Screening

Eighty applications for one role cannot all become interviews. Budget an hour and a half per interview including preparation and notes and that's 120 hours — three weeks of one person doing nothing else, and at the end of those three weeks the earliest applicants have gone elsewhere.

So a filtering step goes in front. The trouble arrived when that step ran without criteria. At Pagedeck, two people reviewed the same application separately: one passed it, the other rejected it. Asked what they had based the call on, both described an impression. One had judged on the size of the previous employer. The other had judged on a résumé that wasn't formatted tidily.

Without criteria, whatever catches the eye first becomes the criteria: the school, the last company's name, the layout of the document. What those share is a weak connection to the work the role actually involves. Narrow eighty candidates to ten on weakly connected criteria and there is no reason to think the surviving ten are the strongest ten. They are the ten who happened to look most familiar to whoever was reading that afternoon.

Filtering nobody isn't the answer either. Put thirty people in front of interviewers and interviewer time runs out, which shortens the time spent per candidate and drags down the accuracy of the interviews themselves. Screening exists to protect the time available in the later stages, and its current shape is an ordering designed to give that filtering a basis.

Five steps.

1. Convert the required qualifications in the job description into three or four items. The job description sets out what the role does and what someone needs to be able to do, and the required qualifications written there are exactly what gets checked in an application. The point of this step is that no criterion absent from the posting gets used in the review.
2. Rewrite each item into something an application can actually show. "Collaboration skills" can't be verified from a document; "shipped a product on a team of three or more" can. Anything that cannot be verified from a document comes out of screening and moves to a later stage where it can be.
3. Have two people review the same application independently. They judge without seeing each other's calls and compare afterward. At high volume, one person reviews first and only the unclear cases go to a second reader.
4. Use three verdicts, not two: pass, unclear, reject. Without an unclear column, a reviewer who isn't confident pushes the application toward reject. Anything marked unclear gets one item confirmed in a 15-minute call and is then judged again.
5. Send results inside a fixed deadline, rejections included. A company that doesn't reply has that fact recorded in reviews on hiring sites, and six months later it will be sending that same person a sourcing email.

The share of applications landing in the unclear column reports on the health of the process. Twenty unclear out of eighty means the items from step 2 are not yet in verifiable form, and rewriting them costs less than making twenty phone calls.

The first reader is usually someone in HR or a recruiter. The hiring manager then reviews whoever passed. That's the person who actually needs the role filled and makes the final call, and unless they set the step 2 items themselves, the criteria shift a second time further down. What that looks like is the manager cutting the ten passes to five. Those five were filtered against the manager's criteria, which means the criteria used to select the original ten went unused.

Speed matters here. Three business days from application to decision is the working standard. Strong candidates don't apply to one company, and most settle on where they're going within two weeks. Take ten days over a screening decision and by the time you're booking an interview they're in someone else's final round. That delay accumulates directly in time to hire — the span from posting the role to a confirmed start.

Two numbers are worth revisiting periodically: the pass rate, and the eventual offer rate among those who passed. A high pass rate with a low eventual rate means screening is under-filtering and spending interviewer time. A low pass rate with an eventual rate near 100 percent suggests the filter is tight enough that people worth meeting are being removed.

"This one can't be verified from an application, so let's check it in the interview."

A proposal to distribute criteria across the stages rather than asking one stage to answer everything.

"The ten unclears get one item confirmed on a 15-minute call, then a second look."

A procedure for not pushing uncertainty into the reject pile, since a reviewer's low confidence is not the candidate's problem.

"Applications have been sitting for three days. Please decide today."

Delay here pushes every later stage back with it, and the candidates lost to it are never recorded anywhere.

Pagedeck changed its approach after two reviewers split on the same application. The required qualifications for the engineering role were converted into three items: shipped a product on a team of three or more, owned a feature through to release, and worked with something overlapping the current stack.

In the next round, 14 of 82 applications passed and 11 landed as unclear. The 11 got one item each confirmed on a short call, and six more moved forward. This time the hiring manager didn't cut the pass list down again, because the items had been written by that manager to begin with — which is the whole reason step 1 belongs to the person making the final call.

The most common reading is that screening is where you pick the best people. It's a filter, so it looks like taking candidates in order from the top. But what this stage can establish isn't ranking — it's overlap. It asks only whether there is overlap with the requirements in the posting. Try to rank and you start hunting for comparable attributes, and the school and the previous employer's name become the criteria again — the exact two things the process was built to stop using.

The other mistake is treating a heavy cut as a good cut. Reducing eighty applications to five makes the later stages easier, so it feels like competence. The way to check is what happened to those who passed. Five passes all receiving offers isn't evidence of accuracy; it's closer to a signal that people worth meeting were deleted in advance. There's no need for screening to pre-remove everyone who would be filtered out by a take-home assignment — practical work completed at home — or a technical interview, which examines job skills directly.