Interview Questions to Ask an Analyst
These are interview questions to ask an analyst candidate, written for the hiring manager, whether the title says business, data or financial analyst. The first three groups suit any analyst and follow the order of the work: how they frame a vague question, how they do the analysis and check their own numbers, and how they explain findings and handle pushback. Three shorter groups then cover each kind of analyst in turn, and the note under each question says how a strong answer and a worrying one tend to sound.
Want questions from the whole vault instead? Try the random question generator.
The questions
Each question, and why to ask it
Framing
Walk me through an analysis you are proud of, from the question you were asked to what happened because of it.
Why ask it
The standard opener for any analyst role, and the last part is the test. Strong candidates end on a decision someone made or a number that moved, while weaker ones stop at the dashboard or the deck. Notice which stage they spend longest on, since it is usually the one they like doing.
A manager asks you, 'Why are sales down?' What do you ask before you open any data?
Why ask it
Good answers narrow the question: down against what, since when, in which product or region, and what will be decided once the answer is known. A candidate who starts naming charts has skipped the step that saves the most time. Swap in a vague request your own team received and play the manager.
From what you know of our business, which three or four numbers would you want in front of you every week, and why those?
Why ask it
It shows whether they prepared, and whether they can tie a measure to how the business earns its money. A thoughtful candidate picks few, says what each one would prompt someone to do, and asks how you define them here. A long recital of every metric they can name tells you about memory, so ask which single one they would keep.
Tell me about a request where what the person asked for was not what they needed. How did you find out?
Why ask it
You are listening for the moment they asked what the number was for. The best stories include a short conversation that replaced days of work on the wrong thing. Blaming the requester for being unclear is the worrying version, because unclear requests are most of the job.
How do you decide when an analysis is finished?
Why ask it
Analysts can lose days polishing work that was good enough on Tuesday. A sound answer ties the stopping point to the decision: more digging would not change what the person does. 'When the deadline arrives' is honest, but it suggests nobody has taught them to set the scope at the start.
You have two days for a question that deserves two weeks. What do you cut, and what do you tell the person who asked?
Why ask it
The strong answer keeps the part closest to the decision and says plainly what was left out and how sure the result is. Ask what they would do with a third day. The answer to worry about is a quiet shortcut handed over as if it were the full job.
A team wants to know whether a change they made worked. How do you decide what to measure?
Why ask it
Look for someone who starts from what the change was meant to do, settles on one measure before looking at results, and names something to compare against. Picking whichever figure happened to rise is how weak analysis gets made, so ask what they would do if the chosen measure did not move.
Tell me about a piece of analysis nobody used. What would you do differently at the start?
Why ask it
Most working analysts have one of these. The useful answers trace the failure to the beginning: no decision was attached, the wrong person asked, or the answer came after the meeting it was for. If they say it has never happened, ask which of their regular reports they suspect goes unread.
Several people send you requests in the same week. How do you decide the order?
Why ask it
A good answer asks each person what the work is for and when it is needed, and takes a real clash to the requesters or a manager instead of settling it silently. First come, first served is acceptable from a junior. From a senior hire, expect some willingness to question a request with no decision behind it.
Method and checks
Which tools do you reach for first, and what is the hardest piece of work you have built in each?
Why ask it
A list of names tells you little, since Excel, SQL, Python and a BI tool turn up on a great many analyst resumes. What they built is where the depth shows: a model with scenarios, a query across several tables, a script that replaced a manual job. Ask what they would still have to look up, and treat a straight answer as a point in their favor.
Describe the messiest data set you have worked with. What was wrong with it, and what did you do?
Why ask it
People who have done the work name particulars: duplicate records, dates in three formats, a field whose meaning changed halfway through the year. Listen also for what they chose to leave unfixed and where they wrote that down. A story with no detail usually belongs to a colleague.
You find a handful of values far outside the rest. What do you do with them?
Why ask it
The careful analyst finds out what they are before deciding: a typing error, a test account, or one very large customer who is entirely real. Where it matters they show the result with and without them and say which version they would stand behind. Ask for the last odd value they chased and what it turned out to be.
How do you check your own numbers before anyone else sees them?
