Forward Deployed Engineer interview questions

Since no certificate exists, the assessment is a conversation. Below are ten questions and, beside each, what it sounds like when somebody actually did the thing against what it sounds like rehearsed. The rule underneath all ten: ask about behaviour rather than knowledge, and listen for texture rather than structure. A rehearsed answer is smooth, sequential and has no names in it.

Here is a problem, badly described. What do you need to know?

This is the whole job compressed into ten minutes. Describe something real and unfinished out loud, the way a manager would, and then stop. What is measured is not the solution but which questions come back, because in most situations only two of them decide anything.

Good sounds like the volume question and the ownership question. How often does this happen, and who is allowed to change the rule. Frequently the third thing said is a request to come and watch it happen once.

The rehearsed version produces an architecture inside a minute. A pipeline, a model, somewhere to hang a dashboard. Fluent, confident, answering a question nobody asked yet.

Can you describe one thing you built so that it could not have been invented?

A made-up project is smooth. A real one has an awkward day in it: an export that changed format, a field that was empty exactly once, a colleague who would not hand over their spreadsheet. Keep asking for detail until the story either produces texture or runs out.

What you want is numbers with denominators attached, the thing that broke, what they would do differently, and who else did the work. The wrong turns carry more weight than the successes, because they place the person at the scene.

Rehearsed, it arrives as a clean arc from problem to percentage, with no names and no dates. Ask what the percentage is a percentage of. Either the answer comes too smoothly or it does not come at all. The format done honestly, on one real case.

What did you talk somebody out of?

The tell is who was annoyed. A real no has a person attached to it: somebody who wanted the thing, and who had to be given a reason rather than a refusal. If the story contains no unhappy person anywhere in it, they are describing a project that was cancelled for them.

A strong answer names the alternative offered instead, because a no that leaves nothing behind is only a refusal. Some candidates also name what it cost them at the time, which is the part nobody thinks to rehearse.

Weak sounds like a principle. “I always challenge requirements”, with no incident underneath it. Ask for the date and the name and the story thins out. The reasoning behind a real one is set out separately.

Who did you speak to that nobody sent you to?

The useful information in any company sits with people who are not on the introduction list. Somebody has to decide, unprompted, to go and find whoever actually does the thing. Nobody grants that permission in advance, which is exactly why the question is worth asking.

Real answers contain a specific person doing a specific job, and what changed in the plan afterwards. The best ones include a moment where a conversation on the floor contradicted the brief and the brief lost.

The polished version says “I engage all stakeholders”. A list of roles rather than a person, in the passive voice. Nobody who genuinely did this calls it engagement.

How would you find out what somebody does too often?

Two different operations hide inside that question, and a good candidate separates them unprompted. Counting tells you how often something happens. Only the person doing it can say whether that counts as too often. Mix the two and you produce management’s irritations dressed up as data.

You want to hear about sitting beside somebody and going through the work with them rather than about them, and about asking what is enjoyable as well as what is not, since those parts are the ones you must not take away.

Rehearsed, the answer is a tool. Process mining, a time study, a workshop with coloured notes. None of it is wrong. All of it answers “how often” when the question was about “too”.

What did you find out that nobody wanted to hear?

In most companies of any age there is a routine done the long way round since before anybody thought to question it, and its author is still on the payroll. Finding that is easy. Saying it without making an enemy decides whether anything else gets heard.

Listen for information rather than a verdict: here is what it costs, here is what changed since you designed it, here is what I would do. Given to the person first, before it went anywhere near their manager.

The answer to avoid is the one where the candidate is the only competent person in the building. That tells you how they will describe your company at their next interview.

Who resisted you, and were they right?

Resistance to this work is usually rational. A colleague near the end of a long career can see automation coming towards them and be right about it. Somebody who has made a task their own can feel replaceable, because in a narrow sense they are. A candidate who has never met this has shipped little.

The answer you want concedes the point, then describes a middle way: the company has to stay viable and the people worth keeping have to feel valued. Ask what they promised that person, and whether it held.

Rehearsed, it comes back as change management. Resistance framed as an obstacle removed rather than as information about risk. In that version nobody who objected ever turns out to have been right.

What stopped you, and what did you do next?

Ask this straight after any story that ended well and watch what fills the gap. You are listening for the point where something was genuinely blocked: a refused export, a supplier who would not answer, a rule nobody would change. The recovery is the answer, not the obstacle.

In a real answer the sentence turns. It could not be done that way, so they found the report that already held the number, or asked a different department, or built the ugly version first and replaced it later.

In a rehearsed one the sentence stops. It was not possible because of the system, the budget, the client. Every reason is true, and the work still sits on somebody else’s desk.

How do you know the model was not just agreeing with you?

A model does what you ask. Ask a second time and the answer improves, but it is still doing what you told it, which is not always what you needed to hear. The question is whether this person has a habit for catching that, or has simply never noticed it happening.

Good answers describe going out of the way to disprove their own idea: checking output against a case where the answer was already known, or asking for the opposite argument and reading it honestly rather than for reassurance.

The thin answer is prompt technique. Better prompts buy better compliance, and compliance was the problem. A complicated solution does not come out of anything that only agrees with you.

What should you be asking the employer back?

If you are the one being interviewed, your questions are half the assessment, and the useful ones are all about access. Who will I be allowed to sit with. Who decides when two systems disagree about a number. What happens if my answer turns out to be that nothing should be built.

The good sign is that names come back straight away rather than roles. Somebody has already worked out who you will sit beside and who settles arguments, which means the role was designed rather than announced.

“We will sort that out once you start” is not flexibility. It means nobody has decided, and your first months go on finding out who is allowed to say yes. What that side of the table should have prepared is here: the employer’s side of this.

Marc Herdes, Almere, 4 September 2026. I have asked these from one chair and been asked most of them from the other.