How RolePark calculates turnover and eNPS
The two methodologies behind RolePark workforce metrics. Formula, what goes into the numerator and the denominator, a worked example, and — separately — the limits beyond which the number stops being comparable.
Employee turnover
Turnover for a period = leavers in that period ÷ average daily headcount of the same period × 100, rounded to one decimal place.
The denominator is an average of daily headcount snapshots, not the headcount at the start or the end. In a company that is growing or shrinking, either edge skews the result: the same number of leavers divided by a smaller starting headcount gives a higher percentage than the same leavers divided by a larger closing one. Averaging over days removes that choice.
Worked example. Four people left during a quarter. Headcount sits around 50, and ten more were hired mid-quarter. The average daily headcount for the quarter is 55. Turnover = 4 ÷ 55 × 100 = 7.3%. With the opening headcount (50) as denominator it would read 8.0%; with the closing one (60), 6.7%. The gap between those options is wider than many real changes, so how the denominator is built matters as much as the data itself.
- The figure covers the selected period and is NOT annualised. A quarterly 7.3% is not "29% a year": annualising assumes the rest of the year behaves the same, and we do not know that.
- With no data for the period the metric is empty, not zero. Zero would mean "nobody left", which is a different statement.
- The numerator counts every employment that ended in the period, including contractors and interns. If you expect classic turnover for permanent staff only, this number will read higher than you expect.
- A breakdown by reason for leaving is available against the same denominator, with an honest "no reason given" remainder that is not redistributed across the other categories.
- Next to the metric you can open the explanation: the formula in words, what the calculation includes and excludes, and the methodology version. If the methodology ever changes, the older version stays attached to the periods it was used for.
eNPS and how anonymity is protected
eNPS uses the standard scale: 9–10 promoters, 7–8 passives, 0–6 detractors. The result is the percentage of promoters minus the percentage of detractors, an integer between −100 and 100. Passives are excluded from the numerator but included in the denominator — so their share lowers the result, and that is not a calculation error.
Anonymity here comes from how the database is built, not from a promise in the interface. The answers table has no person field: no user, no employment, no invitation — only the survey run, the score, the answers, a comment, the department and a tenure bucket. Who responded is known to a separate invitations table, and the application never joins the two. Answers carry no timestamp either, so the order in which they arrived cannot be reconstructed.
A slice is shown only from five answers. One threshold is not enough: if exactly one slice would stay hidden, its value can be derived by subtracting from the total. So in that case the smallest visible slice is hidden as well. The same threshold applies in analytics, so it cannot be worked around through a different chart.
- This is NOT a secret ballot. Anyone with database access sees both tables; the protection is that the link between them does not exist in the data, not that looking at it is forbidden.
- Free-text comments are shown only above the threshold and are detached from the order of answers.
- Participation is reported as three separate numbers — invited, started, answered — rather than one "response rate". Someone who opened the form and closed it is not the same as someone who never opened it.
- Surveys are switched on by the platform: by default only the People section of the HR side is available.