Agree what enters each calculation before asking a dashboard to explain the result.
Two consultants can report different offer acceptance rates and both have done the arithmetic correctly. One divides the acceptances recorded this month by the offers sent this month. The other follows the offers sent this month until each has an outcome. They are answering different questions.
Put that difference in the definition beside the number. Your next review can then ask why offers were declined, rather than spend its time reconciling two reports.
This glossary gives you a set of working conventions to adapt with your team. It covers the calculation, the records counted, the clock and the awkward cases. The conventions are choices, not a claim that every recruitment platform defines these KPIs the same way.
Put the counting rule beside the name
“Time to fill” is a useful place to start. Workable describes a clock from job publication to accepted offer. JobAdder's time-to-hire example starts at first candidate contact. Copying a metric name into your glossary does not settle which events your team should use.
For the definitions below, an application means one candidate considered for one vacancy. The same person considered for two vacancies has two applications. An offer means one formal offer for an application, with revisions attached to that offer rather than counted as new offers. A vacancy means one approved position; split a multi-hire job into positions before using the time-to-fill convention.
Use calendar days for the elapsed-time measures here: subtract the start date from the end date in the agency's agreed reporting timezone. The same day counts as zero. Weekends remain in the result. A business-day measure needs its own name and a specified workweek, holiday calendar and treatment of partial days. Don't compare it with the calendar-day series as if the clock were unchanged.
Each report also needs a selection window and an observation cutoff. The window chooses the records; the cutoff tells you how long you have followed their outcomes. A period count counts events occurring within a window. A cohort measure follows a group selected by a shared entry event, such as offers first issued in October. Later outcomes can belong to that group without becoming October events.
Time to fill and time to hire
These measures use different starting points. Keep them separate so an early candidate search does not quietly change a vacancy clock.
Time to fill: accepted vacancies
Calculation: sum of calendar days from vacancy approval to accepted offer, divided by the number of eligible vacancies with an accepted offer. Unit: mean calendar days per vacancy.
Select vacancies whose offer was accepted during the reporting window. Use the approved-position date as the start and the first valid accepted-offer date as the end. That is this glossary's convention; it differs from Workable's publication-based clock.
Exclude cancelled positions, test records, duplicate vacancies and positions without a reliable start or end date. Show those exclusions as counts. Report still-open positions separately with their age as at the cutoff; a completed-vacancy average cannot tell you how long unfinished searches have already waited.
If an accepted candidate later withdraws, retain the recorded acceptance event and flag the failed start separately. If the vacancy reopens, keep the original episode for this series and begin a clearly identified new episode. Do not overwrite the earlier clock to improve a historical result.
Time to hire: accepted applications
Calculation: sum of calendar days from first recorded contact about that vacancy to accepted offer, divided by the number of eligible accepted applications. Unit: mean calendar days per application.
Select applications with an offer acceptance in the reporting window. Start at the first recorded contact about the particular vacancy, whether initiated by the candidate or consultant. End at the acceptance event. An old general conversation with the candidate does not start this vacancy's clock.
Exclude duplicate applications, test records and applications missing a reliable clock event, and disclose the counts. Applications still in progress sit outside this completed measure and need their own ageing view. If your team instead starts at application received or first interview, rename the definition and preserve the older series rather than merging them.
For either measure, a median is a different summary from a mean. To calculate it, order the eligible durations and take the middle value, or the average of the two middle values for an even-sized group. Label which summary you show and include the record count. Neither summary makes different role types directly comparable.
Conversion measures that keep the same group
The denominator should identify the group you are following. The numerator then counts the qualifying outcomes from that same group, through the stated cutoff. Do not borrow this month's numerator from one group and its denominator from another.
| Measure | Numerator | Denominator | Selection and observation | Exclusions and unresolved cases |
|---|---|---|---|---|
| Submission-to-interview conversion | Applications in the selected group with at least one completed client interview | Eligible applications first submitted to the client in the window | Select by first authorised client submission; observe completed interviews through the cutoff | Exclude duplicates, tests and withdrawn-before-send drafts. Count repeat interviews once. Retain applications withdrawn after submission in the denominator; show pending cases separately |
| Offer acceptance, issued-offer cohort | Formal offers from the selected group accepted by the cutoff | Eligible formal offers first issued in the window | Select by issue date; acceptance must belong to that offer | Exclude tests, duplicate offers and unsent drafts. Keep pending, declined and employer-withdrawn offers in the denominator, with separate status counts. Revisions do not create new offers |
| Starts from accepted offers | Accepted offers in the selected group that led to a confirmed start by the cutoff | Eligible offers accepted in the window | Select by acceptance date; observe the actual start, not its planned date | Exclude tests and duplicates. Retain withdrawals and starts still pending in the denominator; show them separately. Count a start once |
These are deliberately inclusive denominators. They keep unsuccessful outcomes visible, but a recent group may have had little time to progress. Compare groups at a similar observation age, or show how much remains unresolved. If you also want an acceptance rate among decided offers, publish it as a separate measure with its own denominator.
