Sixty applications arrive for a role. The first eight get read carefully. The next twenty get skimmed. By application forty the reader has developed a shortcut that nobody wrote down and cannot describe, and by application sixty they are looking for a reason to stop.
The candidate at position fifty-one was not evaluated by the same process as the candidate at position three. This is the central problem of screening, and it has nothing to do with how much anyone cares. Attention is a depleting resource and consistency is what depletes first.
Software is usually offered as the answer. It is not, or not mostly. A knockout question in a form is only as good as the thinking behind the question, and most screening tools automate a process that was never defined. What fixes screening is defining it, which is free, and then applying it in a fixed order, which is discipline rather than technology.
Write the criteria before you read anyone
The single highest-leverage act in screening happens before the first application is opened.
Criteria written after you have started reading are contaminated by the people you have already read. A strong first candidate quietly becomes the standard, and the requirement list reshapes itself around them. A weak first batch does the opposite, and the bar drops without anyone deciding that it should.
Write three lists, in this order.
Everything in a job description belongs in exactly one of these three, and most descriptions never separate them.
The must be true list is the shortest and the hardest to write honestly. It contains only the facts that make the job impossible without them: availability for the shift the role exists to cover, work authorization, a license the role legally requires, the physical requirements where they are real. If an item can be worked around, it is not on this list.
The hard to teach list is what the shortlist is actually ranked on. Reliability under a bad week. Judgment with a distressed client. The ability to read a drawing. Enough language in common with the team to be safe. These take months or years to develop and are the reason one candidate is genuinely better than another.
The trainable list is everything a competent adult can learn in weeks. Your specific software. Your documentation format. Your escalation path. Trainable items belong in the notes, never in the filter. Filtering on them is the most common way a search removes its own best candidates, and it is invisible when it happens.
Screen in a fixed order
With the lists written, screening becomes mechanical in the good sense.
Check the must-be-true facts first, on every application, before reading anything else. This feels wrong because it treats a person as a set of conditions, and it is right because it protects your judgment for the decisions that need it. A candidate who cannot work nights is not a candidate for a nights role, and discovering that in paragraph four of a careful read is forty seconds you cannot get back sixty times over.
Only then read for the hard-to-teach list, and only among those who passed.
The third pass is the one most people skip: read the rejections for a pattern. If forty of sixty applications failed on the same must-be-true item, the item is worth examining. Sometimes it is a genuine legal requirement. Often it is a preference that was written as a rule, and it has just removed two thirds of the market.
Write down what a pass looks like
Two people screening the same pile against the same list will still produce different shortlists, because "good communication" resolves differently in two heads.
The fix is to write the evidence, not the quality. Not good communicator but has held a role requiring handover to a colleague at shift change. Not reliable but no unexplained gap over three months in the last two years. Not experienced but has done this specific task, in a comparable setting, within the last three years.
Criteria that drift
- Good communicator
- Reliable
- Team player
- Experienced
- Professional
- Culture fit
Criteria that hold
- Has handed over at shift change
- No unexplained gap over three months
- Has worked a rota with shared coverage
- Did this task, this setting, last three years
- Turned up to the scheduled screening call
- Understood the shift pattern when asked
A criterion that cannot be checked by someone else is not a criterion. It is an impression with a job title. The right-hand column is also far more defensible if a rejection is ever questioned, which is a secondary benefit that occasionally becomes the primary one.
The one to be most careful with is culture fit. In practice it is where unexamined preference collects, and where a screening process quietly starts selecting for similarity to the person screening. If it means something specific, write the specific thing. If it does not, remove it.
What a shortlist is for
A shortlist is not the best candidates. It is the candidates worth spending an hour on.
That distinction changes its size. A shortlist of twelve is not a shortlist, it is a slightly shorter list, and it defers the actual decision to a stage that costs far more per candidate. Four to six is usually right for a single frontline opening. Getting there requires being willing to reject someone reasonable, which is the part that feels uncomfortable and is the entire function of the exercise.
- 1SecondsReject on must-be-true, no readingAvailability, authorization, required license. Mechanical, and it protects the attention needed later.
