissals, retirement and planned contract completions.
• The handling of internal moves, rehires and duplicate records.
• The headcount observation frequency and treatment of missing observations.

A transfer between two sites changes local staffing without necessarily being a company departure. Keep that distinction in the data. The US Bureau of Labor Statistics JOLTS definitions count separations at establishment level and distinguish quits, layoffs and discharges, and other separations. An internal company measure can use a different scope, but the difference must be documented before comparison.

Reconcile movement records against the population totals. Opening headcount plus hires, minus departures, adjusted for in-scope transfers and other population changes, should explain closing headcount. A mismatch is a data issue to resolve before interpreting a trend.

Keep counts beside rates

A percentage can change because departures changed, headcount changed, or both. Show the numerator and denominator together.

Turnover rate = departures during the period / average headcount during the same period x 100.

The following is an original fictional example, not a benchmark or customer result. It uses resignations only and the same quarterly headcount convention throughout.

East and West finish with the same rate but warrant different questions. In East, investigate the workforce contraction and whether it altered workload. In West, inspect the additional departures and their timing. Neither result identifies a cause by itself.

For an aggregate rate, sum compatible numerators and denominators. An unweighted average of site percentages can mislead when sites have different headcounts. Keep geography, employment type and departure classification consistent across the combined view.

Use cohorts for early departures

A tenure band in a monthly dashboard and a starter cohort answer different questions. A tenure band describes the people or departures present at a point in the reporting process. A cohort follows a defined set of starters for an equal length of time.

Suppose a fictional January cohort has 40 starters and 8 leave within 90 days. Once every starter has had 90 days of observation, the early departure proportion is 8 / 40 = 20%. A February cohort has 50 starters and 5 departures within the same completed window, giving 10%.

This is a cohort departure proportion, not the period turnover formula based on average headcount. Say whether it counts all departures or only resignations. For the example, all departures count, and internal transfers remain employed.

A newer cohort still partway through its window is incomplete. Do not report everyone currently present as retained for the full 90 days. Show the number whose follow-up is complete, wait for cohort maturity, or use an appropriate time-to-event method with an analyst who can check its assumptions. Do not simply remove early leavers while retaining only successful completers.

Even with complete follow-up, a difference between January and February can reflect role mix, hiring sources, shift allocation or a changing labour market. The arithmetic establishes the difference; it does not establish what caused it.

Add employee context with its coverage visible

Use operational records and voluntary employee accounts to decide what to investigate next. They contribute different kinds of evidence.

Show coverage beside themes. If 8 of 12 participating leavers mention scheduling, say that. If 20 leavers were invited, also show 12 / 20 participation. The theme is present in 8 participant accounts; it is not evidence that scheduling caused 8 departures or describes the views of all 20 leavers.

Define whether an account can receive several theme labels. If it can, totals across themes may exceed the number of participants. Keep examples, coding rules and human review available to the people authorised to validate the analysis, while avoiding identifying quotations in a broad report.

See the exit interview analysis guide for a fuller approach to coding accounts and reviewing interpretation.

Convert a pattern into competing explanations

Ask what else could produce the same result. If new-starter departures increased on an evening shift, plausible explanations might include a mismatch between advertised and actual hours, transport difficulties, limited access to training or a change in the starter mix. These are hypotheses, not findings.

For each explanation, record both supporting and conflicting evidence. A claim about unavailable training becomes weaker if starters consistently received the promised training and describe a different problem. A site comparison becomes weaker if one location recruited mainly temporary workers during a peak while the other recruited permanent specialists.

Use the employee turnover causes guide to build the investigation. Avoid selecting a favoured explanation solely because it matches the intervention already budgeted.

A turnover review template to reuse

The following is a suggested working template. The entries are illustrative and are not reported customer outcomes.

Agree the decision rule before reviewing results. A pilot with better retention but unmanageable trainer workload may need redesign. A pilot that was never delivered cannot establish whether the intended intervention works.

Report what changed and what remains uncertain

A short before-and-after comparison can support a next step without proving causality. Report concurrent changes, the observation window, the number of people involved and the precision the data can reasonably support. Small groups can shift sharply after a single departure.

If a credible comparison is available, examine whether the pilot and comparison groups followed similar patterns beforehand and whether other changes affected them differently. For consequential claims, involve an analyst in the evaluation design. Avoid attributing every avoided departure or every estimated replacement cost to one programme.

Keep measurement proportionate: a checked spreadsheet may be enough for an initial review. A more complex tool becomes useful when data volume, update frequency or coordination genuinely exceeds that workflow.

Where employee conversations fit

When records show a pattern but do not explain daily work, Lontra can help prepare a focused employee conversation campaign and provide a manager brief for human follow-up. The manager brief is distinct from a raw conversation transcript. The business team still owns the turnover calculation, interpretation and action evaluation.

Explore how Lontra supports employee conversations. Bring one question, such as where new starters lack practical support, and decide in advance how managers will respond to what they learn.