Search for recruiting KPIs and you’ll find plenty of lists. Some count six metrics, some twelve, some twenty-three. Whatever the number, the metrics are almost always the same: time to hire, time to fill, cost per hire, offer acceptance rate, source of hire, quality of hire, and funnel conversion.
I have no argument with those lists. My argument is that they stop where the work starts. Knowing which metrics exist is the easy part. Deciding which ones are worth building in your situation is the hard part, and a list cannot do it for you, because the answer depends on what you are trying to do with the numbers and on what your data can support.
Here is the sequence I would use. Start with the job you are doing, take the shortlist that serves it, then put every metric on that shortlist through three checks before you build anything.
Start with the job you are doing
Across our research with recruiters and HR managers, requests for hiring metrics cluster around three jobs. It doesn’t matter who’s asking or how big the company is. What matters is that each job needs different metrics, and mixing them is how a dashboard ends up serving nobody.
Read the three descriptions below and find the one that sounds like your week. Most people recognise themselves in just one. Start there, and don’t worry about the rest of the list until a specific situation asks for it.
1. You are optimising the funnel
Your question is what to change. You treat a hiring process the way a performance marketer treats a landing page: find the step that is losing people, change one thing, see whether it worked.
| Metric | How it is calculated | Worked example |
|---|---|---|
| Stage-to-stage conversion | Candidates who reached the next stage, divided by candidates who reached this one. One rate per step, not one for the whole funnel | 80 reached screening, 20 reached interview. Screening to interview = 20 / 80 = 25% |
| Channel quality, measured downstream | Per channel: candidates who passed the first screen, divided by that channel’s applications. Then hires per channel | Job board: 12 of 120 passed screening = 10%. Referrals: 5 of 10 = 50%. The job board sent twelve times the volume and the same number of screen passes |
| Cost per application and cost per hire, by channel | Channel spend divided by that channel’s applications, then by that channel’s hires | 600 EUR on a job board, 50 applications, 1 hire. 12 EUR per application, 600 EUR per hire |
Order matters here. Conversion tells you where the problem is, channel quality tells you whether the problem arrived at the top, and cost tells you what to do about it. Read them in that order and you will rarely be wrong about what to change next.
One warning on the third. Cost per application means nothing on its own. Twelve euros is expensive or cheap depending entirely on what you compare it against, and most teams have nothing to compare it against. If you have no anchor, use the channel comparison above and ignore the absolute number.
Leave three metrics alone for now: time to hire, offer acceptance rate, and quality of hire. Time to hire is an outcome, so it’s hard to tie a change in it to anything you did. Offer acceptance rate sounds useful, but at most companies the annual number of offers is too small to read as a rate. Quality of hire has its own problem, covered in check two below.
If the symptom sits at the top of the funnel
Two symptoms point at the job ad rather than at your process: too few applications, and applications from people who are obviously wrong for the role. Neither shows up in a conversion rate, because the candidates you needed never entered the funnel at all.
This is worth measuring on its own terms, and it is the one measurement most recruiters I speak to have no way of doing. They rewrite titles and descriptions regularly, on instinct, and never find out whether it worked.
The fix is a deliberate before-and-after test on the job ad. Pick and change one thing: the job title, the seniority wording, the salary range, the opening paragraph, or the length of the required-skills list. Note the date. Then compare applications per week and screen-pass rate for the four weeks before against the four weeks after.
Two conditions make this honest. Change one variable at a time, or you will not know which one moved the number. And check the same period last year if you have it, because hiring volume is seasonal and a January rise is not evidence of a better title.
2. You are reporting upward
Your trigger for opening a dashboard is a recurring obligation: a management update, a board pack, a weekly summary. Your audience was not involved in the process and will not be reading for nuance.
| Metric | How it is calculated | Worked example |
|---|---|---|
| Roles filled against plan | Positions where a candidate accepted in the period, over positions you planned to fill in the period | 7 accepted against 10 planned = 7 of 10 |
| Time to fill, one declared start point, as a median | Days from your declared start point to offer acceptance. Report the median | Five roles at 21, 28, 35, 40 and 96 days. Median 35. The mean is 44, dragged up entirely by the one hard search |
| Candidates in process per open role | Active candidates, excluding rejected and withdrawn, divided by open roles | 48 active across 6 open roles = 8 per role |
| Source of hire | Share of the period’s hires by the channel the hire came from. Declare first touch or last touch | 7 hires: 3 job board, 2 referral, 1 agency, 1 direct |
Rows one and three do most of the work together. One says what you delivered, the other says what is coming, and having both closes the follow-up question you will otherwise get every month.