- 2MinutesRank the remainder on hard-to-teach evidenceRead for the specific evidence written in advance, not for a general impression of the person.
- 3MinutesCheck the rejection patternIf most applications failed one item, examine the item. It is often a preference written as a rule.
- 4One callConfirm the facts that decide the outcomeShift, start date, pay expectation, anything that would end the process at offer stage. Cheaper here than after two interviews.
- 5Same dayShortlist four to six, and reject the rest promptlyA prompt no is the part of the process a candidate remembers, and rejected candidates are the cheapest source for the next opening.
The call that saves two interviews
Before the shortlist goes to interview, one short conversation removes most late-stage failures. It confirms the facts that would end the process anyway: the actual shift, the actual start date, the pay expectation, the commute as the candidate understands it.
Every one of these is capable of ending a hire at offer stage, and every one is knowable in four minutes. An operator who has ever run two interviews and a reference check before discovering a pay mismatch has already paid for this call several times over.
Where a machine genuinely helps
Two parts of screening do not scale with care, and those are the two worth automating.
The first is the must-be-true pass, which is mechanical by construction. The second is replying. Sixty applicants each expecting an answer is a real workload, and it lands on someone whose main job is elsewhere.
This is the shape of what Kariaa does on the employer side. Applications arrive checked against the requirements you set, each requirement marked passed or failed individually, with the evidence that produced the verdict and a confidence level attached, plus an advisory fraud read. The output is a table with a checks column and a fraud column that you can sort and export, so the mechanical pass is already done and the ranking judgment is still yours.
The confidence level matters more than the verdict. A check that says passed with low confidence is telling you the evidence was thin, which is a different instruction from passed. Any screening aid that returns a bare yes or no is hiding exactly the information that should decide whether you read the file yourself.
Moving someone forward is one of four next steps: a video interview, a set of questions, a booking, or a message. The reply reaches the candidate in WhatsApp or Telegram, which for this workforce is the difference between an answer and silence. That is examined in How to Reduce Interview No-Shows.
Key takeaways
- Write the criteria before reading the first application. Criteria written afterwards are shaped by whoever you happened to read first.
- Separate must-be-true from hard-to-teach from trainable. Filtering on trainable items is the most common way a search removes its own best candidates.
- Screen in a fixed order so that expensive judgment is only spent on people who could actually take the job.
- Write the evidence, not the quality. A criterion a second person cannot check is an impression, not a criterion.
- A shortlist is the candidates worth an hour, not the best candidates. Four to six for a single frontline opening.
- One four-minute call confirming shift, start date and pay expectation removes most offer-stage failures.
Common questions
How do I screen job applicants without an applicant tracking system?
- Write three lists before reading any application: what must be true or the person cannot do the job, what is hard to teach, and what is trainable. Then screen in that order, rejecting on the must-be-true list mechanically before reading anything in detail. The method is the order of operations, not the tool.
What should be on a screening criteria list?
- Only facts that would make the job impossible without them belong on the disqualifying list: availability for the required shift, work authorization, a license the role legally requires, and genuine physical requirements. Anything that can be worked around belongs on a lower list. Skills a competent adult can learn in weeks should be noted, never used to filter.
How many candidates should be on a shortlist?
- Four to six for a single frontline opening. A shortlist is not the best candidates, it is the candidates worth spending an hour on. A list of twelve simply defers the decision to a stage that costs far more per candidate.
Why do two people screening the same applications disagree?
- Because criteria like good communicator, reliable or culture fit resolve differently in different heads. Writing the evidence instead of the quality fixes it. Not reliable but no unexplained gap over three months in the last two years. A criterion a second person cannot check independently is an impression rather than a criterion.
Should I use AI to screen candidates?
- It helps most with the mechanical pass and with replying at volume, which are the two parts that do not scale with care. Insist on seeing the evidence and a confidence level behind each verdict rather than a bare yes or no. A pass returned with low confidence means the evidence was thin, which is a different instruction from a pass, and a tool that hides that is hiding the thing that should decide whether you read the file yourself.
Written by the team at Kariaa. Learn more at www.kariaa.com.