This guide is for recruiters and hiring managers who want a problem-solving skills test that predicts how someone handles real work, not how fast they solve brain teasers. Puzzles reward people who've seen them before. Job-like cases show how a candidate frames a messy problem, uses evidence and makes a call. Below are the skills to measure by level, 10 original exercises with model answers, a comparison of formats and a scoring rubric.
What a problem-solving skills test should measure
Listen to the process more than the final answer. Two candidates can reach the same answer, but only one may have checked the data first, and that habit is what you're hiring.
Sample problem-solving test questions
Each one is a short case to talk through in three to six minutes. For prompts about how someone would react in a workplace situation, see our situational interview questions.
1. Junior, estimation: roughly how many support tickets a day should a software company with 500,000 customers expect? Talk through it.
What good looks like: the reasoning, not the number. A candidate might assume 2% of customers contact support each month, which is 10,000 tickets a month, or about 450 per business day across 22 working days. Score stated assumptions, simple math done correctly and a note on what would change the estimate, such as a recent release or the season.
2. Junior, data check: today's report total is exactly 90 more than yesterday's, but nobody added any records. Where do you look?
What good looks like: a difference that divides evenly by 9 often points to two swapped digits, since swapping digits always changes a number by a multiple of 9 (120 entered as 210 is 90 more). Other causes to check: a duplicated row worth 90, a changed filter or date range, or an edited formula. A strong candidate compares the two reports row by row instead of guessing.
3. Junior to mid, diagnosis: online orders in one region dropped 15% this week, and other regions are flat. What do you check, and in what order?
What good looks like: first, whether the data is right: tracking, reporting delays or a broken feed. Then segment the drop by channel, device and product to find where it lives. Then look at what changed: prices, a site release, payments, shipping or a local holiday. Good candidates order checks by how likely and how cheap they are.
4. Mid, prioritizing: you have time to fix two of these today: checkout fails for 2% of users, a typo on the homepage, slow exports for one large customer, password reset emails arriving 10 minutes late, and wrong colors in dark mode. Which two?
What good looks like: checkout failures come first, because they cost revenue on every affected visit. The second pick needs a reason: slow exports if that customer's renewal is at risk, or the reset emails if users are locked out and support is flooded. The best answers ask one or two questions before deciding and say how they'll tell the owners of the other three.
5. Mid, process: your team must cut a 10-day approval process to 5 days with no new hires. How do you start?
What good looks like: map every step and measure working time versus waiting time, since most of the delay usually sits in queues and handoffs. Find the bottleneck, remove or combine approvals that add little, and pilot the change on a small batch before rolling it out. Promising a result before measuring is a red flag.
6. Mid, ambiguity: your manager says "customers are unhappy" and nothing else. What are your first three steps?
What good looks like: define it (which customers, unhappy about what, compared with when), find the evidence (complaints, cancellations, survey scores, support themes), and talk to a few customers directly. Then come back with a short list of likely causes and how to test each one.
7. Mid, iteration: a fix you shipped worked in testing, but the problem came back a week later. How do you approach it now?
What good looks like: treat the first fix as a hypothesis that turned out wrong or incomplete. Find what's different between testing and the real conditions, reproduce the failure, and dig until the cause explains every symptom. Strong candidates add a check that would catch the problem early next time.
8. Senior, trade-off: you can ship a workaround in two days that adds four hours of manual work every month, or a proper fix in three weeks. How do you decide?
What good looks like: puts numbers on both: four hours a month is about 48 hours a year, plus the risk of manual errors. Weighs what else those three weeks would build and how easy each choice is to undo. Often the best answer is both: the workaround now, with the proper fix scheduled and owned.
9. Senior, evidence: a new onboarding flow shows 20% higher completion, but it was only shown to mobile sign-ups in December. What can you conclude?
What good looks like: very little about the flow itself. Device and season are mixed in with the change, so the lift could come from either. The candidate should suggest a randomized test across devices and a longer period, and say what result would change their mind.
10. Senior, cross-team: churn is up, and three teams each blame another. How do you find the real cause?
What good looks like: gets everyone to agree on one data set first, then looks at churn by customer group and start date to see when and where it rose. Lines up the timeline with each team's changes and tests one hypothesis at a time. Keeps the conversation on evidence, not blame, and names an owner for each test.
Test formats compared
Spoken cases like these work well early in the process. For analyst roles, add a short work sample test with a real data set later on.
Scoring rubric
Running a problem-solving test with NTRVSTA
Problem-solving is part of NTRVSTA's soft-skills exam library. The live AI interviewer adapts the difficulty to each answer: a candidate who handles the regional drop case well gets a harder twist, while one who stalls gets a hint, which shows how they take guidance. It follows up on reasoning, asking what they assumed or what they'd check next, and uses the STAR method for questions about past problems. You can tailor the questions to the role.
Reviewers get a score breakdown, video snippets of the moments where reasoning shows and a shareable report. For technical roles, pair this with a coding or design round. To compare standalone assessment tools, read our guide to pre-employment assessment software. Related tests: communication, leadership and DevOps. Hiring analysts? Start with our business analyst job description template.
Questions
How long should a problem-solving test be?
Plan 20 to 40 minutes for three to five cases, with time for follow-ups. One case explored in depth tells you more than five rushed ones.
Are brain teasers a good problem-solving test?
Not usually. They reward people who have seen the puzzle before and say little about how someone handles messy, real work. Cases drawn from the job, like a drop in sales or a failed fix, give you more useful signal.
Should candidates get to use a calculator or notes?
Yes for anything with numbers, since the goal is reasoning, not mental math. Tell every candidate the same rule before the test starts.
See how an adaptive interviewer follows a candidate's reasoning: book a 20-minute demo, or run a sample interview on yourself first.