Why ask it
You want a routine, described without effort: a total tied back to a source the business already trusts, a rough estimate made before running anything, a row count after each step. 'I double-check everything' is not a method, so ask what the routine last caught.
Tell me about a mistake in your work that reached someone else. How was it found, and what did you do?
Why ask it
The answer you hope for arrives quickly and without excuses: what was wrong, who they told and how soon, and which check exists now because of it. Years in the job and no example usually means they have not been looking or would prefer not to say. Either way, ask once more.
A result looks too good to be true. What do you check first?
Why ask it
Experienced analysts suspect their own work before they celebrate. They mention a join that doubled the rows, a filter that dropped a whole group, or this year set against the wrong period. If the first instinct is to send it round, ask what would have to be true for the figure to be wrong.
Finance reports one revenue figure and the sales dashboard shows another. How do you work out which is right?
Why ask it
Often neither is wrong: the two define revenue differently, or count it on different dates. Strong candidates compare definitions, date ranges and filters before they touch the data, and they finish by getting one definition agreed and written down. Ask who they would go to for that agreement.
A number is up 4 percent on last month. How do you decide whether that means anything?
Why ask it
This is not a request for a formula. Listen for how much the figure normally moves from month to month, the same month a year earlier, and how many cases sit behind the percentage. A junior candidate may not name a statistical test, but should know that small counts jump around on their own.
Give me an example of two things that move together where one does not cause the other. How would you tell the difference in your own work?
Why ask it
Anyone can recite that correlation is not causation, so the second half does the sorting. A candidate who has had to do it will look for a third factor behind both, check which came first, find a group to compare against, or admit that the data on hand cannot settle it.
How do you document your work so someone else could repeat it?
Why ask it
Saved queries with comments, sources and definitions listed, assumptions on the first tab, old versions kept: any of these shows the habit. Then ask whether anyone ever did pick the work up, and what that person still had to come and ask.
Which recurring report or task have you rebuilt so it takes less of your time?
Why ask it
This shows whether they treat repeat work as something to fix or something to endure. Follow with how they knew the new version matched the old one. The careful ones ran both side by side for a cycle before switching off the manual copy.
How do you use AI tools in your analysis, and how do you check what they give you?
Why ask it
What may be pasted into these tools differs from one employer to the next, so say what your own policy is before they answer. A sensible reply treats the output as a draft: code is tested on a case with a known answer, and figures are traced to a source. Ask for one time the tool was confidently wrong and how they noticed.
How do you get up to speed on a business or a data set that is new to you?
Why ask it
Reading the definitions and talking to the people who create the data are both good signs. The best one is reproducing a figure the business already knows before producing a new one. Ask how long that took in their last job, and tell them what the first month would look like in yours.
Communicating
Pick a technical idea from your work, such as a margin of error or a discount rate, and explain it as you would to someone with no background in it.
Why ask it
Have them do it live, with you as the audience. A good explanation opens with an everyday comparison, leaves the jargon out of the first sentence and checks that you followed. Which idea they pick is informative too, because it shows what they think of as hard.
Tell me about a recommendation you made from your own analysis. Who had to be persuaded, and what did it take?
Why ask it
This separates an analyst who reports from one who advises. The stories worth hearing have a named person with a reason to doubt, the objection they raised, and what moved them: a second cut, a small trial, or their own figures run through the same method. If the recommendation was turned down, ask what became of the analysis afterwards.
You have five minutes with a senior leader. How do you present a piece of work that took three weeks?
Why ask it
The answer you want leads with the conclusion and what to do about it, gives the one or two figures behind it, and holds the method back unless someone asks. Walking through the steps in the order they were done is the usual weak answer, and a busy listener stops following long before the conclusion arrives.
How do you choose between a table, a chart and a single sentence?
Why ask it
It should depend on what the reader has to do: look up an exact value, see a trend or a comparison, or simply be told. Ask about a chart they have seen mislead people and how they would redraw it. No view at all on this is a concern in a job that ends in slides.
Tell me about a time your findings were not what the person who asked wanted to hear.
Why ask it
Strong candidates raised it early and in private, before the meeting where it would have been a surprise, and held the number firm while staying open about what it meant. Ask what the person did with the finding in the end. A story in which the bad news was softened until it said nothing is the one to probe.
