Where the 33% of salary turnover cost comes from

Provenance: Modelling assumption. The number is an input its author adopted for a calculation, not a result they measured. This describes how the figure is sourced, not whether it is true, and it is not a claim that anyone acted in bad faith.

You will see it in this form: replacing an employee costs about a third of their annual salary, often with a figure of $15,000 attached, usually credited to Work Institute. It is used to size retention business cases and to open a great many sales decks.

The figure is quoted as though someone measured replacement costs across a large sample of employers and found they averaged 33% of pay. Nobody did that, and the report everyone cites does not say they did.

What the source actually says

Work Institute's 2017 Retention Report does not report 33% as a result. It states that the firm uses an estimate of 33% of a worker's annual salary to calculate the cost of turnover, combining industry estimates with census data. That is a stated input to a model, and the report presents it as one.

The report also prints its own working, on page 9.

Read that twice. The percentage is calculated at one wage level, $16,640 a year, and then applied at another, $45,000 a year. Everything downstream follows from that single step, including the $15,000 per employee and the national totals built on it.

Following the endnote

The sidebar containing that arithmetic describes the $5,506 as coming from a meta-analysis. The report's endnote for it points somewhere else: a services page on a consulting firm's website, bylined and dated August 2015. It is not a meta-analysis and does not describe itself as one.

That page returns a 404 today. Anyone checking the figure now reaches a dead end unless they know to look in the Internet Archive, where a 2019 snapshot survives.

The archived page holds a table of fifteen turnover-cost estimates for a notional $8-an-hour employee, gathered from sources with very little to do with one another. An anecdote about one HR director at a company running 53 truck plazas. A supermarket cashier earning $6.50 an hour. A hotel front-desk employee in Miami. A legislative update from an Alaska state senator. Protective services staff, via a university staff senate. Different occupations, different wage levels, different decades, different definitions of what counts as a cost.

The page states its own method in a footer line: the average turnover cost for an $8.00-an-hour employee, using only the lowest ten of the fifteen estimates. So the number underpinning the most-cited statistic in employee retention is the arithmetic mean of a hand-picked subset of a mixed list.

One thing worth noticing, carefully

Among those fifteen inputs is a row attributing to the U.S. Department of Labor the claim that replacing a new hire costs one third of their annual salary. No Department of Labor publication is cited for it.

So a one-third rule was already circulating in the list from which the one-third rule was derived. That is suggestive and we will not call it more than that. It could be coincidence, an unrelated rule of thumb, or genuine circularity. Without a reachable source for that row there is no way to tell, which is itself the point.

What the figure will and will not carry

It is reasonable as an order-of-magnitude placeholder when you have nothing better and you label it as one, as an argument that turnover is expensive enough to be worth measuring properly, and as a prompt to calculate your own replacement cost from your own hiring data.

It will not survive being presented as a measured finding, being applied to senior or specialist roles when the inputs underneath it are hourly and frontline, being multiplied across a headcount to produce a precise-looking total, or being defended to a finance director who asks where it came from.

What to use instead

The honest answer is that replacement cost is specific to a role, a labour market and an employer, which is exactly why a single percentage travels so well and holds up so badly. If you need a number you can defend, build it from your own data: advertising and agency fees, the hours your team spends interviewing, onboarding and training, and the productivity gap until the replacement is fully effective. None of those components are mysterious and all of them are yours to measure.

If you need a published comparison instead, use a figure whose scope you can state out loud. Government wage data by occupation and region will tell you what the salary base actually is, which is the input the 33% rule quietly guesses at.

To be fair to Work Institute

They hid none of this. The estimate is described as an estimate, the arithmetic is printed in their own report, and the endnote is there to follow. Their interview research, which is the substance of the Retention Report, is a separate matter from this one modelling input.

The failure is collective. A figure published as an assumption became a fact through many thousands of repetitions, none of which followed the endnote.