A bad hire costs more than a salary.

Ask a leadership team to name the worst hire it ever made, and the answers arrive at once, with a name attached. Ask what that hire cost, and the room reaches for the only figure anyone booked: the recruiter's fee. That gap between vivid memory and empty ledger is the subject of this briefing, because every serious attempt to audit the cost of a bad hire lands on the same inversion. The items an organization records (agency fees, advertising, a signing bonus) are the small part of the bill; the items that dominate it (output never delivered, a team working around a weak colleague, a manager's quarter spent on remediation) appear in no account at all.

The measurable layers have, in fact, been measured, and they arrange themselves in a revealing order. The costs with invoices attached are the ones organizations quote, because they are the ones somebody had to approve. A wider accounting, which asks what the departure actually consumed rather than what it billed, lands several multiples higher. Beneath both sits a layer that surfaces only when an analyst asks what a stronger hire would have produced instead, and it is the layer least likely to be priced.

This article prices the layers in order and then turns to the variable that governs all of them. The probability that a given hire becomes a regretted one is a function of the validity of the method that admitted them, and it is computable before the offer letter is signed. A bad hire is not weather. It is a draw from a distribution, and organizations choose the distribution, mostly without knowing they are choosing.

The countable layer is real, and it is the small one

Start with what a controller can verify. Boushey and Glynn (2012), reviewing roughly 30 published case studies of turnover costs for the Center for American Progress, found that the median direct cost of replacing an employee was about 21% of the departing employee's annual salary. The figure was fairly stable for jobs paying under $75,000 a year and rose steeply for executive and highly specialized roles, where search firms, long vacancies, and scarce successors compound one another.

That 21% is the cost of employee turnover as the ledger sees it: separation processing, advertising and sourcing, interviewer hours, screening, the administrative side of onboarding. The management literature prices a wider frame. Allen, Bryant, and Vardaman (2010), writing in the Academy of Management Perspectives, put fully loaded turnover costs at 90% to 200% of the departing employee's annual salary once vacancy coverage, ramp-up to proficiency, and operational disruption are counted in.

The two literatures are not in conflict; they are counting different things. The direct studies tally invoices, which is why their totals are stable and auditable. The fully loaded estimates add the costs that arrive as absences: the sales calls not made while a desk sits empty, the veteran who spends her sprint reviewing a newcomer's work; absences resist invoicing. Figure 1 stacks the two accounts as one object, an iceberg with a bookkeeping waterline through it.

For a budget owner, the useful reading of Figure 1 is not the precision of either number but the ratio between them. The booked cost and the experienced cost of a departure differ by several multiples, which means every internal debate about recruiting spend is priced against a fraction of the real exposure. A team that argues over the 21% while absorbing the 90–200% in silence is optimizing the only number it can see. The band is wide because organizations differ in how long seats stay empty and how long proficiency takes, but even the bottom of the band dwarfs the invoice layer.

Two caveats belong in any executive summary of these numbers, and the sources themselves supply them. Turnover-cost studies mix voluntary and involuntary exits, and totals move with the counting method; Cascio and Boudreau (2011), whose book-length treatment sets out how to cost turnover and other staffing decisions, treat the inventory of what to count as the analysis itself, not a footnote to it. And a bad hire is not the same event as a turnover. The overlap is large, since regretted hires tend to leave or be asked to, but the categories diverge in exactly the place the next section prices: the bad hire who stays.

The most expensive bad hire is the one who stays

Replacement-cost studies share a quiet assumption: the loss happens when somebody leaves. But the defining feature of a mis-hire is performance, not departure, and the scale of performance differences inside a single job has been measured directly. Hunter, Schmidt, and Judiesch (1990), pooling studies of measured output in the Journal of Applied Psychology, found that the standard deviation of output, the typical gap between an average and a genuinely strong performer, runs about 19% of mean output in low-complexity jobs, about 32% in medium-complexity jobs, and about 48% in high-complexity and professional work.

Those three percentages are the raw material of every selection-value calculation, and the companion briefing on the ROI of selection charts them and runs them through the utility model. For a cost argument, one implication carries the weight. Because the spread widens with complexity, the price of an identical hiring error is set by the job it happens in, and it runs highest in the analytical, managerial, and specialist roles that organizations are most likely to fill under time pressure.

This sorts exposure by role rather than by seniority, which is not how most hiring governance is organized. Approval thresholds tend to track titles and salary bands, while output spread tracks the cognitive demands of the work. Two seats at the same pay grade can therefore carry materially different selection risk, and the one with the wider spread rarely gets the longer process. Complexity, in this literature, prices the mistake rather than causing it.

