Time-to-Hire Is a Censored Measurement: Open Requisitions, Bimodal Medians, and a Reporting Spec That Survives Audit

Time-to-hire appears on nearly every talent acquisition dashboard, usually as a single average for the quarter. That average is almost always computed only over requisitions that closed, which means the slowest searches — the ones still open — are excluded by construction. The metric is systematically optimistic, and the error grows exactly when hiring gets harder. This piece covers the right-censoring problem, why the median hides structure across job families, why requisition-level and candidate-level definitions disagree by weeks, and a reporting specification we consider defensible.

The right-censoring problem

A requisition opened in the quarter and still open at quarter end has a duration that is at least the time elapsed so far. Its final value is unknown. Survival analysis calls this a right-censored observation, and the correct treatment is well established: censored records contribute to the estimate up to the point of censoring rather than being dropped.

Standard dashboard practice drops them. The resulting average describes only the requisitions that were easy enough to close inside the window. When the market tightens, more requisitions remain open at period end, more of the slow tail is excluded, and reported time-to-hire can improve while actual hiring speed deteriorates.

The fix is mechanical. Estimate time-to-fill with a Kaplan–Meier survival curve over all requisitions opened in the period, with filled_at as the event and the reporting date as the censoring point for open requisitions. Report the median from the survival curve and the share still open at fixed horizons — 30, 60, 90 days. Requisitions cancelled without a hire are a separate competing outcome, not a fill, and should be reported as such.

The median hides bimodality

Analysts who have moved past the mean often settle on the median. That is an improvement for skewed data and still insufficient. Time-to-fill across a whole organisation is frequently bimodal: high-volume roles with standing pipelines close quickly, while specialist and senior roles cluster at much longer durations. The pooled median can fall in the gap between the two modes and describe very few actual requisitions.

To illustrate — numbers constructed for demonstration, not measured: if 60% of requisitions are volume roles filling in about 20 days and 40% are specialist roles filling in about 75 days, the pooled median lands near the upper edge of the volume cluster, and a shift in mix toward specialist hiring moves the headline sharply with no change in either group’s speed.

The requirement is to stratify by job family and level before summarising, and to report mix alongside the summary so composition changes are visible.

Requisition-level versus candidate-level definitions

Time-to-hire is two different metrics sharing a name:

• Requisition-level time-to-fill runs from requisition opening (or approval) to offer acceptance. It measures how long the organisation carries a vacancy.
• Candidate-level time-to-hire runs from the hired candidate’s application or first contact to acceptance. It measures how long the selection process takes for the person who was chosen.

These can differ by weeks for the same hire. A requisition can sit open for a month before the eventual hire applies; a sourced candidate may have been in conversation before the requisition existed. Candidate-level figures are almost always shorter, which is one reason they tend to be preferred in external reporting.

Neither is wrong. They answer different questions, and quoting one under the other’s label is a common source of irreconcilable numbers between talent acquisition and finance.

A reporting specification

For a time-to-hire figure that will be cited outside the recruiting team:

1. Name the clock. State the start event (req_opened_at, req_approved_at, or candidate_applied_at) and the end event (offer_accepted_at or start_date).
2. Include open requisitions. Use survival estimation; report the censored share.
3. Separate cancellations. Report cancelled requisitions as a competing outcome with their own rate.
4. Stratify before summarising. Job family and level at minimum, with mix shown.
5. Handle evergreen and multi-headcount requisitions explicitly. A requisition filling 30 seats is 30 events, each with its own duration; a standing evergreen posting needs a defined start per hire.
6. Freeze the definition. A change in clock convention breaks the trend line and must be marked on it.

What we cannot claim

A correctly specified time-to-hire describes process duration. It does not say whether faster hiring produces better hires, and optimising the metric directly creates obvious pressure to shorten assessment. Nor does survival estimation remove bias from inconsistent timestamps; if requisitions are opened in the system weeks after the search actually began, every definition inherits that lag. The data always wins over the narrative — once the slow requisitions are allowed back into the data.