Defining a Recruiting Funnel Denominator

A recruiting funnel denominator is the eligible set that could have experienced the measured event. It is not automatically an ATS, profile, application or requisition total. The guide below helps an analyst define a process measure with bounded counts. It makes no benchmark, forecast, quality, fairness, causal or outcome claim.

What question and unit does a funnel denominator need?

Write the decision question first: among which units, during which window, what share reached this event? For example (fictional): “Among unique candidates entering the approved first-screen stage for role version v2 from 1 July to 31 August 2026, what share had a screen disposition by the maturity cutoff?” Count one defined unit in the denominator; people, applications, attempts and requisitions are different estimands.

Unit	Example denominator	Appropriate question	Common error
Unique candidate	Candidates who entered a role's first screen	What share progressed to a manager review?	Counting the same person once per message or stage update
Application	Applications submitted to one requisition	What share of applications reached a screen?	Mixing applications with sourced profiles
Screening event	Completed or scheduled screens	What share of screens produced a disposition?	Treating a cancellation as a completed screen
Requisition	Roles opened in a period	What share of requisitions reached an approved close event?	Using requisitions as a candidate denominator

Choose the unit that matches the event. A changed unit is a new metric, not a relabelled one.

What goes in a denominator contract?

Complete this contract so another analyst can reproduce the eligible set.

Decision question and owner:
Unit of analysis:
Role, requisition or segment:
Process and definition version:
Entry event and entry timestamp:
Numerator event and event timestamp:
Observation window and timezone:
Maturity or closure rule:
Inclusion criteria:
Exclusion criteria:
Duplicate and re-entry rule:
Missing, withdrawn and not-applicable treatment:
Source table, query or export version:
Calculation, sensitivity cases and stop conditions:
Verification date and refresh owner:

A source field is not automatically an eligible-population field; see the dated CDC collecting-data source below.

Label each value observed, calculated, assumed, unknown or not applicable. Keep assumptions separate from observations; never place one in the denominator as if observed.

Which records belong in the denominator?

Inclusion describes who or what had the defined opportunity. Exclusion removes records outside the question, never inconvenient negative results.

For a unique-candidate stage measure, require the named role or requisition, an entry event inside the window, a stable approved identifier or documented match rule, a known process version and source, and the same entry definition.

Possible exclusions are test or rehearsal records, out-of-scope roles, post-cutoff entries and duplicates represented by an approved record. State each reason and count. Withdrawals, rejections and non-response remain in the declared event or missingness rules.

funnel share = units with the defined event / eligible units in the denominator

If the denominator is 0, missing or not reproducible, report the counts or design status and leave the share blank. A verified zero event count is not the same as an unavailable denominator.

How to align cohort, window and maturity

Entry and event dates answer different questions. Choose the cohort timestamp (entry, event, close or another approved field), timezone, boundaries and extraction cutoff.

An open record has not had the same opportunity as a mature record. Choose and consistently apply a maturity rule such as “event date at least 14 days before extract.” If it cannot be checked, label a snapshot; never count open as a final non-event.

Show the window and maturity cutoff together:

Field	Example decision record
Entry window	1 July through 31 August 2026, Hong Kong time
Entry event	First approved screen scheduled for role version `v2`
Numerator event	Screen disposition recorded as completed
Maturity cutoff	14 full days after the scheduled screen date
Open records	Reported separately and excluded from completed-event share
Process version	`v2`, before the reminder-policy change

Do not compare mature and immature cohorts as a funnel change. A changed stage definition, reminder policy, source, owner or system needs a new process version or bridge.

How to handle duplicates and re-entry in the denominator

Preserve raw rows, eligible units and every duplicate reduction.

Choose and record one rule:

Situation	Denominator treatment	What to preserve
Same candidate imported twice for one role	Count one approved unique candidate	Both raw IDs, match reason and retained ID
Candidate applies to two separate roles	Count once per role-specific metric, or use a person-level cross-role metric	Role IDs and the approved scope
Candidate re-enters the same stage after a withdrawal	Count one person if the question is person-level; count events only if the question is explicitly event-level	First entry, re-entry reason and version
Two screen attempts for one candidate	Count one candidate for candidate conversion; count two events for a screen-attempt metric	Attempt IDs and candidate link
No reliable match key	Do not guess a merge	Unmatched rows and denominator hold

Never deduplicate by name or similar profile text alone. Use an approved identifier or documented matching review. A re-entry can be a new opportunity, but that changes the estimand and must be stated.

How to treat missing data in a funnel metric

Missing history, unknown closure, withdrawals, deleted identifiers and unavailable exports are data states, not automatic non-progression.

At minimum, separate five data states:

Event observed: the qualifying event is recorded.
Closed non-event: a valid close without the event.
Open: unresolved at the cutoff.
Unknown: the source cannot establish the event or eligibility.
Not applicable: the event does not apply.

Show denominator and missingness counts beside every share. Moving unknowns silently changes the story without changing arithmetic.

