Hiring metrics that show speed, candidate friction and job outcomes

Hiring metrics measure how candidates move through recruiting, what the process costs and what happens after someone starts work. A useful dashboard connects each measure to a decision—not just a target.

Start with three questions: Where are candidates waiting? Where are they leaving or being rejected? Are hires succeeding in the job? If you use AI screening, add a fourth: did the tool improve outcomes, or simply process applications faster?

Eight hiring metrics to start with

The time-to-fill, time-to-hire, offer-acceptance and cost-per-hire definitions below follow common recruiting metric conventions. The stage and post-hire measures are recommended operational definitions; document them before comparing results.

For percentage formulas, multiply the result by 100.

Metric	Calculation	What to check or change
1. Time to fill	Calendar days from requisition approval to accepted offer	Separate time spent attracting applicants from approvals, scheduling and selection.
2. Time to hire	Calendar days from the hired candidate’s application—or first recruiting contact—to accepted offer	Find delays within the candidate’s journey. Use one entry-date rule consistently.
3. Time in stage	Stage exit date minus stage entry date	Locate slow screening, interview scheduling or feedback. Track current age separately for people still waiting.
4. Stage conversion rate	Candidates advancing ÷ candidates entering that stage	Examine selection criteria alongside employer declines, withdrawals and pending cases.
5. Offer acceptance rate	Offers accepted ÷ offers made in the same offer cohort	Review declined-offer reasons, compensation, role expectations and elapsed time. Show pending and employer-withdrawn offers separately; do not treat them as candidate declines.
6. Cost per hire	Internal plus external recruiting costs ÷ hires in the same reporting period	Include recruiter and interviewer time, agency fees, advertising, assessments and allocated software costs. Agree the cost boundary with finance.
7. 90-day retention	Eligible hires still employed on day 90 ÷ all hires whose day-90 date has been reached	Review early departures alongside onboarding, job expectations and working conditions. Separate voluntary and employer-initiated exits.
8. Job-outcome attainment	Hires meeting predefined job milestones at a fixed review point ÷ eligible hires assessed	Check whether hires can do the work. Report missing assessments separately.

For time metrics, show the median, number of cases and a tail measure such as the 90th percentile. An average alone can hide a small group waiting much longer. Time to fill and time to hire here describe accepted-offer cases; report open requisitions and still-waiting candidates separately so their delays do not disappear.

Treat job-outcome attainment as one component of quality of hire, not a universal score. Define observable, role-specific milestones before hiring—for example, completing a task independently to an agreed accuracy standard. Keep retention, manager feedback and milestone results separate rather than hiding them in a weighted composite.

These outcomes also reflect training, management and workload. They are not solely a recruiter’s performance measure.

Set the counting rules before setting targets

Every dashboard needs a short data dictionary:

• Unit: use a candidate–requisition record for funnel measures. One person applying to two jobs creates two records; duplicate submissions to one job should not inflate the funnel.
• Timing: distinguish a cohort report, which follows the same applicants, from a period report, which counts activity during a month.
• Status: separate advancement, employer rejection, candidate withdrawal and pending decisions. Record known technical failures separately.
• Scope: compare similar role families, seniority, locations and hiring routes. Record changes to assessment thresholds and process steps.
• Ownership: name who verifies the data and who can fix the underlying problem.

For retention, include only cohorts with enough elapsed time to observe day 90—including people who left earlier. For job outcomes, show assessment coverage: a strong result based on half the eligible hires is incomplete. Label conversion and offer-acceptance rates as provisional while decisions remain pending.

Use your own comparable cohorts as the starting baseline. Before adopting an external benchmark, check its population, dates and start/end definitions. “Time to hire” measured to acceptance is not comparable with a measure ending on the employee’s first day.

Read conversion alongside candidate exits

Consider a fictional, fully resolved screening cohort:

• 100 applications entered screening.
• 30 advanced.
• 50 were declined by the employer.
• 20 candidates withdrew.

Screening conversion is 30 ÷ 100 = 30%. Employer declines are 50%; withdrawals are 20%. Calling all 70 non-advancements “screening rejections” would conceal candidate exits.

A lower conversion rate could reflect stricter job-related criteria, a different applicant pool or a process candidates abandon. Inspect the reason codes before changing sourcing or screening thresholds. For a longer calculation, see the worked hiring-funnel example.

Add a brief candidate-experience survey at selected stages, including for unsuccessful applicants. Report response counts and response rates, not just satisfaction scores: respondents may differ from people who never answer.

When AI is involved, add checks beyond speed

A faster résumé-screening stage is a process result, not evidence that the tool identifies better employees. The U.S. Office of Personnel Management’s assessment guidance distinguishes validity evidence, including whether assessment performance relates to subsequent job performance.

For an AI-assisted stage, retain the tool version, configuration, threshold, recommendation and final human decision under defined access and retention rules. Compare:

• Stage advancement rates and counts by role and, where lawfully collected, demographic group.
• Assessment completion, technical failures and reported accessibility problems.
• Human overrides and their documented reasons.
• Later job outcomes for comparable hiring cohorts.

Do not label rejected applicants “false negatives” merely because a reviewer disagrees with the AI. Their performance in the job is generally unobserved. A review sample can identify overlooked job-relevant evidence, but it does not establish a job-performance error rate.

An internal dashboard is not a statutory bias audit. As of October 28, 2026, NYC’s official AEDT guidance describes independent-audit calculations covering selection or scoring rates and impact ratios across sex, race/ethnicity and intersectional categories. Whether those requirements apply depends on the tool and its use. This is informational guidance, not legal advice; confirm jurisdiction-specific obligations with qualified counsel.

Assign each recurring review one decision and one owner: fix scheduling delays, investigate assessment withdrawals, review changed selection rates or revisit job milestones. For tool-specific review triggers and pause rules, use the post-deployment monitoring guide.