In January, a talent acquisition director is handed a single objective for the year: bring average time-to-hire down from 88 days to 60. The objective is sensibly motivated. Every day a senior engineering seat sits empty costs real money, and the strongest candidates leave the market within a fortnight. By December the objective is met at 58 days. The team receives its bonus and a slide with a falling line goes to the board.
Over the same twelve months, the proportion of hires who leave within their first year rises from 30 per cent to 41. Nobody places those two figures side by side, because two different functions own them, they report on different cycles, and the second one closes long after the first has been filed away. Hiring was made 31 days faster and 11 percentage points worse, and the organisation celebrated exactly one half of that sentence.
This article defends a claim that is uncomfortable for anyone carrying that metric in their objectives: optimising for time-to-hire systematically degrades hiring quality, because it rewards shortening precisely the stages that filter out bad hires. The problem is not that the metric can be implemented badly. The problem is that implementing it correctly produces this outcome, because once queueing time is exhausted the only remaining days sit inside the filtering stages.
What does time-to-hire actually measure, and what does it leave out?
Time-to-hire counts the days from a candidate entering the process to that candidate accepting an offer. Many UK organisations track time-to-fill instead, running from requisition approval to signature. The distinction matters operationally and not at all for this argument, because both share the same structural defect: they stop counting on the day of the signature.
Everything that determines whether the hire was a good one happens after that date. Whether the person delivered what the role required. Whether they integrated with the team. Whether they are still there a year later. Whether the hiring manager, given the same decision again, would make it the same way. A metric that closes before any signal about consequences exists is not a measure of process quality. It is a measure of process throughput.
What makes it attractive is exactly what makes it dangerous. Time-to-hire is available immediately, computed automatically by every applicant tracking system, comparable across companies, and intuitively legible to a board that does not need to understand hiring to understand that 60 is less than 88. Quality-of-hire metrics have none of those properties. They require waiting, qualitative judgement and an agreed definition.
So most organisations end up choosing between a metric that is precise, fast and nearly unrelated to the goal, and one that is slow, fuzzy and measures the thing they actually care about. They pick the first, then justify it on the grounds that the second cannot be measured well. The premise is true and the conclusion does not follow. That we cannot measure something precisely is not an argument for managing something else instead.
Why does shortening a process almost always mean removing a filter?
This is the core of the argument and it deserves to be set out carefully, because it marks the difference between criticising an implementation and criticising the metric itself.
A senior hiring process decomposes into two categories of elapsed time. The first is queueing: the gap between application and first contact, diary collisions, the wait for a salary band decision, the week spent getting headcount signed off. The second is filtering: the technical interview, the practical exercise, the team conversation, reference checks, and a background check where the role warrants one.
Cutting the first category is pure profit and every organisation should do it aggressively. The difficulty is that there is a finite supply of it and it runs out quickly. A company that tidies its diaries, shortens its approval chain and stops leaving candidates unanswered for ten days typically recovers fifteen to twenty-five days. Once. By the second year of the same objective that reserve no longer exists.
The objective does not expire with the reserve. If the target requires moving from 70 days to 60 and all the queueing has already gone, the only remaining days sit in the filtering category. What follows is a series of decisions that each sound entirely reasonable in isolation: two technical interviews become one, a six-hour take-home becomes a two-hour pairing session or disappears entirely, the team conversation is merged into the hiring manager interview, references are taken after the offer if at all.
Each of these has its own justification and none of them is stupid. Together they remove the process’s function. Filtering stages are not the cost of the process, they are its product — they are what generates the information the decision rests on. Cutting them does not buy time, it buys the same decision made on less evidence.
What does a vacant day actually cost, and what does a bad hire cost?
This claim would be rhetoric if it could not be priced. It can be, and the numbers are worth being explicit about because the argument turns on their ratio rather than on either figure alone.
On the vacancy side, a senior engineer in a UK product business is typically carrying £15,000 to £35,000 a month of attributable output, and that range is the honest one. On the failure side, the Recruitment and Employment Confederation’s much-quoted estimate puts the cost of a bad hire at mid-manager level at up to three times salary. On a £90,000 senior engineering salary that spans roughly £135,000 to £270,000 once the agency fee, the lost ramp-up, the second search and the team disruption are included — consistent in structure with the breakdown in our own IT recruitment cost calculator, which reaches comparable magnitudes on Polish figures.
Put the numbers in. Cutting a process from 88 days to 58 saves 30 days of vacancy. At £25,000 a month that is £25,000 saved per hire. The saving is real and I am not going to argue it away.