Workable's offer acceptance formula divides offers accepted by offers extended. The table makes the local handling of pending offers and month boundaries explicit. It is that handling, rather than the percentage sign, that needs agreement.
Hypothetical arithmetic: the month-end offer report
The following numbers are invented solely to explain the calculations. They are not customer results, benchmarks or targets.
Suppose a team issues ten eligible formal offers in October. By 31 October, six have been accepted, two declined and two remain pending. One pending offer is accepted on 3 November; the other is declined on 5 November. Two other acceptances recorded in October belong to offers issued in September.
| Question | Calculation | Result and meaning |
|---|---|---|
| What share of October-issued offers was accepted by 31 October? | 6 ÷ 10 × 100 | 60%; two outcomes remain pending |
| What share of decided October-issued offers was accepted by 31 October? | 6 ÷ (6 + 2) × 100 | 75%; pending offers are excluded from this separately labelled denominator |
| What share of October-issued offers was accepted by 5 November? | 7 ÷ 10 × 100 | 70%; all ten offers now have an outcome |
| How many acceptance events occurred in October? | 6 from October-issued offers + 2 from September-issued offers | 8 acceptance events; this is a period count |
Dividing those eight October acceptance events by the ten October offers would produce 80%. It would mix groups, so it is not the issued-offer cohort acceptance rate defined here. The extra two acceptances belong in October's activity count and in September's offer cohort.
If no eligible offers were issued, the cohort rate is not available, with denominator zero. It is not 0%. A zero rate means eligible offers existed and none had been accepted by the cutoff.
This is why a revised cohort result need not mean the earlier report was wrong. The cutoff changed. Keep the original snapshot and its cutoff, then show the later observation as an update to the same group.
Source share and recruitment cost need their own boundaries
Application source share
Calculation: eligible applications attributed to a named source, divided by all eligible applications in the selected group, multiplied by 100. Unit: percentage of applications.
Select applications first received in the reporting window. For this convention, attribute each to its first recorded acquisition source for that vacancy and keep “unknown” as a source category within the denominator. Observe source records as at the report cutoff. Exclude tests and merged duplicates, with their counts disclosed; do not exclude an application simply because the person later withdrew.
This measures the composition of applications. It does not show which source caused a hire or returned the best value. A candidate's first acquisition source and the channel that reopened a conversation may be different fields. Agree which question you want before changing the attribution rule.
Recruitment cost per start
Calculation: agreed recruitment costs recognised in the reporting period, divided by confirmed starts in that same period. Unit: currency per start. State the currency, cost policy and reporting period beside it.
This is a period-based management measure, not the cost of the exact cohort of people who started. October costs can support people who will start later, while October starts may reflect work paid for earlier. Workable's recruitment-cost guidance stresses matching the variables' time period; even then, you still need to explain what the result represents.
Agree the included costs with finance, such as allocated consultant delivery cost and recruitment advertising, and how shared costs, credits and tax are treated. Exclude worker wages passed through on temporary assignments from this chosen recruitment-cost pool; do not present the resulting measure as temporary-desk gross margin. Exclude test starts and duplicates, and disclose missing costs rather than treating them as zero. If there are no eligible starts, report the rate as unavailable and show the cost total separately.
Employer cost per hire, an agency's cost per start and agency profit answer different questions. If your decision concerns margin or an AI investment, use the agency profitability and ROI guide for that separate calculation. No benchmark in this glossary tells you what your agency should spend.
Give each definition an owner and a change record
Copy this entry into your team's glossary and complete it before adding a metric to a report:
Metric name and version:
Decision it supports:
Unit counted: candidate, application, offer, vacancy or another specified unit
Numerator and denominator, or duration calculation:
Entry event and reporting window:
Outcome event and observation cutoff:
Calendar or business-day rule and timezone:
Exclusions, duplicates, pending cases and missing values:
Source fields and correction process:
Accountable owner and date agreed:
Have the owner reconcile a small authorised sample to the source events. A status entered late may change an event-date report, while a corrected duplicate can change a denominator. Retain the reason for the correction and identify which report version readers are seeing. Where Australia's Privacy Act applies, OAIC APP 10 guidance also addresses reasonable steps for personal-information quality. That guidance does not choose your KPI conventions for you.
Once the definitions are settled, use them to ask operational questions. The legal recruitment dashboard guide shows how a review can lead to a named next action. This glossary supplies the counting rules underneath that conversation.
FluxHire's AI builds recruitment reports for people to review. Its recruitment analytics reports never change your records. Bring your agreed glossary to an early-access evaluation and ask which definitions, source records and review steps the team can support. The benefit to test is whether the prepared report helps you inspect the desk's work with a shared understanding of the numbers. This glossary is not a promise that each formula is available as a built-in report.
Request early access with one reporting question and the definition behind it. Confirm supported records, integrations and review steps with the team, then inspect the prepared report against your source sample before relying on the result.