Cut Calendar Time, Not Evidence, When the Start Date Is Fixed

Where is the time actually going?

Into waiting, mostly. In SHRM's benchmarking data the median time to fill a nonexecutive role was 44 calendar days from the requisition opening to the offer being accepted, and within that the median organization spent 5 days screening applicants, 7 days interviewing, 4 days making a final decision and extending an offer, and 2 days from offer to acceptance ¹. Those are medians of separate quantities and will not add up, but the shape is clear.

That 44 hides more than it shows. It is a median, the mean in the same data is 54 days, and the middle half of organizations sat between 28 and 73 days ¹. It counts calendar days including internal approval delay, it stops at offer acceptance rather than at a start date, and it is 2021 data collected from SHRM member organizations. Any sentence beginning with the average hire takes 44 days is wrong twice over.

Published benchmarks disagree because their clocks differ and their samples differ, and no published figure separates the two causes. One large recruiting dataset reports a median Time to First Fill, from a job being opened to the first hire being made, of 56 days for business roles and 76 for technical ones ⁴, measured on venture-backed technology employers hiring knowledge roles. Time to First Fill stops at the first hire on a requisition, so it is a different quantity from SHRM's, and neither of them is your calendar.

So start from your own calendar instead. Lay the process out backwards from the start date, mark every interval, and label each one waiting or working. Pre-booking a block of interview slots before the posting goes live, answering submissions within a day, and giving one person authority to schedule without asking usually recovers more calendar than deleting a stage does, and costs nothing in evidence.

Which stage produces the most evidence per hour?

The one where the candidate produces something. The stages above it mostly re-read what the candidate already sent. Rank by evidence per hour of your attention and the order inverts the usual one: in one large recruiting dataset the later stages convert at roughly 95% after the onsite and 81% through offer, while recruiter screens pass about 35% of those who reach them ². A stage that passes nearly everyone who reaches it is ratifying a decision made earlier.

Each of those numbers is narrower than it looks. They are stage-to-stage passthrough among candidates who reached each stage, so they do not multiply into an end-to-end rate, the heaviest filtering happens at application review which is not in the numbers at all, and the 81% is passthrough through the offer stage, which is not the offer acceptance rate it gets mistaken for. Stage names are configured per employer, and the dataset skews toward venture-backed technology companies hiring knowledge roles. What survives all of that is the ordering.

The top of a funnel is structurally weak before anybody runs it badly. Modelling a screening stage built only from methods that scale to volume, with no structured interview available, the validity-maximising combination tops out around .51 against .61 when the interview is in the mix ³. That ceiling is a modelling result under stated assumptions, nobody has observed it at an employer, and resume screening is not in the model at all. The comparison is the usable part: the methods that scale to the top of a funnel are collectively weaker than the one that does not scale.

So the cut order, roughly, for a small team against a date: the cover letter, the recruiter-style introductory call that repeats the application, the fourth conversation added because somebody wanted to meet the candidate, and the culture chat with no questions written down. Dropping the resume screen entirely in favour of a work sample is a real option, and it changes what the top of your funnel has to survive: see going straight to a work sample.

Keep the stage where the candidate makes something

Keep it, and shrink it to fit the calendar you have. A forty-five minute working session on a real decision from your own business beats a week-long take-home nobody has time to read, and the margin widens when a deadline is why you are choosing between them. That stage is the only one in the loop producing evidence that did not exist before the candidate arrived.

The asynchronous version of it is rarer than the debate suggests. In the same recruiting dataset, around 13% of hires included a take-home component ², which makes it a minority practice, and the real question under time pressure is usually whether to add anything here at all. Add the short version. Forty-five minutes of live work with your material in front of both of you gives you more to act on than three unstructured conversations, and it fits inside a week.

Four compressions that keep the evidence:

• Run it live. The submission is rougher and the reasoning is visible, which is the part you wanted.
• Use one real decision from last quarter. No authoring time, and the answer key is your own experience.
• Give the whole loop to two people. Scheduling cost grows with the number of calendars, not with the number of hours.
• Book the debrief before the interviews. The slot itself is usually the delay.

Testing how somebody works with an assistant belongs inside the exercise you already run rather than in an hour added beside it, which is the whole of testing AI skills without lengthening the loop. A week spent shortening the loop is the worst week to lengthen it. And if your loop got longer this year, the cause is usually a stage someone added to handle volume, which is worth diagnosing before you cut anything: where the time went after AI entered every stage.

Set the bar before the calendar sets it for you

Set the bar in writing before the first interview, and write down what would make you say no. A deadline does not lower standards by itself; it lowers them by letting the threshold drift as the date approaches, in a process where nobody wrote the threshold down and so nobody can watch it move. One paragraph in a document is most of the defense against that.

Write the no conditions as specifically as the yes ones. Somebody who cannot describe a decision they got wrong. Somebody whose work sample is clean and whose reasoning for it is empty. Somebody who needs the role to be a different role. Those are easier to hold onto at day 25 than a general sense of standards, because they were written when you were not yet under pressure.

Then decide the fallback in advance, because that decision is much worse made late. If nobody clears the bar by the date, the honest options are contract coverage for the gap, a reduced scope for the first hire, or a later start with the work triaged in the meantime. Each of those costs something you can name. Hiring somebody you have already decided is below the bar costs a search you will run again before long, plus the cost of ending the first one. Contract coverage here means a genuine contractor engagement with a defined deliverable, and whether an arrangement is really that is a classification question with consequences: contractor or employee for a first hire. Check the shape of it with counsel before the paperwork goes out.

Protect one thing more while you are moving fast. Ask every candidate the same questions in the same order, and keep whatever notes you take, however brief. Consistency costs no calendar time, it is the cheapest thing to lose under pressure, and it is the only record you will have if the decision is questioned later.

See a sample report >