Hiring Data

Technical recruiting funnel metrics: a worked example

Calculate technical recruiting funnel metrics with a complete fictional ledger, explicit stage denominators and separate pending and withdrawn applications.

An open ledger with a blue pencil beside a metal funnel and blank paper cards.
On this page
  1. 1. Define the cohort, opening and cutoff
  2. 2. Distinguish applications from duplicate records
  3. 3. Reconstruct the complete stage-entry ledger
  4. 4. Calculate observed advancement
  5. 5. Show what remains unresolved
  6. Keep timing and interpretation within scope
  7. Frequently asked questions
  8. Sources

Calculate technical recruiting funnel metrics by following a defined group of applications through recorded stage entries. Divide the applications entering the next stage by those entering the preceding stage, and state the reporting cutoff. Pending cases and withdrawals need explicit treatment before you call that percentage a conversion rate.

The complete fictional example below lets you reproduce each count. It contains no real applicants, Prepin results or industry benchmarks.

1. Define the cohort, opening and cutoff

Choose the applications first, then follow their progress. Greenhouse distinguishes cohort, activity and snapshot reports: a cohort follows a selected group, activity counts events during a period, and a snapshot describes a point in time. Those views answer different questions.

For this example, use the following report definition:

Field Definition for this worksheet
Opening R1, one hypothetical backend engineering opening
Counting unit One canonical application to R1
Cohort Applications received November 2–6, 2026, inclusive, using UTC dates
Measurement cutoff November 20, 2026, at 23:59:59 UTC
Progress recorded All stage-entry and disposition events through that cutoff
Stage sequence Applied → recruiter screen → technical screen → final interview → offer issued → offer accepted
Special paths No skipped stages, reentry or backward moves in this example

An October application that had an interview in November is outside this cohort. A November 6 application that reached an offer on November 14 stays inside it. The cohort dates select applications; they do not restrict every later event to November 2–6.

Define the stage before counting it. The technical screening guide explains that assessment’s purpose and limits. Here, entering the technical stage does not establish that an interview was completed or passed.

Use the actual responsibilities when separating openings. A title in the backend engineering catalog provides role context, not a reporting definition or permission to combine different jobs.

2. Distinguish applications from duplicate records

One person can have more than one application. Ashby’s candidates-versus-applications documentation explains that job and stage information belongs to individual applications. Someone applying to two openings therefore represents two application units when both openings are in scope, even though they are one person.

Duplicate export rows are a different issue. Preserve an application ID, opening ID, event ID, event type, timestamp and source-record reference so you can inspect what repeated.

This invented raw export has two known exceptions:

  • Repeated event: A03’s technical-entry event, e03-tech, appears twice with the same event ID. Keep one occurrence; it adds one technical entrant.
  • Duplicate application record: A10-copy is explicitly confirmed to be an accidental second record of A10’s same application to R1. Map that source record to A10 and retain the mapping. Eleven raw application records become ten canonical applications.

Those identity facts are supplied as part of the exercise. They are not a rule to merge real people because their names match. Uncertain duplicates need review; legitimate applications to different openings must remain separate.

For this worksheet, count each canonical application once per stage it actually entered. Keep the underlying event history and correction notes instead of silently deleting inconvenient rows.

3. Reconstruct the complete stage-entry ledger

Every cleaned application and its later entries appear below. All dates are in November 2026, UTC. The numbers in the third column are dates, not scores or durations.

Stage key: R = recruiter screen; T = technical screen; F = final interview; O = offer issued; A = offer accepted. No listed entry means that stage had not been entered by the cutoff.

Application Applied Later stage entries: stage and date Status at cutoff
A01 Nov 2 R 3; T 5; F 9; O 12; A 13 Offer accepted; employment start not recorded
A02 Nov 2 R 4; T 6; F 10; O 14 Awaiting offer reply
A03 Nov 3 R 4; T 9; F 12 Rejected Nov 16 after final
A04 Nov 3 R 5; T 10 Awaiting technical decision
A05 Nov 4 R 6; T 11 Withdrew Nov 13 after technical
A06 Nov 4 R 6 Rejected Nov 9 after recruiter screen
A07 Nov 5 No later entry Rejected Nov 6 before recruiter screen
A08 Nov 5 No later entry Withdrew Nov 7 before recruiter screen
A09 Nov 6 R 12 Awaiting technical scheduling
A10 Nov 6 R 10; T 13 Rejected Nov 17 after technical

Count actual entries here, rather than assuming someone visited every earlier stage because they reached an offer. All preceding entries are explicitly recorded in this fixture.

That choice matters when reading vendor reports. Greenhouse’s pipeline-history report includes bypassed intermediate stages in its totals. Ashby’s funnel explanation also describes inferred progress through skipped or earlier stages. A reported stage total therefore need not mean that many interviews occurred.

If your process allows skips, document whether you report actual entry or inferred reach. Label inferred reach separately; do not manufacture an interview event. This example rules out skips so its arithmetic stays fully inspectable.

4. Calculate observed advancement

For this linear cohort, the calculation is:

Observed advancement = applications entering the next stage by cutoff ÷ applications entering the preceding stage by cutoff × 100.

The cumulative entry counts are 10 applied, 8 recruiter, 6 technical, 3 final, 2 offers issued and 1 offer accepted. Each later group is a subset of the preceding group in this fixture.