The median in row two is not a stylistic preference. One six-month executive search will drag an average far enough to make the number useless, and somebody in the room will know it is wrong without being able to say why.
Leave stage-by-stage conversion and cost per hire alone for now. Stage-by-stage conversion is out because your audience won’t read it. Cost per hire is worth including only if finance already owns a spend figure you can reuse without building anything new. And any metric you can’t reproduce next month is out too, since this kind of reporting only works if the numbers stay comparable over time.
Your real constraint in this job is rarely insight. It is the hour you spend rebuilding the same view every month, and a consistent time window plus a fixed definition removes most of it.
3. You are keeping the process moving
You care about timeliness. No candidate sitting too long, response commitments met, hiring managers nudged. You are acting on these numbers today, not reporting them next month.
| Metric | How it is calculated | Worked example |
|---|---|---|
| Days in current stage | Today minus the date the candidate entered their current stage. Set the threshold before you look | Threshold 7 days. Three candidates at 9, 12 and 15 days. That is your morning worklist |
| Time to first response | Working days from application received to the first substantive reply | Standard of 5 working days. Applied Monday the 3rd, answered Monday the 10th = 5 working days, exactly at the limit |
| Days since last activity, per open role | Today minus the date of the last recorded action on any candidate for that role | Role open 40 days, last activity 12 days ago. That role has quietly stalled |
| Interview-to-feedback turnaround | Working days from interview held to the hiring manager’s decision being recorded | Four interviews at 1, 2, 8 and 11 days. Median 5, and two conversations to have |
The first two are thresholds with an action attached rather than metrics in the usual sense. That is what makes them work. A cap of about a week in any stage, with somebody pushing the moment a candidate exceeds it, does more for a hiring process than any chart of average dwell time. The teams we talk to run informal versions of this constantly. One commits to a five-working-day first response and escalates to managing directors when it slips.
Row four is the one people avoid, because it measures somebody else. Measure it anyway. Interview feedback is the most common delay in a hiring process and the hardest to raise without evidence, and a median turnaround by hiring manager makes the conversation factual.
Leave alone for now: cost per hire, source of hire, conversion rates. All useful to somebody. None of them tell you what to do this morning.
If time to hire is the only number your team currently reports, compare it with five hiring metrics that beat time to hire.
The three shortlists side by side
| Your job | Track first | Leave alone for now |
|---|---|---|
| Optimising the funnel | Stage-to-stage conversion, channel quality downstream, cost per application and per hire by channel | Time to hire, offer acceptance rate, quality of hire |
| Reporting upward | Roles filled against plan, time to fill on one declared start point, candidates in process per role, source of hire | Stage-level diagnostics, cost per hire, anything not reproducible next month |
| Keeping things moving | Days in stage with a threshold, time to first response, days since last activity per role, interview-to-feedback turnaround | Cost per hire, source of hire, conversion rates |
Which of these have an actual standard
Three do. The rest do not, and knowing which is which saves you a great deal of arguing.
Cost per hire is the most standardised metric in recruiting and the least consistently reported. ANSI and SHRM formalised the formula in 2012: total internal recruiting costs plus total external recruiting costs, divided by the number of hires in the same period. ISO/TS 30407:2017 covers the same ground. The internal half is where it comes apart, because recruiter time, hiring-manager time, interview panel time and tooling are exactly what most teams leave out.
Time to fill is covered by ISO 30414:2025, the current human-capital reporting standard. Its public page confirms the scope, but not the detailed calculation, so declare your start point rather than relying on the label alone.
Quality of hire has had its own technical specification since 2018, ISO/TS 30411. A standard has existed for eight years. Only one in five organisations measures the metric.
The other eight metrics in the tables above have no standard at all. Days in stage, time to first response, screen-pass rate by channel, candidates per open role: there is no authority to appeal to. The definition is yours to write, and what matters is that you write it down and that everyone reading the number can see it.
Then put the shortlist through three checks
A shortlist is a hypothesis. Three questions turn it into a plan, and all three are cheap to ask now and expensive to ask after you have built a dashboard.
Check one: if this number moved, what would you do?
Start here, because there is no sense in measuring something you would never act on. This check is also what earns the word shortlist above.