Now the other side of the ledger. If that shortened process raised first-year attrition by ten percentage points, then every hire carries an additional ten per cent chance of incurring the full cost of a failed hire. At £180,000 that is £18,000 of additional expected cost per hire. At £270,000 it is £27,000.
So the trade lands close to even, and that is its most interesting property. For cheap, easily replaced roles, speed wins. For critical ones it loses, and it loses before we add the costs that resist being priced at all: the team’s loss of confidence in the hiring process after a second consecutive departure, and the cascade in which a badly filled leadership seat pulls two or three further resignations behind it within six months.
The decisive point, though, is not which side wins. It is that almost nobody performs this calculation, because the two figures belong to different people. The vacancy saving is immediate, visible and attributed to talent acquisition. The cost of the bad decision is deferred, diffuse and absorbed by the team that suffered it. The accountability structure guarantees that the first number gets optimised and the second gets tolerated.
Where is the break-even point, and why is it not where you expect?
Since the trade can be written down, the threshold can be solved for. It is one of the few things in hiring you can work out on the back of an envelope in five minutes, and almost nobody does.
Shortening pays as long as the vacancy saving exceeds the increase in expected failure cost. Stated plainly: days saved times daily vacancy cost must exceed the increase in failure rate times the cost of one failed hire.
For a senior engineer at £25,000 a month of vacancy cost and £180,000 of failure cost, saving 30 days yields £25,000, and that saving is wiped out by a 13.9 percentage point rise in the failure rate. At a leaner £15,000 a month the threshold falls to 8.3 points. For an engineering manager at £360,000 of failure cost and £30,000 of monthly vacancy cost, it is again about 8.3 points.
It is worth pausing on these, because they do not say what people expect them to say. The threshold is not microscopic. Eight to fourteen percentage points is a substantial move — first-year attrition going from 30 per cent to 38 or 44. Anyone hoping the arithmetic would immediately rule out shortening gets a more awkward answer: on a single hire, this is a bet at reasonable odds.
The question is what the bet is about. The scenario that opened this article, a rise from 30 per cent to 41, sits squarely inside that band. Which is to say: an effect the same size as the threshold is a completely ordinary consequence of dropping a practical exercise or a second technical interview. It is not a worst case, it is the expected case. So an operation the organisation performs every quarter has its outcome decided by a quantity nobody measures, whose plausible magnitude straddles the break-even point.
That is the real indictment and it is sharper than the simpler claim that shortening never pays. Shortening the process is a transaction of unknown sign, entered into in the belief that the sign is known. The board slide shows one side of the ledger — the calculated, immediate, favourable side — and the other side is neither estimated, nor given a range, nor listed as a line item at all.
Why do our own English-language articles promise both at once?
Honesty requires starting the argument at home, because our own collection carries exactly the promise this article disputes.
Our IT talent acquisition checklist states that its structured process “reduces time-to-hire by 30-40% while improving candidate quality and hiring consistency”. That is the conjunction under dispute, asserted rather than demonstrated. If both effects genuinely occur together, it is because the quality of the funnel’s input changed, not because the number of gates fell — and the checklist does not distinguish between the two.
The same article is more useful than it looks, though, because it supplies the number that decides the whole argument, and supplying it exposes how much weight it carries. It puts the cost of an unfilled critical IT role at $10,000 to $25,000 per week. Take the top of that range: roughly £84,000 a month, at which point saving 30 days is worth £84,000 and the break-even threshold jumps to about 47 percentage points. At the bottom of the range it is still nearly 19 points. On the checklist’s own vacancy figures, speed wins almost unconditionally.
That is not a refutation of this article. It is the demonstration of its central claim. The entire question turns on the cost-of-vacancy figure, which is quoted without a source and which the reader’s organisation has almost certainly never measured. The checklist’s advice is sound if the number is right for you and expensive if it is not, and nothing in the advice tells you which case you are in. Our opportunity cost analysis is the piece that actually tries to derive that figure rather than assert it, and it is the one worth reading first.
The narrower criticism applies to our candidate experience guide, which recommends celebrating with the team when time-to-fill falls. A falling time-to-fill is not a success in itself. It is a symptom, consistent with tidier diaries and equally consistent with a deleted take-home exercise, and the dashboard cannot tell those apart. Celebrating the fall without establishing its mechanism teaches the team that the number is what gets rewarded.
Do referral programmes disprove the thesis?
