Employee Turnover: What It Costs and How to Count It

The rate is an argument about the denominator

The formula everyone quotes is departures divided by average headcount over the period, multiplied by 100. The numerator looks obvious and the denominator looks like a technicality. It is the other way round. Take one company's year: 200 employees on 1 January, 240 on 31 December, 33 people left. Average headcount is (200 + 240) ÷ 2 = 220, so the rate is 33 ÷ 220 = 15.0%. Now compute it the three other ways people actually compute it. Against closing headcount: 33 ÷ 240 = 13.75%. Against opening headcount: 33 ÷ 200 = 16.5%. And of those 33 departures, 8 were fixed-term contracts that simply reached their agreed end date; strip those and the rate against average headcount is 25 ÷ 220 = 11.36%.

Four defensible methods, one dataset, answers from 11.36% to 16.5% — a spread of 5.1 percentage points, and the highest is 1.45 times the lowest. Nothing in that range is a lie. What makes it useless is comparing your 11.36% to somebody else's 16.5% without knowing which convention either of you used. The rule that fixes it is not "use my method", it is: publish the convention next to the number, and never change it mid-series.

One more denominator trap, for growing companies. The (opening + closing) ÷ 2 shortcut assumes headcount rose smoothly. If the whole of that 40-person increase landed in December, the true average — the mean of thirteen monthly readings — is 203.08, and the rate is 16.25% rather than 15.0%. In a company that hires in bursts, average headcount should be computed from monthly payroll counts, not from two dates.

Voluntary, involuntary, regretted, not

A single blended rate hides the only distinction that matters. Split the same 33 departures: 11 people resigned and the company wishes they had not; 9 resigned and the company is relieved; 5 were dismissed; 8 were fixed-term contracts ending. On average headcount those are rates of 5.0%, 4.09%, 2.27% and 3.64%. The regretted voluntary rate — 5.0% — is the one that measures whether good people want to stay. It is a third of the blended 15.0%, and it is the only one of the four that a retention programme can be judged against.

This is also where national law makes cross-border comparison genuinely hard, and it is worth being explicit rather than presenting one country's rules as universal. What counts as a leaver, whether a fixed-term contract reaching its end date is a departure at all, how long a notice period runs, what a dismissal requires and how much it costs: all of these differ substantially across markets, and the statistical offices that publish turnover series do not all classify them the same way. The mechanism is the same everywhere — a person stops working for you and the seat has to be refilled — but the legal packaging around it is not, so a rate imported from another country needs its definition imported with it.

Costing one departure, component by component

You will see turnover costed as a multiple of salary — half, one, twice. Treat every such figure with suspicion unless you can trace it to a named published study with a stated method, because the multiple depends entirely on the role, the market and what the author chose to include. Build it instead. Take a $60,000 role with employer social contributions of 30% — a rate you must replace with your own, because it is one of the things that differs most between countries. That gives a fully loaded cost of $78,000 a year, or $6,500 a month, which is the unit every component below is measured in.

Vacancy is the output not produced while the seat is empty. Assume two months and assume the team covers half of the work through overtime or a contractor: 2 × $6,500 × 50% = $6,500. Recruitment is agency fee plus internal time: 20% of salary is $12,000, and forty hours of hiring-manager and interviewer time at a loaded $50 an hour is $2,000, giving $14,000. Onboarding, equipment, IT provisioning and induction training: $3,000 as a stated flat figure. A colleague spending a tenth of their time mentoring for three months: 0.10 × 3 × $6,500 = $1,950.

The ramp is the term that dominates and the term nobody books. A new hire is paid in full from day one and is not fully productive for months. Assume six months to full effectiveness and an average productivity gap of 50% across that period — the shape matters less than the area under it — and the shortfall is 6 × $6,500 × 50% = $19,500. Add the five components and one departure costs $44,950: 74.9% of salary, 57.6% of fully loaded cost, with the ramp alone at 43.4% of the total. Every one of those numbers moves with your assumptions, which is exactly the point of writing them down. Change nothing but the ramp — 3, 6, 9 or 12 months — and the total moves to $35,200, $44,950, $54,700 or $64,450, that is 58.7%, 74.9%, 91.2% or 107.4% of salary.