Volume metrics fail it most often. Application count is the easiest thing in recruiting to measure and, for many teams, the least useful. Many of the recruiters we talk to say the same thing: the number of applications arriving is not their problem, and what they care about is how many of those people signed. Teams running high volume put it most bluntly, because their difficulty is finding the good candidates and knowing which channels produced them.
This matches where the wider field says it is going. The Josh Bersin Company and AMS argue that business outcomes are displacing operational hiring metrics as the primary measure of success, with organisations moving away from time to fill and cost per hire toward productivity and workforce capability. I would treat that as a stated position rather than a measurement, since the release discloses no sample or method, but the direction matches what I hear.
A metric that fails this check can still be worth producing. Just call it reporting and keep it separate, so it stops competing for attention with the numbers you actually act on.
The check also finds the opposite case, where the metric that would change a decision appears on no list at all. One recruiter told us that retrospective performance data didn’t help him much. What he actually needed was a way to justify spend before a role went live, not an explanation for it afterwards.
That raises a fair question: can you get a forward-looking number without past data? Not really. Every expectation about a role you have not posted yet comes either from somebody’s aggregated past or from the state of the market right now. Concretely, the variables are how many applications a comparable role attracted, how long it took to fill, what it cost, and what salaries comparable roles are advertising today. The first three are past data. If you have run enough comparable roles you can build them from your own history, though most teams have too few to trust the average. The fourth is present-tense market data, and it does not live in your ATS at all. That is why this particular ask feels permanently unmet, and it is worth being clear with yourself about which of the four you are actually missing.
Check two: can you compute it?
This check kills more KPI lists than any other, and it is the one almost never mentioned.
Every speed metric depends on stored stage-transition timestamps. Time to hire, time to fill, days in stage: each needs a record of when a candidate entered and left each stage. Plenty of systems keep no such history in a form you can reach. A common export gives you a status snapshot, where every candidate has a current stage and no trace of how they got there. You can count people in that file. You cannot recover a single day of elapsed time from it.
Funnel conversion has a similar dependency and catches people out more often, because conversion needs the people who left. If your export carries only active candidates, rejected and withdrawn candidates never make it into the count, and the funnel you build will look far healthier than reality.
Quality of hire is the extreme case. It needs data that mostly lives outside recruiting, in performance reviews and retention records, joined back to a hire from a year ago. Which is presumably why SHRM’s 2026 benchmarking report, surveying 4,657 respondents between 24 November 2025 and 23 January 2026, finds that only one in five organisations measures it at all. Not that they measure it badly. They do not measure it.
This is where recruiting analytics starts: before adopting any metric, ask the boring question. Does the underlying event exist in my data, with a timestamp, for candidates who are no longer in the process? Where the answer is no, you are looking at a data request with a different owner and a different timeline, and it belongs on a different list.
Asking this first reorders your priorities in a useful way. Teams commit to a speed dashboard, discover three weeks in that the timestamps were never stored, and end up with a spreadsheet maintained by hand. Checking first points you at source and funnel metrics, which usually carry a lower data bar, and gets you something real in the same three weeks.
Check three: does everyone read it the same way?
A metric is a word plus a calculation, and both of them drift. The failure is quiet: two people read one number as two different facts and never discover it.
The word. Cost per hire is the clearest example in the field, and you can watch it happen inside one organisation’s own publications. The most-quoted SHRM figure, an average of 4,129 USD, comes from a page published on 8 August 2016 reporting fiscal year 2015 data from 2,048 members, and it does not say which costs are included. SHRM’s 2025 benchmarking release puts the nonexecutive average at 5,475 USD. SHRM’s 2026 recruiting benchmarking report puts the nonexecutive median at 1,300 USD.
Three figures, one label, one organisation. They do not contradict each other, because they are different studies measuring different things: two averages and a median, drawn from different samples, almost certainly with different cost definitions. That is precisely the problem. A standardised formula has existed since 2012 and the published numbers still cannot be compared. Before you report a cost per hire to anyone, write down which costs you counted and whether you are giving them a mean or a median.
Time to fill and time to hire are the second example, and this one has a standard. ISO 30414:2025 is the current human-capital reporting standard. Its public page does not expose the detailed calculations, so this article does not attribute a precise time-to-fill definition to it. The practical problem remains: teams count from requisition submitted, requisition approved, or job posted. A standard exists, and internal definitions still need to be explicit.
The calculation. Even with an unambiguous word, defining a single funnel chart takes several independent decisions, and getting them wrong is what makes a chart unreadable rather than merely inaccurate.