This is the strongest counter-argument I know, it comes from our own collection, and it cannot be waved away.
Our article on employee referral programmes reports that referrals deliver 40 per cent better retention and 55 per cent faster time-to-hire. If both figures hold, then a lever exists that compresses the process and simultaneously improves hiring quality. That looks like a direct contradiction.
It is not. It is the thesis’s best confirmation, provided you look at the mechanism.
A referral compresses the process by raising the quality of the population entering the funnel. Someone referred by an engineer already on the team has been pre-screened by a person who knows both the candidate and the reality of the job — a filter of a quality no formal stage can match. That candidate sits the same set of interviews as everybody else and simply fails them less often. The process gets shorter because fewer candidates are rejected along the way, not because any gate was removed.
Now compare that with the other lever producing the same dashboard movement: deleting the practical exercise. That also cuts time-to-hire, but by reducing the number of gates while the input quality stays exactly where it was. The first operation raises hiring quality, the second lowers it, and in the quarterly pack both appear identically as a fall in the mean number of days.
That is precisely the argument against the metric rather than against speed. Time-to-hire cannot distinguish an improvement in input quality from a degradation of selection, so it rewards both equally. Since the second is far easier to implement — one decision in one meeting, rather than two years of building a referral culture — the second is what wins wherever the number enters annual objectives.
How does the metric distort recruiters, hiring managers and candidates?
A target metric changes the behaviour of everyone who can see it, and the changes are worth separating out, because each is rational for the person making it.
The recruiter stops presenting risky candidates. A career changer, someone with an unusual trajectory, anyone who needs an extra conversation, lengthens the process and damages the number. A safe candidate matching the job description word for word closes quickly. Across a year, selection drifts towards literal profiles, which independently deepens the problem we described in our piece on the junior developer crisis: anyone who needs developing is by definition slower to hire.
The hiring manager stops admitting doubt. Saying “I would like to see two more candidates” costs measurable days and shows up in the statistics as delay on their side. Saying “let us proceed with this one” costs nothing and reads as decisiveness. In an organisation that measures days, hesitation becomes expensive and agreement becomes free — an exact inversion of how a large, hard-to-reverse decision ought to be weighted.
The candidate gets an offer faster and joins an environment they had less chance to assess. The stages the employer treats as filters are the candidate’s only window onto the team, the codebase and the way work actually happens. A compressed process reduces their ability to withdraw, which raises first-year attrition regardless of how accurate the employer’s judgement was. Selection runs both ways, and the shortcut cuts both. This is the mechanism sitting underneath much of what we described in our work on IT talent retention: people leave situations they never got to evaluate.
None of these behaviours is misconduct. All of them are correct responses to what the organisation measures and rewards. Which is why appeals to good judgement do not fix it — nobody here is exercising bad judgement. There is an objective everyone meets correctly and a side effect that appears in nobody’s objectives.
Why does a metric with these flaws remain so popular?
The answer is not that talent teams fail to understand it. They understand it perfectly well and report it anyway, and the reasons are structural.
The first is asymmetric availability. Time-to-hire is in the applicant tracking system on day one and requires no agreement from anyone. First-year attrition requires joining recruitment data to HR records, attributing departures to candidate sources, and settling on a definition of which departures count as failures. A metric that needs no implementation project beats one that does, irrespective of which measures the right thing.
The second is demand for comparability. Boards want to know whether hiring performs better than at other companies. Time-to-hire has market benchmarks; hiring quality has none and cannot have any, because it depends on organisational context. The appetite for external comparison by itself favours shallow metrics.
The third is the accountability boundary. Talent acquisition is answerable for delivering a candidate, not for that candidate’s effectiveness a year on. A metric ending at signature fits the function’s remit exactly — but the remit is arbitrary and it stops precisely where the thing the business cares about begins.
The fourth is the least comfortable: time-to-hire is almost never caught out. Because failed hires surface with a delay and admit many possible explanations, they can nearly always be attributed to something other than the shortened process — cultural fit, the manager, the market, compensation. The metric is never convicted, because the consequences of optimising it are always over-determined.
Why does the UK correction window make this worse rather than better?
There is a specifically British version of this conversation and it deserves separating out, because the UK’s legal framework has quietly subsidised fast hiring for decades.
