Frame the workforce decision and its evidence · Lesson 3

Metric denominators and cohort construction

12 min read

Lesson objectives

  • Calculate FTE, vacancy, turnover, and retention with explicit denominators and entry/exit rules.
Reading tools

An HR leader asks, 'Turnover rose to 18%; do we have a crisis?' Before answering, establish numerator, denominator, period, and population. Does 18% mean leavers over twelve months divided by average workforce size, or departures from a new-hire cohort? Does it include fixed-term contract completion, internal transfers, or people who never started? A bare number does not distinguish an actionable risk from expected movement. A useful metric is a definition contract: it says which event counts, who was eligible to enter the denominator, how long they were observed, and what happens to people joining or leaving during that period. There is no formula that suits every decision. Consistency and fitness for purpose matter more than a familiar label. For a workforce plan, expected departures from a scarce skill may matter more than organization-wide turnover because the gap depends on capability and replacement lead time.

Write a metric card. In this example, a descriptive annual turnover rate equals external leavers during the year divided by the average of opening and closing workforce counts. That is a simplified denominator when daily exposure is unavailable, and the approximation should be stated. A position vacancy rate equals authorized fillable positions unoccupied at a snapshot date divided by all authorized fillable positions on the same date; an unapproved requisition is not a seat. Retention for a January intake equals January starters still employed after a year divided by all January starters, provided each has had a full year of follow-up. FTE is not a rate of events: it sums effective work fractions for a defined period. Before comparing two measures, check that snapshots, locations, contract types, and exclusion rules match. Arithmetic can be flawless while the managerial inference is wrong because the populations differ.

For a cohort, fix both the entry event and observation length. Suppose 40 people started in January 2025. On January 31, 2026, 32 remain employed and eight have left the company. One-year organization retention is 32/40 = 80%. If five of the 32 moved to another team, they are retained by the company but no longer available to the original team. Original-team retention is 27/40 = 67.5%. Neither figure is inherently wrong; each answers a different supply question. Do not add later starters to this cohort's denominator or treat a newer intake with less than a full year of observation as directly comparable. Where follow-up lengths differ, a time-to-event method may be more appropriate than a simple cohort percentage. At this point it is enough to recognize when the simple denominator becomes unfair.

Validation goes beyond repeating the division. Plot numerator and denominator separately. A falling rate may reflect a rapidly growing denominator even as more people leave. Look for spikes in entry dates rather than event dates, and for HR-system migrations or changed vacancy definitions. Break the measure down by critical occupation and tenure without making categorical claims from tiny groups or exposing identifiable individuals. If a dashboard shows 0% vacancies for a team with zero authorized positions, the denominator is zero and the rate is undefined, not evidence of perfect staffing. The rate records what happened. It does not establish why it happened, which requires a comparison and operational context.

When rates look comparable but follow-up differs

Two teams launch hiring programmes and a manager asks for six-month retention. The first team's cohort joins in January, while the second finishes joining in April. A June report gives each cohort's retained share. The rates appear ready to compare, but the second team has not had equal follow-up. Record each person's entry, observation end, and six-month anniversary. The current report answers retention through June, not six-month retention for everyone. Both results can be useful if their timing is explicit. Hiding the difference can make hiring speed look like a difference in retention quality.

Define the exit event before counting people. It might mean leaving the organization, leaving a team, or completing a temporary assignment. A worker transferring departments leaves her team while remaining with the organization. For organizational knowledge loss, she counts as retained. For a particular shift's capacity, her move reduces that shift's supply. Put the event definition in the table heading and preserve movement type in source records. This protects comparisons when another analyst reuses the report for a different purpose, and prompts the decision owner to clarify what the worker's departure actually changes.

Watch for records that disappear without a confirmed exit. A changed employee identifier, delayed branch update, or incomplete import may explain disappearance. Mark follow-up as incomplete until status is established and keep the unresolved count visible. Silently removing such people from the denominator can raise retention merely because information was lost. Calling all of them leavers can misclassify a register problem as a talent problem. First reconcile identities and movements. Pending that work, show bounds with all unknowns retained versus all exited. The interval shows how missing information affects the decision without pretending to know individual status.

For instance, a fifty-person cohort has forty confirmed retained workers, six confirmed exits, and four unknown statuses. Retention is at least eighty percent if all unknowns exited and at most eighty-eight percent if all stayed. If a fictional internal escalation threshold is eighty-five percent, resolving those four statuses could reverse the escalation decision. Reconciliation is then operationally important. With a ninety-five percent threshold, both bounds give the same decision, allowing an operational review to start while records are reconciled. These are case assumptions, not universal professional thresholds. They show how denominator uncertainty can change which task should be prioritized.

A report owner may ask to include permanent contracts only. Define that choice before measurement and apply it to numerator and denominator together. Suppose a cohort includes twenty permanent and ten temporary workers. Two permanent and three temporary workers leave. Permanent-contract retention is eighteen of twenty, or ninety percent, while all-contract retention is twenty-five of thirty. The first rate is valid when the question truly concerns permanent contracts. Presenting it as retention for the whole organization would hide an entire worker group. If a person's contract type changes during observation, state whether classification follows the entry contract or report-date contract. Choose the rule before seeing results so the definition remains consistent.

Use full denominator labels rather than the word rate alone: six-month cohort retention, external exits per one hundred person-months, or vacancies among approved positions. Each represents a different operating question. At a monthly review, test one transferred worker, one external leaver, and one missing record. If colleagues classify them differently, repair definitions before adding another chart. That small test helps preserve reporting meaning over time. Record definition changes and effective dates so readers can distinguish workforce change from changes in how the workforce was counted.

Lesson summary

An HR leader asks, 'Turnover rose to 18%; do we have a crisis?' Before answering, establish numerator, denominator, period, and population.