High-volume hiring without a screening bottleneck

• In high-volume hiring the bottleneck is the first human read, and it is an arithmetic problem before it is a hiring problem. Two thousand applications at four minutes each is over a hundred and thirty hours of reading. One recruiter working on nothing else finishes in about three and a half weeks, by which point the strongest candidates have taken other work.

• The way out is to decide what evidence separates candidates before the campaign opens, collect that evidence in a comparable form at application time, rank the pool on it, and spend the reading time on a ranked band instead of on arrival order.

Decide the evidence before you advertise

• Pick three to five things that genuinely predict whether someone can do this job. For frontline roles that is usually availability against your actual shift pattern, a legal right to work in the location, one piece of relevant practical experience, and a short scored answer to a scenario from the job.

• Then collect them as structured questions in the application, not as free text to be mined from a CV later. A yes or no field about weekend availability is comparable across two thousand people in a way that a paragraph about flexibility never is. Every criterion you leave to the CV is a criterion your team will read for, one applicant at a time.

• The scenario question is worth the extra minute you are asking candidates for. OPM's structured interview guidance covers how to build one: describe a realistic situation from the job, ask what the person would do, and score the answers against a scale with written levels rather than a general impression. Done at application stage it gives you a work-relevant signal on everyone, including people whose CVs are thin because they are early in their careers.

Rank the pool, then read in bands

• A ranked pool changes what the reading time is for. Instead of a queue, you have an order, and the job becomes checking whether the order is right.

• The practice that holds this together is that every applicant stays in the pool and stays visible. A band gets reviewed, the band under it gets spot-checked, and anyone can be pulled up by a recruiter who sees something the ranking missed. What does not happen is software declining people on its own. A ranking exists to put a person's attention in the right place; the decision to decline a candidate is made by a named human against stated criteria, and it is recorded.

• That is a design choice with a cost, and it is cheaper than the alternative. A filter that quietly discards the bottom of a pool discards it permanently, and nobody ever audits a folder they cannot see. We went through what those invisible rejections cost a team in [the hidden cost of false rejections](/spaces/cost-of-ats-false-rejections).

Make a screen mean one thing

• At volume, several people run screening calls at once, which is where consistency dies. OPM's three conditions apply whatever the length of the conversation: the same questions in the same order, one rating scale for everyone, and agreement on what an acceptable answer sounds like before the first call.

• A ten minute screen with four fixed questions and a five point scale gives you something you can compare across screeners. Ten minutes of rapport does not.

• Two of OPM's named interviewing mistakes get worse as volume rises. Pressure to hire is what happens when a team facing an unfilled rota starts accepting thinner evidence as the deadline approaches. Contrast effects are what happens when a candidate gets rated against whoever was screened immediately before them instead of against the scale. UK government guidance recommends interviewing in batches, which helps with scheduling and makes the second problem more visible, so the scale has to be written down and used.

Watch the funnel while it is running

• A volume screen applies the same rule to thousands of people, so an error in it compounds at a speed that a single-hire process never reaches. Count how many candidates enter and pass each stage, and where you can lawfully collect the data, break the pass rates down by group.

• The EEOC's guidance sets out the standard in the US. A neutral selection procedure that disproportionately excludes people on a protected ground has to be job related and consistent with business necessity, and where an alternative would predict job performance as well with less exclusion, the employer is expected to adopt it. Employers stay responsible for the procedures they run even when a vendor supplied them and says they are validated. In practice that means asking your vendor what the screen measures and checking the answer against your own pass rates, rather than accepting a documentation pack.

Seasonal campaigns start with last season

• A seasonal campaign that begins from an empty pipeline every year is paying twice. Before the advert goes out, pull the people who worked the last peak and left in good standing, then the candidates who reached the final stage and declined or lost out. Both groups already have evidence attached and can go to the front of the queue.

• Acas advises advertising in more than one place, because a single channel reaches only the people on it. For seasonal work that usually means the local channels your last cohort actually came from, which the previous campaign's data can tell you.

• Put the working pattern and the pay in the advert. Peak-season candidates are comparing several offers in the same week, and an advert that hides the shift pattern buys itself screening calls that end at the first question. That is the same discipline described in [how to write a job description that attracts the right people](/spaces/write-job-description-that-attracts), under more time pressure.

• Everything here depends on the speed of the first cut. When a ranked shortlist comes back the same day the applications land, the reading time goes to the candidates sitting near the line. That is the argument for [building shortlists in hours](/spaces/build-shortlist-in-hours), and for being precise about [what shortlisting is supposed to produce](/spaces/what-is-candidate-shortlisting) before a campaign this size opens.