A Must-Have Is Any Requirement You Are Willing to Test For

How do you tell a must-have from a nice-to-have?

Ask whether any stage of the process will test it, and what happens to a candidate who fails. If both answers exist, it is a must-have. If either is missing, it is a preference, however important it feels. That rule is useful because the answer sits inside your own process, where you can go and look.

Run it against three lines from a real posting and the sorting is quick:

• Written and spoken fluency in Spanish. The recruiter screen is conducted in Spanish and a candidate who cannot hold the conversation does not advance. Tested, with a rule. Must-have.
• Experience in a regulated industry. Nobody asks about it, nobody scores it, and no answer would end a candidacy. Preference. It goes in the description or it goes.
• Comfortable with ambiguity. No stage owns it and no answer fails it. This is a mood, not a requirement, and it belongs in the paragraph about the team.

The rule also moves the argument off the candidate and onto the process, where it can actually be settled. Two people can argue forever about whether a regulated-industry background matters. They cannot argue about whether anybody is going to ask.

Why does the shorten-the-list advice not work any more?

Because its mechanism was applicant self-selection, and an AI-assisted application removes it. The advice says a long list deters qualified people, especially women, so cutting to five or seven bullets widens the pool. Whatever that effect was, it ran through a reader deciding not to bother, and the cost of bothering has collapsed.

The circulating version of that claim says women apply only when they meet 100% of the requirements and men at 60%. Its usual citation is a 2014 Harvard Business Review piece, and that piece attributes the figure to an internal Hewlett-Packard report the author was openly skeptical of, then offers her own survey against it ¹. What the survey found, from over a thousand predominantly American professionals, is that the top reason both sexes gave for not applying, twice as common as any other, was believing the employer would not hire them because they did not meet the listed qualifications (41% of women and 46% of men), and the least common answer of all was not being able to do the job well ¹.

Measured, the gap is much smaller. In an online experiment with 10,468 working-age participants, the Behavioural Insights Team found men said they would apply when they met 52.1% of an advert's requirements and women at 55.7%, and the gap appeared only among the less qualified half of the sample ². Nobody applied to anything: participants rated willingness on a seven-point scale for one fictional role, and both headline percentages move with where the researchers drew the line for would apply ².

The same trial carries a warning for anyone about to rewrite the list in more detail. Willingness to apply was greater for the vague high-level-skills version of the advert than for the version spelling out required behaviors, and the specific version failed its own manipulation check: readers did not rate it as any clearer ². That is an exploratory result from a manipulation the authors say failed, not a finding that detail repels people. The usable part is narrower and still worth having: adding detail does not, on its own, mean anything was communicated.

The collapse in the cost of applying shows up at the other end of the funnel. In the recruiting software vendor Ashby's own customer base, which skews to technology companies, the average recruiter now processes 291 applications per hire against roughly 100 in early 2021, and Ashby names automated submissions among the drivers without sizing how much of the rise they account for ³. A shorter requirement list does not move that number.

Map every requirement to the stage that tests it

Open one document with the requirement list down the left and the interview plan down the right, then draw a line from each requirement to the stage that will test it and name the person who owns that stage. Anything with no stage and no owner is an orphan. Orphans are the edit: delete them, move them into the description, or add the stage.

The exercise takes about thirty minutes and it usually produces three findings. Some requirements are tested twice, by two stages that do not know about each other, which is where a loop gets long for no reason. Some are tested by nobody, which is where a rejection with no explanation comes from. And one or two turn out to be tested by a stage that cannot actually see them: a phone screen cannot assess whether somebody writes clearly, however carefully the recruiter listens.

Do it before the requisition opens, not after applications arrive. Once the pipeline is full, every line is attached to somebody who has already been rejected under it, and the list stops being editable. This is the same discipline as writing requirements you can actually test, applied to a list that already exists.

One caution when the requirement is about AI use, because it is the line most likely to arrive without a stage attached. A bullet saying experience with AI tools has no test behind it and no rule that fails anybody, so it sorts on whether the applicant thought to write the words down. Naming the task and the check turns it into something a stage can own, which is the whole of how to write the AI requirement without drowning in unqualified applicants.

What an untested must-have costs when a rejection is questioned

An explanation you cannot give. US federal selection law treats the whole range of assessment as one category: the Uniform Guidelines define a selection procedure as any measure, combination of measures, or procedure used as a basis for any employment decision, and say so explicitly for everything from paper tests through informal or casual interviews and unscored application forms ⁴.

That definition cuts in the direction people do not expect. Calling a stage just a conversation does not move it outside the rules, and a requirement applied by feel is applied by a selection procedure with no record of how. Once a complaining party shows that a particular practice causes a disparate impact under Title VII, the statutory test added by the Civil Rights Act of 1991 asks the employer to demonstrate that the practice is job related for the position in question and consistent with business necessity ⁵. A requirement no stage tested leaves nothing to point at when that question arrives.

Those two rules are public legal fact rather than advice, and both are United States federal law: the Uniform Guidelines date from 1978 and the burden-shifting text is 42 U.S.C. 2000e-2(k), added in 1991. Whether a particular requirement is defensible for a particular role belongs in front of counsel. The practical version needs no lawyer, though. A requirement with a stage, a rule and a written record of how it was applied is a requirement you can describe to a candidate, to a hiring manager, and to anybody who asks later.

So the list gets shorter in the end, for a reason the templates do not give. It shortens because each surviving line has to earn a slot in a loop somebody is running, and loops are finite. That is a better constraint than a bullet count, because it comes from the work the team has agreed to do.

See a sample report >