Calculate Hiring Funnel Conversion Without Losing the Denominator
Follow a fictional 40-applicant cohort through four stages, separating advancement, employer declines and candidate withdrawals.
A hiring funnel becomes misleading when it combines different candidate groups or treats every person who leaves as a rejection. Build one small table before drawing conclusions. The following example is entirely fictional, with an intentionally closed cohort so every outcome can be reconciled.
Define the cohort before calculating
Imagine one vacancy with 40 completed applications received during a fixed application window. All 40 have reached a final outcome by the reporting date. Each row below counts people, not interviews or assessment attempts. We exclude duplicate applications from this invented dataset and count no candidate twice within a stage.
| Stage entered | People entering | Advance | Employer declines | Candidate withdraws |
|---|---|---|---|---|
| Application review | 40 | 20 | 16 | 4 |
| Initial interview | 20 | 8 | 10 | 2 |
| Work sample | 8 | 4 | 3 | 1 |
| Final interview | 4 | 1 offer | 3 | 0 |
Every row balances. For example, the work-sample row has eight entrants: four advance, three are declined and one withdraws. The last row ends at offer creation, not acceptance or employment start.
Calculate each transition with its own denominator
Application-to-initial-interview conversion is 20 divided by 40, or 50%. Initial-interview-to-work-sample conversion is 8 divided by 20, or 40%. Work-sample-to-final-interview conversion is 4 divided by 8, or 50%. Final-interview-to-offer conversion is 1 divided by 4, or 25%.
Overall application-to-offer conversion is 1 divided by 40, or 2.5%. Multiplying the transition rates gives the same result: 0.50 × 0.40 × 0.50 × 0.25 = 0.025. That agreement is a useful arithmetic check, not evidence that the process is good.
Keep withdrawal separate
Seven people withdraw across the four stages: 4 + 2 + 1 + 0. That is 17.5% of the original cohort. Thirty-two receive an employer decline and one receives an offer; 7 + 32 + 1 = 40.
Do not describe all 39 non-offer outcomes as rejections. Likewise, do not call the offer a hire until acceptance and start outcomes are recorded. If it is declined, add that outcome without rewriting the earlier interview counts.
Ask a question the table can answer
The table shows where people left this fictional process. It cannot explain why. The two withdrawals at the initial interview might reflect timing, changed interest or something else entirely. A follow-up reason field can provide context if available; the rate alone cannot diagnose a scheduling problem or an assessment defect.
It also cannot show whether a work sample measures the right competency. OPM describes work samples as job-like tasks; stage conversion measures movement through the process, a different question. OPM: Work Samples and Simulations.
For an open cohort, add an “awaiting outcome” column and a reporting date. Never quietly treat candidates still in progress as failures just to make the table look complete.
When an automated tool influences those transitions, monitoring the AI hiring tool requires additional checks beyond funnel arithmetic.