Structured interviews: asking everyone the same questions

When each interview follows wherever the conversation goes, you finish the process holding a set of impressions that cannot be placed alongside each other. One candidate was asked in depth about handling a difficult customer; another was not, because that conversation went somewhere else. You now have detailed evidence about one person on a dimension where you have nothing about the other, and no honest way to compare them.

What fills that gap is a general sense of how each interview felt, which is exactly the input a hiring process should be trying to reduce. It also has a specific distorting effect worth knowing: the candidate who is most comfortable in unstructured conversation appears strongest, because a free-flowing interview rewards conversational ease. For a small number of roles that is the skill being hired. For most it is unrelated to whether someone can do the job, and it correlates with confidence, class background and familiarity with interview conventions rather than with capability. Structure does not remove judgement from hiring — it makes sure the judgement is exercised on the same evidence for each person.

The question set should be derived from the job description rather than assembled from a list of good interview questions. Take the things the role actually requires — the ones you identified as genuine must-haves — and write one or two questions per requirement that would produce evidence about it. If the role involves handling upset customers, there is a question about that. If it involves working unsupervised, there is a question about that.

The most reliable form is a question about something the candidate has actually done, rather than what they would do. 'Tell me about a time a customer was angry with you and what you did' produces a real account with details you can probe. 'How would you handle an angry customer?' produces a description of good practice that anyone can supply, which tells you they know the right answer rather than that they have ever managed it. Aim for a small set — five or six core questions is usually enough for a first interview and leaves room for follow-up. It is also worth writing down what a good answer would contain before you start interviewing, because deciding that in advance is what stops the standard drifting between the first candidate and the fifth.

Structured does not mean scripted, and this is the misunderstanding that puts people off the idea. The core questions are asked of everyone in the same words, because that is what makes answers comparable. What happens after each answer is entirely open: probing, asking for specifics, asking what happened next, asking what they would do differently. That is where most of the useful information actually comes from.

The distinction is between the stimulus and the exploration. Keeping the stimulus identical means every candidate got the same opportunity to demonstrate the same thing. Letting the exploration run freely means you are not throwing away the thing an experienced interviewer is genuinely good at, which is noticing that an answer is vague in a particular way and finding out why. In practice this also solves the objection that structured interviews feel robotic. They only feel robotic if the follow-up is suppressed as well, which nothing requires. An interview with six fixed questions and unlimited probing feels like a normal conversation to the candidate, and produces a record you can lay next to five other candidates without pretending the comparison is valid when it is not.

A rubric is just a written description of what a weak, adequate and strong answer looks like for each question, decided before the interviews start. For a question about handling an angry customer, a strong answer might describe a specific incident, what they actually said, how it resolved, and what they changed afterwards; a weak one describes general principles with no incident attached.

Writing these down in advance does two things. It stops the standard moving as you go, which otherwise happens invisibly — the fourth candidate is judged against a bar shaped by the first three rather than against the job. And it makes disagreement between interviewers productive, because two people who scored the same answer differently can look at the rubric and establish whether they heard different things or are applying different standards. Score immediately after each interview rather than at the end of the day, since recall degrades quickly and the details that distinguish answers are the first to go. Record the evidence next to the score, not just the number: 'described the refund dispute in Feb, escalated appropriately, no follow-up action' is reviewable months later, while a score of three out of five is not.

For a customer-facing role, the questions that produce the most signal are about specific past incidents: a time a customer was angry and what they did, a time they did not know the answer to something a customer asked, and how they handled a busy period when several people wanted attention at once. Each of these has a right answer that cannot be guessed from general knowledge, because it requires having been there.

For an administrative or back-office role, ask about a time they found an error in someone else's work and what they did about it, how they keep track of tasks arriving from several people, and what happened the last time they missed something. That last question is the most informative single question in most interviews, because the answer reveals whether a person can describe their own mistake accurately — and someone who cannot name one is either inexperienced or unwilling to be candid, both of which matter. For any role with responsibility for money or stock, ask what they would do on discovering a discrepancy, and listen for whether their answer involves telling someone. What you are testing is not cleverness but whether their instinct is to escalate or to quietly resolve it.

Structure makes evidence comparable. It does not make it correct, and it is worth being clear about the limits. A candidate can give a strong, specific, well-evidenced answer about an incident that did not happen the way they describe, or at all — an interview cannot verify a past event, which is what reference checks and work samples are for. A structured process applied to the wrong criteria produces consistent, defensible, well-documented decisions about things that do not predict performance in the role.

It also cannot tell you how someone will behave once the novelty of a new job has worn off, or whether they will get on with the specific people they will work beside. Those are only established by working together, which is what the probation period is genuinely for. The honest claim for structured interviews is narrower than the enthusiasm around them sometimes suggests: they stop you comparing candidates on evidence you only gathered for some of them, they reduce the weight of conversational ease, and they leave a record you can explain. That is a substantial improvement over an unplanned conversation, and it is not a prediction of how the hire will work out.