HR metrics and people KPIs that actually matter (and how to calculate them)
Six numbers, each pointing at a specific part of the system — starting with the one everybody asks about first. Not a dashboard project: a way of finding out what is failing before someone resigns.
HR metrics earn their place when a bad reading tells you which part of the organisation to go and look at. Employee turnover is the clearest example: it is the share of people who left over a period, and you calculate it by dividing leavers in the year by the average headcount across that same year. Below three hundred people, turnover and five other people KPIs cover almost everything a leadership team needs.
People analytics usually arrives in a growing company as a dashboard nobody opens. Fourteen tiles, most of them headcount cut four ways, answering a question nobody asked. The six below are chosen on the opposite principle: each one exists to send you somewhere.
How to calculate employee turnover
The formula, in words: take the number of people who left during the period, and divide it by the average number of employees over that same period. Average headcount is the headcount at the start plus the headcount at the end, divided by two. Use the same period boundaries every time, or the trend is measuring your calendar rather than your company.
A worked example, with round invented numbers purely for illustration. A company starts the year with ninety employees and ends it with a hundred and ten, so its average headcount for the year is a hundred. Twelve people leave during that year. Turnover is twelve leavers for every hundred employees.
Then do the part most companies skip. Of those twelve, decide how many were departures you were sorry about. If seven were, regretted turnover is seven for every hundred employees, and that is the number a leadership team should be looking at. The headline figure mixes together people you wanted to keep, people you did not, and people who moved country.
Six HR metrics worth the space
| Metric | Formula in words | How often | What it signals |
|---|---|---|---|
| Turnover, total and regretted | Leavers in the period divided by the average headcount over the same period | Quarterly | Management quality, more often than compensation |
| Rating distribution by manager | People at each point of the scale, per manager, over that manager's team size | Once per cycle | Calibration, or its absence |
| Time in level, by level | Average months people have already spent at the level they are currently on | Twice a year | Whether progression is real or theoretical |
| Pay dispersion within level | Highest salary in the level divided by the lowest salary in the same level | Annually, and at every pay review | Whether the pay framework survives contact with hiring |
| Time to productivity for new joiners | Weeks between the start date and the date the manager considers the person autonomous in the role | At every joiner, reviewed quarterly | Onboarding, and role clarity before that |
| Span and layer count by function | Direct reports per manager, and the number of layers between the front line and the leadership team | Twice a year, and at every reorganisation | Structure, and management load |
The second one only means anything if there is a performance management system behind it. Without one, a rating distribution measures the manager rather than the work.
When each one is worth a conversation
- Regretted turnover — when it concentrates in one team, or in the six-to-eighteen-month band.
- Rating distribution — when one manager's distribution is visibly different from their peers' and the work is comparable.
- Time in level — when a level has people well past the typical time and no promotions out of it.
- Pay dispersion — when the spread inside a level starts to overlap the level above.
- Time to productivity — when it is longer than a quarter, or when nobody can say what it is.
- Span and layers — when a manager has more than nine reports, or a layer exists with spans of two.
Each metric maps onto a specific part of how the company runs its people — the same parts a People Diagnostic examines. That is the point of the selection: the numbers are not there to be reported, they are there to tell you which part to examine.
Two of them deserve more explanation
Regretted attrition, not attrition
Regretted attrition requires a judgement recorded at the time rather than reconstructed later. Ask the manager, in the week of the resignation, one question: would you rehire this person tomorrow? The answer is unreliable six months on, and unrecoverable a year on.
Cut by tenure, it becomes diagnostic. Departures inside six months usually indicate a hiring or onboarding problem. Departures between six and eighteen months usually indicate a management problem: the person understood the job, and chose to leave anyway. That distinction is worth more than the headline rate.
Pay dispersion within level
Widening dispersion inside a level is the earliest visible sign that a pay framework is being eroded by hiring. It shows up long before anyone complains, because complaints only start once two people compare offers.
The reading that should worry you is overlap: when the top of one level extends past the bottom of the next. At that point the levels no longer explain the pay, which means salary bands are describing something other than how the company actually pays.
What to skip
Below three hundred people, several popular HR metrics cost more than they return.
- Engagement scores as a headline number. A single figure moving from 7.2 to 7.4 tells a leadership team nothing they can act on. The free-text answers underneath it are worth reading; the number is not worth reporting.
- Cost per hire. Real, and dominated by which roles you happened to open. It measures your hiring mix more than your hiring.
- Training hours. An input measure. It records that time was spent, not that anything changed.
- Nine-box talent grids. They assume a calibration discipline most companies this size do not yet have, and they produce confident-looking labels from unreliable inputs.
How often, and with whom
Quarterly, in the leadership meeting, on one page. Monthly is too frequent to show signal on any of these, and annual is too late to act on the ones that matter.
The review should take twenty minutes and produce at most one investigation. Six numbers that reliably generate one good question a quarter are worth considerably more than a dashboard that generates none.
Data does not fix anything. It tells you where to look, which is the part most companies are guessing at.
If none of these are currently measurable, that is itself the finding, and it is the one a People Diagnostic usually surfaces first.
Co-founder — Organisation & Operations
Francisco Campos
Built growing companies from the inside: first employee to COO at Onport through its acquisition by Farfetch.
More from Francisco