Transition Calculation Observed advancement at cutoff
Applied → recruiter screen 8 ÷ 10 80%
Recruiter → technical screen 6 ÷ 8 75%
Technical → final interview 3 ÷ 6 50%
Final → offer issued 2 ÷ 3 66.7%, rounded
Offer issued → offer accepted 1 ÷ 2 50%

For example, technical entrants are A01, A02, A03, A04, A05 and A10. Three reached final: A01, A02 and A03. That produces 3 ÷ 6 = 50% observed advancement through November 20.

A03’s later rejection does not erase the earlier move from technical to final. A05’s withdrawal also does not erase the technical entry. Stage history and ultimate disposition answer different questions.

Overall observed application-to-offer-acceptance is 1 ÷ 10 = 10%. Call it that, rather than a hire rate: the fixture records no employment start. These percentages describe the invented applications only.

Copyable report caption:

R1 applications received November 2–6, 2026, UTC; one canonical application per unit. Actual stage entries measured through November 20, 2026, 23:59:59 UTC. Observed advancement uses all preceding-stage entrants, retaining pending cases and withdrawals. No skipped, repeated or backward stage moves in this example. Offer acceptance is not employment start.

5. Show what remains unresolved

The technical stage reconciles into four disjoint outcomes at the cutoff:

Technical-stage outcome Applications Count
Advanced to final A01, A02, A03 3
Rejected without advancing A10 1
Withdrew without advancing A05 1
Still awaiting a decision A04 1
Total technical entrants 6

The headline fraction, 3 ÷ 6 = 50%, retains the pending application. It is not final cohort conversion. Its other 50% is not a rejection rate: that remainder contains a rejection, a withdrawal and an unresolved case.

You can separately calculate advancement among resolved technical-stage outcomes, retaining withdrawals:

3 advanced ÷ (3 advanced + 1 rejected + 1 withdrawn) = 60%.

This denominator excludes pending A04. It answers a different, selective snapshot question. It is not automatically a better or unbiased comparison, and it does not predict how the full cohort will finish. A08 withdrew before technical, so it belongs in the overall cohort but cannot enter technical’s exit denominator.

Reconcile the whole cohort too: 1 accepted + 4 rejected + 2 withdrawn + 3 still active = 10 applications. The active cases are A02, A04 and A09. Counting only current occupants would miss applications that entered a stage and later left it.

Ashby’s documented historical funnel focuses on completed considerations; our worksheet deliberately shows both all-entrant and resolved-outcome fractions. Do not assume that a similarly labelled dashboard uses either denominator. Read its filters and stage rules before comparing results.

Keep timing and interpretation within scope

Report calendar boundaries alongside any duration. A01’s November 2 application and November 13 acceptance are 11 calendar-date boundaries apart, excluding the starting date. The ledger supplies day stamps, so it cannot establish exact elapsed hours or business days.

A02 has no acceptance date. Its November 2 application is 18 calendar-date boundaries before November 20; that is age at cutoff, not a completed time to acceptance. Entering zero for its missing duration would misstate the record. A completed-case average would need that label and explicit pending exclusions.

Before comparing cohorts, align the opening scope, stages, denominators and follow-up time. A small unfinished funnel cannot tell you why someone withdrew, whether an assessment measured the right skill, or whether a tool improved hiring. The AI hiring-tool pilot guide addresses a separate evaluation task.

Keep this report definition beside the export and repeat the reconciliation when pending cases resolve. Prepin runs technical interviews by voice and live coding and provides a scored report for human review. See how screening works.

Frequently asked questions

Is a recruiting funnel the same as current pipeline counts?

No. This funnel counts cumulative entries from a fixed application cohort through a cutoff. Current pipeline counts describe where applications are now; someone who advanced or exited may no longer occupy a stage they previously entered.

Should withdrawn applications disappear from conversion rates?

Not from the all-entrant denominator defined here. Keep withdrawals visible, and give any measure that excludes them a separate name and denominator. Attribute a withdrawal to its recorded stage, rather than every stage the application never reached.

Can one person count twice?

Yes, in application-based reporting when the person has separate applications to two in-scope openings. A repeated export of the same application is not another application. This worked example filters one opening and uses confirmed record mappings.

What if a stage has no entrants?

A conversion fraction with a zero denominator is undefined, not 0%. Show “not applicable” or an equivalent explicit label and the underlying counts. Do not manufacture a rate to fill an empty dashboard cell.

Sources

Prepared with AI assistance. Official documentation checked November 21, 2026. The ledger, counting policy and calculations are original fictional teaching material, not an ATS report reproduction or measured hiring results.

Oleh Savchuk, founder of Prepin

Oleh Savchuk

Founder, Prepin

Oleh Savchuk is the founder of Prepin, an AI hiring marketplace where recruiters screen engineers with AI-run technical interviews and candidates practice for theirs. He is a software engineer in Philadelphia and writes about technical hiring, AI in recruiting and interview preparation.

Screen engineers before the first call

Send candidates one link. Prepin runs the technical voice interview and live coding, then gives you a scored report.

See how screening works →

Practice the interview out loud

Run a realistic AI mock interview for your target role and get feedback on every answer.

Start a mock interview →