Problem Solving Interview Questions

These questions assess how someone behaves when the path is not obvious - whether they structure the problem, or start moving and hope structure appears. Brainteasers measure something real - they load on general mental ability, which does predict performance - but they measure it worse than a plain reasoning test would, and add practice effects and a hostile candidate experience on top. The questions below use problems the candidate actually faced, where you can check whether the reasoning survived contact with reality.

Ten problem solving interview questions

Question	A strong answer	A weak answer
Tell me about a problem where your first diagnosis was wrong. How did you find out?	Describes the evidence that contradicted them and how quickly they let go of the first theory. In our experience this is the question that separates candidates fastest.	Cannot recall being wrong, or the 'wrong' diagnosis turns out to have been someone else's.
Walk me through the last problem you solved that had no clear owner.	Explains how they decided it was theirs to take, and who they told. Ownership under ambiguity is rarer than analytical skill.	Escalated it and waited.
What is a problem you decided not to solve?	A deliberate call with reasoning - cost, reversibility, or something more important. Shows judgement about where effort goes.	Treats every problem as worth solving, which usually means none were prioritized.
Describe a time the data disagreed with what everyone believed. What did you do?	Checked the data first, then made the case with specifics. Best answers include what it cost them socially.	Either deferred to consensus, or claims they were vindicated without describing how they persuaded anyone.
How did you break down the largest problem you have worked on?	A real decomposition with the seams they chose, and why those seams. Vague 'broke it into smaller pieces' is not an answer.	Describes the solution without the decomposition, which usually means someone else did that part.
Tell me about a fix you shipped that you knew was temporary. What was the plan?	Honest about the trade-off, explicit about what would trigger the real fix, and says whether it ever happened.	Claims never to have shipped a temporary fix. Everyone has.
When did you last solve a problem by removing something rather than adding?	A concrete deletion - a step, a feature, a meeting - and the effect. Tests a habit most candidates do not have.	Cannot produce an example, which is common and worth noting rather than penalizing heavily.
What did you do the last time you were properly stuck?	A specific unsticking method - changed the representation, asked a particular person, went back to the raw data - rather than 'kept trying'.	Describes persistence with no change of approach.
Give me an example of a problem where speed mattered more than getting it right.	Recognizes that such cases exist, names one, and describes what they consciously traded away.	Insists on always getting it right, which in most operating environments is a way of being slow.
How do you know when a problem is actually fixed?	Names a signal they watched afterwards. Strong candidates have been burned by declaring victory early.	Treats deployment or handover as the end state.

Give the candidate the space to think out loud. Watching someone structure a problem badly is more informative than hearing a tidy account of one they solved months ago.

Scoring rubric

Rating	What it looks like
Strong	Structures out loud, changes approach under new evidence, and can say what they chose not to do.
Mixed	Solves it, but the reasoning is reconstructed after the fact and the dead ends have been tidied away.
Weak	Jumps to a solution, or describes a process with no specific problem attached to it.

Score against evidence, not delivery, and write your rating down before the panel discusses it - the first opinion spoken aloud anchors everyone else's.

Running these consistently

The signal is not whether they solved it. It is whether the reasoning changed when the evidence did - so push on the first wrong theory in every answer.

Three or four questions asked the same way of every candidate produce comparable evidence; the same questions asked ad hoc across five interviewers produce five impressions and an argument. The Cognitive works the same way - you set the rubric, the AI interviewer composes each follow-up live from the role and the candidate's answers rather than reading a fixed list, and every rating comes back anchored to a quote and a timestamp.

Frequently Asked Questions

Are brainteasers ever useful?

Only where the job is literally puzzle-solving under time pressure. The best-known evidence against them is practitioner evidence rather than a study: Google's then head of people operations said in 2013 that brainteasers predicted nothing and the company had stopped using them, based on internal analysis it did not publish. Treat that as a strong signal, not as settled research.

Should we use a take-home instead?

A take-home shows the output; these questions show the reasoning and the abandoned paths. They answer different questions, and the reasoning is usually the part that transfers.

Interview questions for other roles

Interview questions by role · STAR method interview questions · Build an interview scorecard