How to calculate 90-day new-hire retention by cohort

A hotel can report improving retention simply because its latest recruits have not had time to leave. The remedy is a cohort measure: follow a defined group of starters for the same length of time, then compare like with like.

For hospitality HR teams, 90-day new-hire retention is useful for reviewing recruitment expectations, induction and early working conditions. It does not tell you why someone stayed or left. Start with a defensible calculation, then use operational evidence to decide what needs attention.

Define exactly what counts as staying

Use this working definition: the percentage of eligible starters who remain continuously employed within the agreed organisation boundary through their 90-day milestone.

Define the start date as day zero and the milestone as start date plus 90 calendar days. If your organisation calls the start date day one instead, document that convention and use it consistently. Do not confuse three calendar months with 90 days.

State whether an employment end date means the last employed day or the first day no longer employed. Under a last-employed-day convention, an employee whose end date equals the milestone is retained through that date. Check this with payroll before writing the formula.

• Organisation boundary: a hotel, employing company or group. Report site retention separately from group retention.
• Eligible starters: for example, external recruits into roles intended to last at least 90 days.
• Continuous employment: leaving and later returning does not erase the original exit.
• Employment status: an authorised absence is not automatically an exit. A missing payslip is not sufficient evidence of termination.

A transfer between hotels can preserve group retention while reducing retention at the original property. The dashboard should show both facts, rather than allowing payroll system changes to decide the answer.

Keep the denominator tied to starters

90-day retention = starters continuously employed through their milestone ÷ eligible starters in the fully observed cohort × 100.

The denominator is not average headcount, the number of current employees or the people who completed induction. Removing employees who left during induction would remove precisely the experience you are trying to measure.

Group starters by their actual start month or another declared intake period. Record department and property at entry so that later transfers do not silently rewrite the cohort. Keep subsequent moves in separate fields.

Decide seasonal eligibility before inspecting results. If a six-week contract is outside your 90-day measure, exclude that contract type consistently and report contract completion separately. Do not exclude a permanent recruit retrospectively because their employment ended early. Show how many starters were outside scope and why.

Wait until the cohort has matured

A monthly cohort is fully observed only when its latest eligible starter has reached the milestone date, or would have reached it had they stayed. Apply the same reporting cut-off to everyone.

For the headline rate, leave the whole immature cohort pending, including its known early leavers. Including the leavers while excluding people still working towards day 90 produces an unfairly low result. Counting all current starters as retained produces an unfairly high one.

The statistical issue is incomplete follow-up, often called right censoring. NIST’s explanation of censoring describes observations whose eventual event time is not yet known. Its examples concern reliability testing; the relevant principle here is that an unfinished observation is not a completed outcome.

If you need an earlier view, show separate 30-day and 60-day results for cohorts eligible for those milestones, plus observed exits to date. Label each clearly. A survival analysis can use differing follow-up lengths, but it needs appropriate expertise and assumptions about censoring; it is not the same calculation as a completed-cohort percentage.

A worked hospitality example

Hypothetical example, not an industry benchmark. A group reviews two fully observed monthly intakes using continuous group employment:

• January: 40 eligible starters, 32 retained through their milestone. Retention is 32 ÷ 40 × 100 = 80%.
• February: 50 eligible starters, 45 retained. Retention is 45 ÷ 50 × 100 = 90%.
• Combined: 77 retained from 90 starters. Retention is 77 ÷ 90 × 100 = 85.6%, rounded to one decimal place.

The simple average of 80% and 90% is 85%, but that gives the smaller January intake the same weight as February. Aggregate the underlying counts instead.

Suppose four of January’s retained employees moved to another hotel in the group. Group retention remains 80%. If all four left their original property and the other 28 stayed there continuously, original-property retention is 28 ÷ 40 × 100 = 70%. Neither figure is wrong; they answer different questions.

Build a small, auditable cohort file

Use one row per employment spell. Restrict identifiable records to the people who need them; the management report normally needs counts and rates, not names.

• Employee identifier and employment-spell identifier.
• Actual start date, cohort and milestone date.
• Entry property, department, role family and contract type.
• Eligibility flag and documented exclusion reason.
• Employment end date, its date convention and exit category.
• Internal movement dates and organisation boundary.
• Reporting cut-off, source-system refresh date and validation status.

Investigate duplicate starter records, retrospective payroll corrections and rehires. An accepted offer that never starts belongs in a pre-start conversion measure, not among employment starters. A person who starts and leaves on the same day belongs in the cohort if they meet its eligibility rules.

Do not treat an unresolved record as a confirmed exit or quietly remove it. Resolve it before final publication where possible. Otherwise show the unresolved count and qualify or withhold the rate. Keep a revision note when later corrections change previous cohorts.

Turn the result into a useful decision

Read the count beside the percentage. One person’s outcome changes a cohort of ten by ten percentage points. Avoid league tables built from tiny departmental intakes; combine appropriate periods while retaining the original monthly detail.

Compare similar seasonal periods, role mixes and contract types. A higher rate after an induction change is encouraging, but changing pay, recruitment sources or occupancy pressure may also explain it. Use a comparison group before making a causal claim.

Pair the outcome with evidence from first shifts, roster changes and new-starter conversations. WINC Wire’s guide to onboarding surveys offers related reading on listening during entry. Its analysis of the weekly rota and retention provides another operational lens, not proof of the cause in your own cohort.

Finish each review with one decision: what will change, who owns it and which future fully observed intake will test the result. A stable definition and an honest waiting period make the measure more useful than a faster number that cannot withstand scrutiny.

Sources

• NIST/SEMATECH e-Handbook: Censoring

AI-generated editorial illustration.