Under section 108 of the Employment Rights Act 1996, an employee generally requires two years of continuous service before they can bring an ordinary unfair dismissal claim. Compared with most European jurisdictions this is an unusually long and unusually cheap correction window, and it has shaped UK hiring practice more than most practitioners notice. If a mis-hire can be unwound at relatively low legal risk for up to two years, then in strict expected-cost terms a shorter process really is more defensible in London than in Warsaw, Berlin or Paris. That is a genuine argument and it should be stated at full strength before it is answered.
The answer has two parts. The first is that legal exposure is not the same as economic cost, and the arithmetic above was never about tribunal risk. By the time the correction window is exercised, the business has already paid a recruitment agency fee of 15 to 25 per cent of first-year salary, funded three to six months of reduced output, absorbed the management time consumed by a performance process, and taken the morale hit of a departure the team watched coming. Every one of those costs sits inside the failure figure the break-even calculation uses, and none of them is affected in any way by section 108.
The second is that the subsidy is being withdrawn. The direction of UK employment reform is towards day-one unfair dismissal protection, with a statutory initial period of employment doing the work the two-year qualifying period does today. Whatever the final commencement timetable, the movement is one way, and it makes the correction window narrower rather than wider. Any hiring model whose economics quietly depend on the cheapness of reversing a mistake becomes more expensive as that assumption weakens — and organisations that shortened their processes on that implicit basis will discover the dependency only after it has stopped holding.
There is a parallel worth noting for contract engagements. Since April 2021 the off-payroll working rules in Chapter 10, Part 2 of ITEPA 2003 have required medium and large private-sector clients to determine a contractor’s employment status and issue a Status Determination Statement. This is a compliance step with a real elapsed-time cost, and it is the one part of a contractor onboarding process that must not be compressed for the sake of a metric — because the consequence of getting it wrong is not a bad hire but a tax liability.
What should replace time-to-hire without recreating the same trap?
Swapping one target metric for another simply reproduces the problem under a new name. Quality of hire introduced as a quarterly objective will be optimised by managers inflating their ratings, exactly as time-to-hire is optimised by deleting stages. How the set is constructed matters more than which single indicator is chosen.
Three rules make it work.
Rule one: the only target metric is one that closes after the consequence. In practice this means first-year attrition, split by candidate source and by role. It is slow, awkward and impossible to improve within a quarter — and those are its virtues, not its defects. A metric that cannot be improved quickly does not invite shortcuts.
Rule two: speed metrics stay, without numeric targets. Time-to-hire is reported as an observed value with its spread, and split into queueing time and filtering time. That split is the important part, because it answers the question the headline average forbids: where did the reduction come from. A fall in the queueing component is a success. A fall in the filtering component is a decision that requires justification.
Rule three: add a metric owned by the receiving side. Time-to-productivity — weeks from day one to independently closing work of typical complexity for the role, assessed by the hiring manager. It is fuzzy and assessor-dependent, but it measures exactly what the hire was opened for, and it is immune to optimisation by the recruiter because it lies outside their control.
To that set, add one purely diagnostic figure: the proportion of offers accepted at the first financial proposal. It detects salary-band drift faster and more cheaply than any compensation benchmark, and it is close to impossible to game in a harmful direction.
One implementation detail decides whether any of this survives contact with an organisation, and it is usually treated as a formality: you must write down who owns the metric that closes after a year. If first-year attrition remains the property of HR while days remain the property of talent acquisition, the new set reproduces the original failure under new names, because once again nobody holds both numbers. The after-the-fact metric has to sit with the person accountable for filling roles, even though their influence over it is only partial. Partial influence over the right thing beats total influence over the wrong one.
What does a maturity map for hiring measurement look like?
The table below sets out the four levels organisations actually occupy. The levels describe what the measurement system rewards, not what it aspires to, because rewarded behaviour rather than stated intent determines the real level.
| Level | Target metric | What the system actually rewards | Typical symptom | First move out |
|---|---|---|---|---|
| 0. Unmeasured | none | Filling the seat at all | Nobody knows how long hiring takes or how many people leave in year one | Start measuring first-year attrition before you start measuring days |
| 1. Speed measured | time-to-hire or time-to-fill as a target | Cutting filtering stages once the queueing reserve runs out | Falling average days alongside rising first-year attrition, reported separately | Split time-to-hire into queueing time and filtering time |
| 2. Split measurement | time-to-hire observed by stage, no target | Removing queueing only, because only that is rewarded | The average stops falling after a year and this is correctly treated as success | Add first-year attrition split by candidate source |
| 3. Consequence measured | first-year attrition plus time-to-productivity | Accurate decisions, including slower ones | A manager can say “I need one more conversation” at no cost | Tie candidate sources to twelve-month outcomes and cut sources, not stages |
Moving from level one to level two is cheap and achievable in a quarter, because it only changes how existing data is presented. Moving from two to three requires joining recruitment and HR data and agreeing a single definition of a problem departure. That is a two-quarter project and the most common reason organisations stall at level one.
