Analysis for people teams.
Three people can compute attrition correctly and get three answers, which is why the definition matters more than the arithmetic.
- Read5 min
- Sections4
How do I calculate staff turnover correctly?
Choose one denominator, write it down, and use it everywhere. Attrition on opening headcount, on closing headcount and on average headcount give three different numbers, and comparing figures built on different denominators is the most common mistake in this analysis. Then split by tenure before anything else, because losing first-year starters and losing five-year veterans are different problems.
01
The denominator decides the number
Opening, closing and average headcount are all defensible bases and produce visibly different rates, particularly in a year with growth or a restructure.
Pick one, state it next to every figure, and never compare against a published benchmark without knowing which basis it used. A rate computed a different way is not a comparison, it is a coincidence.
02
Tenure before team
Most leaving happens early. An organisation losing people in their first year has a hiring or onboarding problem; one losing five-year veterans has something else entirely, and a blended rate hides both.
Cohort survival from hire date is the view that separates them, and it is more useful than any split by department.
03
Small teams get counts
A team of six with one leaver is not a 17% attrition rate in any useful sense, and no manager should be ranked on a number that one departure would have reversed.
Report groups below the threshold as counts. It is less satisfying, it is honest, and it makes the sample size impossible to overlook.
Free-text leaver reasons and the structured field frequently disagree. That disagreement is the finding, not something to resolve quietly.
04
What to upload
A headcount export with hire dates, leave dates and team, which is enough for attrition by tenure and by team. Leaver reasons make the analysis considerably more useful and are usually in a separate report.
Identifiers are hashed as the file is read, so the analysis can follow a person between records without knowing who they are. No personal detail beyond dates and grouping fields is needed for any of this.
See it on a real project
The 22.4% rate combined three different events; the unrecovered no-show rate was 11.4%.
Why is our no-show rate 22%?Related
Try it on your own file.
Every check described here runs automatically, and your first findings arrive in full before you pay for any of it.
No account needed to start. You only pay when you like what you see.