Most interviews fail before anyone enters the room, at a step most teams skip without noticing they skipped it: nobody decided what the interview was for. In my recruiting years, when I asked a hiring manager what we were actually looking for, the most common answer was a recitation of the job ad. Then we would pick questions the way you pick songs for a party, by what felt good, and afterwards compare candidates on whatever each interviewer happened to remember.
Job analysis is the unglamorous fix, and it is the first step of every structured interview worth the name. This article gives you the lightweight version: three questions, maybe three minutes of honest thinking, and you will know exactly what your interview should measure and why.
What a job analysis actually is
Strip away the HR-textbook apparatus and a job analysis is one decision: what does this job actually require, and which of those requirements should the interview test? You are looking for a short list, the four to six capabilities that separate people who succeed in this role from people who merely survive it.
The formal discipline behind that decision is real. Industrial-organizational psychologists run task inventories, structured questionnaires, and panels of subject-matter experts; the US government's O*NET database rates about a thousand occupations against a common item pool this way. If you hire hundreds of people into one role, that machinery is worth its cost. If you are hiring one support lead this quarter, you need the same logic at a fraction of the dose, and the good news is that the logic compresses well.
A cook decides what to eat before shopping for ingredients. That is the whole idea. The rest of this article is about doing it deliberately instead of letting the job ad decide for you.
Why it comes first
The evidence for structured interviews is unusually strong: in the largest modern re-analysis of selection research, they predict job performance better than any other method, roughly doubling the validity of unstructured conversations. But that headline number is an average, and the average hides a wide spread. Some structured interviews approach the ceiling of what any selection method achieves; others perform barely better than winging it. How well you structure is the whole game.
And the very first component of structure, in the classic taxonomy of what "structured" even means (Campion, Palmer and Campion, 1997), is this: base the questions on a job analysis. Not "use good questions". Questions chosen for the wrong role are wrong questions, however well-crafted. An interview is a measurement instrument, and an instrument can only measure what you point it at.
There is a fairness dividend too. Criteria decided in advance and tied to the job are exactly what closes the openings bias walks through, which is why the US federal government's interviewing guide builds every structured interview on a job analysis: it is what makes a hiring decision defensible, to a candidate or to anyone else who asks.
The job description is not your job analysis
The tempting shortcut is to open the job description and interview against it. Resist it, for two reasons.
First, most job descriptions describe a person, not a job: years of experience, tools, degrees, and a trait salad ("self-starter with excellent communication skills"). Veteran hiring writers like Lou Adler have made a career of pointing out that these are person descriptions, lists of credentials someone once hoped would correlate with the work. What the work actually demands is hiding a level deeper.
Second, your job ad is marketing. You wrote it to attract applicants, possibly by inheriting the previous ad, possibly under a title the market searches for rather than the one that fits. It is an advertisement for the role, and interviewing against your own advertisement is grading your own homework.
The raw material for a real analysis is not the ad. It is the work, and the people who have done it well.
How to do a job analysis in three steps
Here is the lightweight method. It is distilled from what the serious practitioner playbooks converge on: Greenhouse's structured hiring kickoff, Google's attribute-first interviewing, the scorecard method in the book Who. They differ in ceremony but agree on the moves.
Step 1: name the outcomes
Ask yourself, or better, ask the hiring manager: what should this person have accomplished six to twelve months from now for you to call the hire a success? Write down two to four answers.
The discipline is in the phrasing. Outcomes are observable: "shipped the new onboarding flow", "cut the support backlog by half", "two new hires ramped and holding the quality bar". If you catch yourself writing trait words ("brings energy", "is strategic"), you have drifted from what the job produces back to what the person is like, and the whole point of this step is to hold that question off until the evidence is on the table. Greenhouse's role kickoff opens with almost exactly this question, and it opens with it for a reason: everything else in the intake follows from the answer.
Keep your own words. Do not translate "stop us from losing enterprise customers over slow responses" into "customer focus". The verbatim phrasing is the sharpest version of your requirement, and you will want it later, when you are scoring answers against what you actually meant.