For a closed-subset diagnostic, the analyst may calculate:

closed-subset share = observed event / (observed event + closed non-event)
closure coverage = (observed event + closed non-event) / eligible entering units

Label the first result as a closed-subset diagnostic when open or unknown records remain. It is not final cohort conversion. Without an approved imputation method, material missingness stays unknown with a named source or process check.

How to show definition risk with sensitivity cases

Sensitivity changes one documented rule or input at a time while keeping the base case visible; it is neither a confidence interval nor a benchmark.

Fictionally, 80 unique candidates entered a screen: 32 qualifying events, 28 closed non-events, 12 open and 8 unknown. The denominator is 80, but unresolved maturity or unknown rules prevent calling its share final.

An analyst can show bounded descriptive cases:

Case	Numerator	Denominator	Share	Interpretation
Base, all eligible entering units	32	80	40.0%	Snapshot; 12 open and 8 unknown remain visible
Closed-subset only	32	60	53.3%	Diagnostic among 32 events plus 28 closed non-events
Unknowns resolved as qualifying events	40	80	50.0%	Sensitivity case only; not an observed result
Unknowns resolved as closed non-events	32	80	40.0%	Sensitivity case only; not an observed result

These figures are fictional arithmetic, not a benchmark, target or Talent Summoner result. Investigate unknowns, verify open records and decide whether the metric remains on HOLD; the range does not estimate unseen outcomes.

Change one rule at a time (unit, duplicate match, maturity, scope, version or missingness). Do not combine favourable choices into an “adjusted” result without an owner-approved method. Keep raw, eligible, excluded, open and unknown counts.

When should an analyst put a funnel metric on HOLD?

Mark the metric HOLD when:

the question, unit, events, role, window, timezone, process version or maturity cutoff is vague;
units are mixed;
open or unknown records are treated as non-events or zeros;
duplicate, re-entry, exclusion or source transformations cannot be reproduced;
the denominator is zero, unavailable, target-derived or from another population;
a rate is being used as a benchmark, probability, quality, causal or ROI claim; or
no named analyst and decision owner can verify it.

Next disposition: use as scoped context, repair the source, collect a mature cohort, run a defined sensitivity, revise the metric or close the comparison. A documented unknown is valid; do not rank from one unstable share.

Where Talent Summoner fits

Talent Summoner is our product for candidate sourcing and candidate ranking, not funnel measurement. Its candidate-sourcing workflow, checked 28 September 2026, starts from a role brief and returns public-source discovery material for human review. Your team defines stages, verifies entry, deduplicates and owns the calculation.

Talent Summoner does not validate denominators, close cohorts, resolve identities, infer missing events, provide benchmarks or predict replies, interviews, offers, hires, quality, speed or ROI. It is not an ATS or application-pipeline manager. Use pricing only for current purchase options, then return to candidate sourcing when approved discovery is in scope.

What is the denominator in a recruiting funnel?

The denominator in a recruiting funnel is the eligible units that could have experienced the defined event under the same role, stage, process version and time rule. Name whether the unit is a unique candidate, application, event or requisition, and show the count beside the share.

Should I use all candidates in the ATS as the denominator?

Only if the question and inclusion rule say that every ATS record belongs to the same eligible population. Otherwise, filter to the defined entry event and scope. Profiles, applications, screens and requisitions are different units.

How should open or missing records affect the metric?

Keep open and unknown records separate from closed non-events. You may report a labelled closed-subset diagnostic, but do not present it as final cohort conversion. If the missingness rule is not approved, mark the result unknown or HOLD.

How do duplicates and re-entry change the denominator?

They change what is being counted. Count one record per person for a person-level measure, or one event per attempt for an event-level measure. Preserve raw IDs, matching decisions and re-entry reasons so another analyst can reproduce the rule.

Not by itself. A share can change because the role mix, eligibility, timing, stage definition, source coverage, missingness or maturity changed. It is a bounded process description, not a quality, fairness, speed, hiring or ROI conclusion.

Can Talent Summoner calculate our recruiting funnel denominator?

No. Talent Summoner can support role-based sourcing and supplied-material ranking. Your authorised analyst and recruiting owners define the population, verify records, handle deduplication and missingness, calculate the measure and decide what action follows.

Sources checked

SHRM: Assessing Recruitment Conversion Rates Can Lead to Better Hires, published 20 September 2017; checked 5 September 2026. Funnel-methodology reference, not a benchmark.
CDC Field Epidemiology Manual: Collecting Data, chapter published 8 August 2024; checked 5 September 2026.
CDC Field Epidemiology Manual: Describing Epidemiologic Data, chapter published 8 August 2024; checked 5 September 2026.
CDC Field Epidemiology Manual: Analyzing and Interpreting Data, chapter published 6 January 2025; checked 5 September 2026.

Reopen the selected methodology sources before you rely on them, preserve the source and query versions, and keep the metric on HOLD when the unit, denominator, cohort maturity, missing-data rule or owner is unclear. After the measurement brief is approved, use candidate sourcing only for its separate discovery workflow.