What the whole year costs, and what it costs to fix

Scale the unit cost. Thirty-three departures at $44,950 is $1,483,350 in a year, against a payroll of 220 × $60,000 = $13,200,000 — 11.24% of salary spend, none of it appearing on any line of the income statement. Take only the eleven regretted resignations and it is $494,450. That second figure is the honest budget for a retention programme: if a change in pay, management or working conditions costs less than $494,450 a year and removes even half of those eleven, it pays for itself. The blended figure of $1,483,350 would not be a fair target, because the eight fixed-term endings and most of the five dismissals were not failures to be prevented.

Why a small team's turnover rate is noise

One person leaving a team of five is a 20% turnover rate. Two is 40%. There is no value in between, because the rate can only take the values 0, 20, 40, 60, 80 and 100 — the arithmetic has no resolution below twenty points. Model it properly: if every one of those five people independently has a 15% chance of leaving in the year, the number who actually leave is binomial. The probabilities are 44.37% that nobody leaves, 39.15% that one does, 13.82% that two do, 2.44% for three, 0.22% for four and 0.01% for all five.

Read those numbers back as a management problem. A team with a perfectly ordinary 15% underlying rate will report 0% turnover 44.37% of the time and 40% or worse 16.48% of the time. Put together, it reports either "perfect" or "crisis" in 60.85% of years, and the manager did nothing different in either case. The standard deviation of the observed rate makes the same point in one number: 15.97 percentage points at five people, 11.29 at ten, 7.98 at twenty, 5.05 at fifty and 2.52 at two hundred. At n = 5, an observed rate of 30% or more happens 16.48% of the time by chance alone; at n = 50 it happens 0.53% of the time.

The practical consequence is a reporting rule. Do not publish a turnover rate for a unit small enough that one departure moves it by more than a few points, and do not run a performance conversation off one. Aggregate small teams into a function, or lengthen the window to three years so the denominator is person-years rather than people. If you must look at a single small team, look at the count of departures and the reasons given, which carry information, rather than at a percentage that is mostly arithmetic artefact.

Frequently asked questions

Should fixed-term contracts that reach their end date count as turnover?

It depends what you are measuring, and the honest answer is to report both. If you are asking how much replacement work the organisation generates, they count — the seat still has to be refilled. If you are asking whether people want to stay, they do not, because nobody chose to leave. In our dataset the difference is 15.0% including them and 11.36% excluding them. National practice varies too: countries differ in how freely fixed-term contracts can be used and in whether their statistical offices treat an expiry as a separation, so state your treatment explicitly.

Is turnover cost really about the same as annual salary?

Sometimes, and only by coincidence. Our build gave 74.9% of salary for a role with a two-month vacancy and a six-month ramp. Lengthen the ramp to twelve months and the same build gives 107.4%. Shorten it to three and it gives 58.7%. The answer is not a constant; it is a function of time-to-fill, time-to-productivity and how expensive your hiring channel is. Any single multiple quoted without those inputs is describing somebody else's role in somebody else's market — build your own and label every assumption.

Why is the ramp the biggest component if the new hire is working?

Because they are paid in full while producing less than full output, and the gap runs for months rather than weeks. Six months at an average 50% shortfall on a $6,500 monthly loaded cost is $19,500 — 43.4% of our total and more than the recruitment fee. It never appears in any budget because there is no invoice: payroll pays the same amount it would pay a fully productive person. That invisibility is why it is systematically left out, and leaving it out is what makes turnover look cheaper than it is.

What is a good turnover rate?

There is no universal number, and anyone quoting one without naming a published source, a sector and a counting convention is guessing. Turnover varies enormously by industry, by seniority, by contract type and by country, and even official series are built on different definitions. The useful comparison is against yourself: the same convention, the same population, quarter after quarter, split into regretted and not. A regretted voluntary rate that is rising is a signal whatever its level; a blended rate compared to a benchmark you cannot trace is not.

How long a window should the rate be measured over?

Long enough that the denominator carries some statistical weight, which for most organisations means a rolling twelve months rather than a quarter. A quarterly rate on a small population moves violently for reasons that have nothing to do with management: at five people the standard deviation of the annual rate is already 15.97 points, and a quarterly window makes it worse. Rolling twelve months, computed monthly, gives you a smooth series that still turns when something real changes.