A senior stakeholder says your number is wrong and theirs is right. What do you do?
Why ask it
Curiosity should come first: where their figure came from, how each side defined the thing, whether the analyst is the one in error. If their own number holds up, they then say so with the evidence in hand. Folding at once and digging in at once are the two ways to fail this.
Have you been asked to make numbers say something they did not? What did you do?
Why ask it
A good answer offers what could honestly be said, such as a different cut clearly labeled or the favorable figure with its caveat attached, and refuses to drop the inconvenient rows. If they have never been asked, make it concrete: their own manager wants the weak quarter left off the chart.
How do you tell people how sure you are of a result?
Why ask it
Ranges, plain words such as 'likely' tied to something concrete, and a line on what would change the answer all count. Every figure quoted to two decimal places with no caveat is a warning, and so is a page of caveats nobody could act on. A good follow-up: the range gets dropped and only the middle figure is repeated around the building, so what happens next?
What do you do when someone keeps asking for one more cut of the data?
Why ask it
Repeated requests usually mean the person has not decided or does not trust the answer. A candidate who sees that asks what would be done differently depending on the result, and offers to agree the last cut before running it. Ask how they put that to someone senior without sounding as if they were refusing the work.
Business analyst
How do you get requirements out of people who are not sure what they want?
Why ask it
Look for more than one technique: watching people do the work, asking about the last time the problem bit, playing their words back, putting a rough mock-up in front of them to react to. Sending round a template to fill in is the weak answer, because it returns what people think they should say.
Two departments want opposite things from the same project. How do you handle it?
Why ask it
An analyst's part is to make the conflict visible, not to pick a winner quietly. Strong answers write both needs down, dig for what each side is really after, and take what remains to whoever owns the decision. Ask about a time the two sides turned out to want the same thing.
Walk me through how you would map a process you had never seen before.
Why ask it
The best answers start with the people who do the work, not the manual. They follow one real case from start to finish, note where practice departs from the written procedure, and take the finished map back to those people to confirm. Ask which notation they use, but do not weigh it heavily.
How do you write a requirement so that a developer and a business user read it the same way?
Why ask it
You want to hear that it can be tested: a condition, an expected result, plain words and a worked example. Then ask for one of theirs that was misread and what they changed afterwards. If they brought a writing sample, read a single requirement aloud and say what you would build from it.
A stakeholder asks for a change after the requirements were agreed. What happens next?
Why ask it
Neither an automatic yes nor an automatic no. A sound answer works out what the change touches in time, cost and other requirements, records it, and takes it to whoever decides. How formal that step is varies between teams, so describe yours and see whether they adjust.
What was your part the last time users tested something before it went live, and what did the testing turn up?
Why ask it
How much of user testing falls to the analyst differs from team to team, so say how it works on yours. Where they have done it, expect test cases written from the requirements, real users at the keyboard instead of the project team, and a clear rule for what could wait until after launch. A round of testing that found nothing at all deserves a question of its own.
After something you specified went live, how did you find out whether it solved the problem?
Why ask it
Many analysts move to the next project on launch day. The ones worth hiring went back to the users, compared a measure from before and after, or discovered the workaround people had invented in the meantime. Ask what they found and whether anything was changed because of it.
Data analyst
What is the difference between an inner join and a left join, and when would choosing the wrong one have changed your result?
Why ask it
The definition is only a screen. Someone who writes SQL every week can describe rows vanishing without an error, such as customers with no orders dropping out of a count, or rows doubling when the match was not one to one. If your team works in another tool, ask the same about combining two tables there.
A dashboard number dropped overnight. Walk me through how you would investigate.
Why ask it
Good candidates check the plumbing before the business: whether the data loaded, whether a definition or a tracking tag changed, whether the fall is in one segment or all of them. Only then do they look for a real cause. Reaching straight for a business story is the common miss.
How would you set up a test to find out whether a new version of a page performs better than the old one?
Why ask it
Listen for visitors split at random, a single measure named up front, a run length fixed in advance and no stopping early because the result looks good. If the role involves no experiments, ask how they would make a fair before-and-after comparison.