Step 2: remember who was good at this
Now the oldest trick in the job-analysis book, descended from Flanagan's critical incident technique (1954), which built selection criteria by collecting concrete incidents of notably effective work: think of someone who did this kind of job well. What made them good?
Be specific, and think in incidents rather than virtues. Not "Maria was great", but what Maria did: never escalated a problem without a proposed answer, stayed level with furious customers, told you within the hour when the team had promised something it could not deliver. One or two sentences is plenty.
This question works because it breaks the vagueness that step 1 can leave behind. Outcomes tell you what the job must produce; the person who produced them shows you what producing them takes. Between the two, the next step almost answers itself.
Step 3: choose four to six competencies, and stop
Now convert what you wrote into the things your interview will test. Work from a stable, finite list; ours is ten competencies, distilled from the frameworks that hiring research keeps converging on. Read your outcomes and your incident notes, and pick the competencies they require. Then apply the two rules that give this step its teeth:
Cap the list at six, and flag at most two as must-assess. Every serious framework lands on a handful: Google interviews against four attributes, and the UK Civil Service defines nine behaviours but assesses only a few per role. The cap is not a limitation to work around. Every competency you add costs interview minutes you are not spending going deep on the ones that matter, and a scorecard that tries to rate everything ends up discriminating nothing. Four to six is the sweet spot; the must-assess flags are your tiebreaker when interview time runs short.
Let some requirements stay job-specific. If the role genuinely turns on something outside any general framework, medication administration, say, or fluency in Finnish, write it down as its own requirement rather than shoehorning it into the nearest competency. Just be honest about whether an interview question is the right way to check it: technical skills are often better verified with a work sample or a practical task, while the interview does what it does best, which is past behavior.
A worked job analysis example
Say you are hiring a support lead for a forty-person company. The three steps might produce this.
Outcomes. First response times back under four hours by summer. Two new support hires onboarded and holding quality without daily supervision. An escalation playbook that exists and gets used.
Who was good. "Elin, at my last company. She never forwarded a problem without a proposed answer. Angry customers calmed down talking to her. And when we had promised something wrong, she said so before the customer found out."
The pick. Response times and angry customers point at customer orientation; flag it must-assess, because the outcomes say the role lives or dies there. Onboarding two hires is leadership, the developing-others half. The playbook, and "holding quality without supervision", suggest planning and delivery. Elin's saying-so-first is integrity. Four competencies, one flag, and a defensible reason for each written in your own words. Notice what didn't make the list: nothing about the helpdesk stack, which a short practical walkthrough will verify better than any question.
The payoff is that question selection, the step most people start with, has become almost mechanical. You are no longer browsing for clever questions; you are filling four labeled slots. For the must-assess slot above, a card practically volunteers itself:
That is what a question chosen by a job analysis looks like: you can say, in one sentence, which requirement it tests and why this role needs it.
From analysis to interview plan
The analysis is not a warm-up exercise to be thrown away once the questions are picked; it is the spine of the interview guide. Give each chosen competency its own section with two or three questions, ask every candidate the same ones, and score each answer against anchored criteria rather than against charisma. Put the analysis itself at the top of the printed guide: the role, the outcomes verbatim, the must-assess flags. Interviewers who know what the interview is for probe better and drift less, and when you compare candidates afterwards, you compare them against the role instead of against each other's charm. The guide-creation step of this path picks up there.
One honest caveat: a lightweight analysis is a hypothesis, not a revelation. If the first two interviews teach you that the role turns on something you missed, revise the brief and carry the revision forward to the remaining candidates. The failure mode is not writing an imperfect analysis; it is interviewing without one.
Do the three minutes
You can run this whole method with a notepad. Or you can use the planner, which walks you through exactly these steps: the outcomes prompt, the who-was-good-at-this prompt, and the capped competency pick, and turns your answers into a working interview plan with the analysis printed on top. It is free, it takes about three minutes, and no account is needed to think.
Either way, do the three minutes. No other preparation pays off like it: every question, probe, and score downstream inherits its relevance from this one decision about what the role actually requires.