It has its own almost-opposite problem in only counting employees who did leave.

Then you get into the field of survival analysis, where you get methods that can handle this problem (called right censoring IIRC).

https://en.m.wikipedia.org/wiki/Kaplan%E2%80%93Meier_estimat...

Exactly. If half your workforce is happy and has been there 20 years, that’s entirely non reflected in such a figure. There must be a better metric.

Well you can also just count current employees' tenure. That fixes the problem and is qualitatively consistent IMO.

That’s how you get 3 years because everyone is new.

You're right - my mistake.

I thinkit is impossible to get this number

What do you mean? Every employer I've worked for had data for the starting date of their employees, when someone leaves you have an end date of employment. It's a pretty simple metric to calculate.

I think by "this number" they mean "a better metric", maybe?