Now put the two sections together. A replacement fee is paid once — the output gap is paid every year of tenure. A mediocre hire in a high-complexity role who stays three years forfeits the performance difference annually while generating none of the turnover charges that would flag the situation to finance, which is why mis-hire cost estimates keyed to replacement spreadsheets systematically understate the exposure. The full conversion of this spread into currency is the province of utility analysis, treated in the companion briefing on the ROI of selection; the directional point suffices here. Much of the submerged mass in Figure 1 is unpriced for exactly this reason.

One person can reprice a team

The layers so far assume the damage stays with the individual. It does not. Felps, Mitchell, and Byington (2006), reviewing the experimental and field evidence on negative group members in Research in Organizational Behavior, found that a single persistently negative, disruptive, or free-riding member measurably depressed the performance of otherwise capable teams, and that on cooperative tasks groups tended to sink toward the level of their worst member rather than average across their talent.

The mechanism is structural rather than mysterious. Interdependent work gives every member a measure of veto power over shared rhythm: colleagues divert attention to monitoring, quietly redo work, stop volunteering, and protect themselves instead of the task. The bad hire's own performance file registers almost none of it, recording one modest contributor where the real entry is a team producing below its capability.

The descent-to-the-floor finding is worth unpacking, because it inverts a common intuition about hiring bars. If team output on interdependent work is set by the weakest member rather than the average, then the marginal hire does not merely add their own expected contribution; they reset the floor under everyone else's. A hiring standard, on this reading, is not a filter on individuals. It is a price support for the team.

Two further channels deserve a line even though neither carries a defensible number. Managers spend disproportionate time on their weakest reports, and that attention is diverted from the team's best opportunities; customer-facing roles convert internal quality problems into external ones, because the customer who meets the wrong employee does not file the experience under staffing. The honest accounting for this layer is qualitative, and this article will keep it that way rather than decorate it with an invented percentage.

The cost of a bad hire is a probability you can buy down

Everything to this point prices the event. The lever, though, is the probability, and the probability has been computable for decades. Taylor and Russell (1939) published, in the Journal of Applied Psychology, the tables that translate a validity coefficient, the correlation between assessment scores and later job performance, into the quantity a hiring manager actually experiences: the share of hires who work out.

Frame the question the way a risk officer would. Define a regretted hire, deliberately setting a demanding bar, as one who lands in the bottom half of performers for the role — a definition under which the no-signal rate is 50 per 100 by construction. Assume next that the employer is selective enough to hire the top 30% of its applicant pool, which describes any funnel attracting several plausible candidates per seat. Under the standard bivariate-normal model behind the Taylor-Russell tables, the mis-hire rate is then computable for any level of validity, before a single offer goes out.

The validity anchors come from Sackett, Zhang, Berry, and Lievens (2022), whose re-estimate of selection-method validity applied less aggressive range-restriction corrections (adjustments for validation samples containing only the people who were hired) than earlier syntheses: unstructured interviews carry a validity of .19, structured interviews .42. The full ranking of methods is the subject of the series' flagship briefing on what predicts job performance; here the coefficients do a different job. They set the price of risk.

The computed results are in Figure 2. With no signal at all, the outcome is a coin flip: 50 of every 100 hires land below the median. At r = .19, the unstructured interview, the count falls to 41. At r = .42, a structured process, it falls to 30; at r = .50, within reach of a strong multi-method composite on the 2022 corrections, 26.

Read the bars as a procurement document. Nothing in the figure eliminates risk; even a strong composite leaves 26 regretted hires per 100, and any vendor implying otherwise is promising something validity cannot deliver. What the figure establishes is narrower and more useful: mis-hire risk is a function of measurement quality, the function is known, and it is the only term in the entire cost structure that can be moved before the exposure begins.

That caption's "computed" is a warning label. Read the bars the way an actuary reads a model rather than the way an auditor reads a receipt: change the assumptions and they move. a role where most candidates could succeed has a friendlier base rate, a funnel with few applicants cannot select from the top 30%, and real performance distributions are messier than a bivariate normal. What does not move is the ordering, or the direction of the slope. Under any defensible parameterization, more validity means fewer regretted hires, and the improvement is largest across the very range where most organizations currently operate, between an unstructured conversation and a structured, multi-method read of the candidate.