- The start point. Time to hire counted from the day the job went live and time to hire counted from the internal kickoff meeting answer different questions. One recruiter deliberately counts from kickoff, to separate delay she caused from delay the hiring manager caused. Where the metric is used to hold anyone accountable, that choice is the whole metric.
- First arrival or every arrival. Move a candidate back a stage and forward again: do they count once or twice?
- Skipped stages. Someone sourced straight into an interview never entered screening. Do they count at screening? Leave them out and your bars show a drop that never happened.
- Reach or dwell time. These cannot share a population coherently. A candidate who skipped a stage still reached the next one, and contributes nothing to how long people spend in the stage they skipped.
- Automatic rejections. Candidates screened out by knockout questions were never assessed by a person. Counting them as reviewed overstates your workload and hides your real screening ratio.
A shared recruiting pipeline still needs one declared event model. None of these choices has a universally correct answer. Each needs an answer you can say out loud, because two people reading one chart with different assumptions are reading different charts.
Three habits close this out. Hand the metric to someone who did not build it, ask what it means and how they think it is calculated, and stay quiet until they have answered. If you have to explain it, rename it. Treat approval carefully, because liking a number is no evidence of understanding it, and that failure never gets flagged. And anchor everything to one declared total: if a summary card says 120 candidates and the rows beneath add up to 113, people will notice, they will do the arithmetic you did not show them, and they will quietly stop trusting the page.
The checklist
For every metric on your shortlist, in this order:
- Job. Which of the three jobs does this serve? If none, cut it.
- Decision. If this number moved by a third, what would I do differently? If nothing, move it to the reporting pile.
- Data. Does the underlying event exist, with a timestamp, including for candidates who left? If not, it is a data request.
- Definition. Written down: population, start point, denominator, and how automatic rejections and skipped stages are treated.
- Label. Tested on one person who did not build it, with no explanation first.
- Threshold. For anything you intend to act on: the number that triggers action, who acts, and what they do.
Six lines, and most candidate metrics fail at line two or three. That is the point of running them. Four metrics that pass are worth more than twenty-three that were never checked.
I would rather defend four metrics I can compute, explain and act on than publish twenty-three I cannot. That choice is open to you today, and it costs an afternoon.
Sources
- Join Research, our method and scope: who we talk to, how sessions are run, what we ask about, and what we never publish.
- SHRM, 2026 Recruiting Executives Benchmarking: Attracting Critical Talent. 4,657 respondents, fielded 24 November 2025 to 23 January 2026. https://www.shrm.org/topics-tools/research/recruiting-benchmarking/full-data-brief
- SHRM, Benchmarking Report: $4,129 Average Cost-per-Hire, published 8 August 2016, reporting fiscal year 2015 data from 2,048 members. https://www.shrm.org/topics-tools/news/shrm-benchmarking-report-4129-average-cost-per-hire
- SHRM, SHRM Releases 2025 Benchmarking Reports, published 15 October 2025. Nonexecutive cost per hire, average 5,475 USD. https://www.shrm.org/about/press-room/shrm-releases-2025-benchmarking-reports—how-does-your-organizat
- ISO, People power, ISO/TC 260, published 12 July 2021. Cited as background on ISO/TS 30407:2017 (cost per hire) and ISO/TS 30411:2018 (quality of hire), not as the current human-capital reporting standard.
- ISO, ISO 30414:2025 Human resource management — Requirements and recommendations for human capital reporting and disclosure, accessed 25 August 2026.
- ANSI/SHRM 06001.2012, the cost-per-hire standard. The standard itself is a paid document, so this article cites its existence and shape rather than its clause wording. ANSI’s approval of it as the first American National HR standard sponsored by SHRM is reported by CBIA, published 1 June 2012. https://www.cbia.com/news/hr-safety/ansi-approves-first-american-hr-standard/
- The Josh Bersin Company and AMS, Navigating the Talent Acquisition Revolution, press release 7 July 2026. Cited as a stated position; the release discloses no sample or method. https://www.prnewswire.com/news-releases/josh-bersin-company-research-shows-talent-acquisition-moving-beyond-ai-experimentation-302819410.html
Where the rest comes from
Everything not attributed above comes from Join’s own research with recruiters and HR managers, described in full on Join Research. Individual participants and their employers are confidential, so those observations appear here as patterns. Where an observation rests on a single conversation, I have said so.