When does this thesis not apply?
An article that does not mark the boundaries of its own claim is a sales document rather than an analysis. The boundaries here are clear.
It does not apply when the cost of error is low and reversible. In volume hiring for roles with repeatable, well-documented scope, a wrong decision is cheap and replacement is quick. Speed is the correct objective there, and elaborate selection wastes money on both sides of the table.
It does not apply when the vacancy cost is immediate and counted in days. If one missing person halts a delivery with a contractual date or blocks revenue from tomorrow, the arithmetic tilts decisively towards speed, and shortening the process becomes a deliberate, calculated choice. That is a completely different thing from an annual objective to reduce an average.
It does not apply to pathologically long processes. A 150-day process with seven stages does not suffer from excessive filtering, it suffers from organisational disarray, and shortening it improves hiring quality by stopping the outflow of good candidates. This argument concerns processes already within market norms.
It does not apply to queueing time at all. There was never a case for leaving a candidate unanswered for ten days, and nothing here supplies one. That portion of the process should be cut without hesitation and driven towards zero.
How does staff augmentation change the arithmetic?
It is worth saying plainly where our commercial interest sits in this analysis, because without that the rest reads as a run-up to a pitch.
Staff augmentation changes one variable in this calculation, and it is not the variable usually advertised. Obviously it shortens time to a filled seat — our own benchmark is around two weeks against the 52 to 88 days a senior permanent search typically takes in this market. That is not what matters for the thesis.
What matters is that it changes the cost of error and its reversibility. The whole break-even calculation rests on a failed permanent hire costing £135,000 to £360,000 and being hard to unwind. On a contract engagement the same error costs a fraction of that and is reversible in weeks. When the cost of error falls by an order of magnitude, the break-even threshold moves by an order of magnitude too, and speed genuinely does become the right objective.
Which yields a more honest conclusion than the usual recommendation: our model does not solve the problem described in this article, it changes the conditions under which the problem occurs. If you are filling a strategic role intended to stay in the organisation for years, the failure-cost arithmetic applies in full and shortening the process is a bad idea regardless of engagement model. If instead you need a capability for a defined horizon, or you want to validate a headcount need before opening a permanent role, a contract engagement lets you buy speed without paying for it in decision quality — because the decision is reversible.
The sequence we see working most often is exactly that: contract first, to start without a vacancy and learn the real shape of the need, then an unhurried permanent search for the role you now know is strategic. The vacancy stops pressing on the selection process, which is the only durable way to take the pressure off time-to-hire.
What are the key takeaways?
- Time-to-hire stops counting at signature, before any signal about decision accuracy exists. It measures throughput, not quality, and has no justification as an annual target.
- Once the queueing reserve is exhausted, the only remaining days sit in filtering stages. Shortening then stops meaning a tidier organisation and starts meaning a decision made on less information.
- The break-even threshold is calculable and sits between roughly 8 and 14 percentage points of additional failure rate on UK numbers, depending on role and vacancy cost. That is not microscopic — it is the same size as the typical effect of removing a stage, which makes the trade one of unknown sign.
- The whole answer hinges on your cost-of-vacancy figure, which most organisations quote from an article rather than derive from their own data. Derive it before you act on any advice about hiring speed, including ours.
- Referral programmes confirm the thesis rather than refuting it. They compress the process by improving input quality, not by removing gates — and that distinction is invisible to a metric that rewards both identically.
- The UK’s two-year unfair dismissal window subsidised fast hiring and is being withdrawn. Models whose economics depend on cheap reversal get more expensive as day-one protection arrives.
- The set that works has one target metric that closes after the consequence — first-year attrition by source — with time-to-hire reported without a target, split into queueing and filtering.
If your average time-to-hire has fallen three years running, there is one question worth asking before that chart goes to the board again: which part of the process did those days come out of. If the answer is “the stages where candidates used to fail”, you have been making a hiring-quality decision every quarter without knowing it.
If you want to put your own numbers against this — your cost of vacancy, your cost of a failed hire and your actual first-year attrition rate — and find out which side of the threshold your process sits on, let us talk. We will work the calculation through on your data and help design a staffing model that buys speed where mistakes are reversible and preserves decision time where they are not.