The figure also reframes what an assessment is for. The instrument is not a formality at the mouth of the funnel; it is the parameter that sets how often the organization pays the iceberg in Figure 1 and the recurring output gap beneath it. The cost of a bad hire, multiplied by the probability in Figure 2, is the expected loss per offer, and expected loss per offer is a number a CFO can govern.

The bill arrives as a curve, not an invoice

One property of the mis-hire remains, and it is the one budget cycles are worst at seeing: timing. A bad hire is not a single charge but a curve that steepens, sketched in Figure 3. The account opens before day one, with a vacancy someone else is covering; it climbs through a ramp in which a full salary buys partial output; it grows through the long middle stretch in which the performance gap is visible but unaddressed; and it spikes at the end, when documentation, difficult conversations, and severance turn a performance problem into a project.

Then the curve does the thing that makes it a systems problem rather than an HR anecdote: it ends where the next curve begins. The managed exit reopens the vacancy, restarts the recruiting spend, and returns the organization to the top of the funnel, where, if the selection instrument is unchanged, the odds in Figure 2 are unchanged too. Organizations that treat each mis-hire as bad luck repeat the draw; organizations that treat it as a sample from their own selection process get to change the process.

The dashed contrast line is the point of the figure. A sound hire passes through the same vacancy and much the same ramp, then diverges permanently: value returned, compounding with tenure, while the mis-hire's curve keeps climbing. The distance between the two trajectories at any month is the running total of a decision made before either curve started.

Of the five phases, the drag segment invites particular suspicion, because it is the longest and the one most governed by psychology rather than process. Sunk-cost reasoning extends it: the organization has paid for the search, the ramp, and the training, and each month of patience is framed as protecting that investment rather than as buying another month of the gap. Hope extends it too, since performance problems rarely announce themselves as permanent. The curve in Figure 3 has no natural alarm on that stretch, which is precisely why it accumulates so much of the area under the line.

The curve also explains why this exposure is chronically underweighted in budget debates. Its segments land in different fiscal periods and different cost centers (recruiting, the hiring team's payroll, the manager's calendar, severance), so no single owner ever sees the total. A figure that is never summed is never compared to the cost of preventing it.

Price selection risk like any other operational risk

The prescription follows from the structure of the problem: expected loss equals probability times exposure, and the two factors are managed in different offices. Exposure is set by role design, salary, and complexity; probability is set by the selection method. The same arithmetic already governs credit risk, safety incidents, and supply failures elsewhere in the business, and none of those registers would accept "we interviewed them and liked them" as a control. An organization that wants to manage the cost of a bad hire, rather than merely absorb it, needs four disciplines.

1. Price the exposure per role, not per policy. The published bands bracket it: about 21% of salary in direct replacement cost (Boushey & Glynn, 2012) and 90–200% fully loaded (Allen et al., 2010), with the forgone-performance layer scaling on top as complexity rises (Hunter, Schmidt, & Judiesch, 1990). A mis-hire in a high-complexity role is a different financial event from one in a routine role, and the steep rise Boushey and Glynn document for executive and specialized positions means the tail needs its own entry.
2. Treat validity as the control variable. It is the one input an organization can move, and it moves in documented steps: structure, standardization, and multi-method measurement separate .19 from .42 between unstructured and structured interviews alone. Method quality belongs in the procurement file next to price, with validity claims documented rather than asserted (what documented validity evidence looks like).
3. Instrument the funnel backward. Early exits and weak first-year reviews are outcome data about the selection method that admitted them, and the signals that precede an early exit are themselves measurable before the offer. Route them back, so that every regretted hire updates the organization's estimate of its own mis-hire rate, per method and per role, using the costing discipline Cascio and Boudreau (2011) supply for exactly this bookkeeping.
4. Book the submerged layers somewhere. What appears in no ledger gets managed by no one. Even a published internal range for vacancy drag, ramp cost, and team spillover changes who attends the meeting where the assessment budget is set. The range does not have to be precise to be useful; it has to exist, and it has to be applied consistently enough that two requisitions can be compared. Precision can improve later, once the organization has its own exit and performance data to calibrate against.

Volume sharpens all of this. A hundred offers a year at a 41-per-100 regret rate and at a 30-per-100 regret rate describe different companies three years later, and the gap widens with every additional offer, making enterprise hiring programs the natural first place to manage selection risk formally. The cost of employee turnover is a lagging indicator of a selection process; the mis-hire rate is the leading one.

The worst hire anyone in the room remembers was never an isolated accident. It was a draw from a distribution the organization configured: priced in Figure 1, governed by Figure 2, billed on the schedule in Figure 3. The distribution, unlike the memory, is adjustable.