What makes people keep opening a dashboard after the first week?
Why ask it
Answers worth hearing are built around the few questions a reader has every week, with the definitions printed on the page and not much else. Ask how they know whether anyone opens theirs, and whether they have ever retired one. Pride in the number of charts is a bad sign.
Tell me about a query or script that was too slow or too large to run. What did you do?
Why ask it
Here you learn whether they understand what sits under the tool. Filtering earlier, working on a sample, aggregating before joining and asking an engineer for help are all reasonable. Skip it for a junior role where the data fits in a spreadsheet.
Before you rely on a table you did not build, how do you check that it holds what its name says?
Why ask it
The habits to hear are reading a few raw rows, counting them, checking the date range, finding out who maintains the table and comparing a total with a figure the business already reports. People who have been burned answer with a story, and the story is the better evidence.
Describe a time you found something in the data that nobody had asked about. What did you do with it?
Why ask it
A test of curiosity, and of judgment about what to do with it. The good version checks the finding is real, sizes it, and takes it to someone who can act. Ask how they fit that kind of exploring around the requests already waiting.
Financial analyst
If depreciation rises by 10, what happens on each of the three financial statements?
Why ask it
The classic screen for whether a candidate sees the statements as one system. A sound answer has profit fall, the charge added back on the cash flow statement because no cash left the business, and the balance sheet still balancing through lower assets and lower retained earnings. Give them a tax rate and the fuller answer has cash ending slightly higher by the tax saved, and count it in their favor if they ask for the rate first.
Tell me about a forecast you built. What were the main assumptions, and how close did it come?
Why ask it
Expect two or three named drivers, such as volume, price or headcount, and a clear account of where each assumption came from. They should also know the size of the miss and the reason for it. Not knowing how the forecast turned out means nobody went back to look.
Actual costs came in well over budget in one department. How would you explain the variance?
Why ask it
A good answer splits the gap before explaining it: price against volume, timing against a permanent change, a one-off against something that will recur. It also includes a conversation with the department before the commentary is written. A paragraph that only restates the two figures is not an explanation.
How do you build a model so that someone else can follow and audit it?
Why ask it
A model built to be read keeps inputs apart from calculations, has no typed-in numbers hidden inside formulas, carries checks that flag when something stops balancing, and has a tab of notes. Then turn it around and ask how they would hunt for an error in a model they inherited. That second answer is harder to prepare.
How would you decide whether a project or a large purchase is worth the money?
Why ask it
Listen for cash laid out over time, a rate that reflects what the money costs, and a test of what has to be true for the spending to pay back. Payback alone is fine from a junior candidate, while a senior one should also raise what the calculation leaves out. For a valuation or investment seat, ask them to walk through a discounted cash flow instead.
Which assumptions in a model do you test hardest, and how?
Why ask it
The right answer is the ones the result swings on, found by flexing each in turn, with the outcome shown as a range or a few scenarios. A single figure presented with no sensitivity around it suggests the model has never been questioned by anyone senior.
How would you explain to a department head why profit and cash are telling different stories?
Why ask it
It tests the accounting and the plain speaking together. Good explanations use timing the listener already knows: customers who have not paid yet, stock bought ahead of sales, equipment paid for up front. If the answer leans on terms like accruals and working capital with no example, ask for one from a business like yours.
How to interview an analyst
Practical guidance for the conversation itself
Before you meet the candidate
Decide which analyst you are hiring
The title covers very different jobs. Write down what this person will produce in the first three months: requirements and process maps, queries and dashboards, or models and forecasts. Take most of your questions from Framing, Method and checks and Communicating, which apply to any analyst, then add the group that matches. Plenty of roles sit across two, such as a financial analyst who writes SQL, so borrow from both.
Pick eight to ten
Eight to ten is what fits in an hour if each one gets a proper follow-up. A workable spread is two on framing, three on method and checking, two on communicating and two or three from the specialty. Keep that core identical for everyone you see, as an analyst would keep the measure fixed across the groups being compared.
Bring a real question from your own team
Your own material makes the scenarios sharper: a vague request that landed last quarter, a figure that two teams argued over, a report nobody trusts. Strip out anything confidential and change the numbers. You already know how the story ended, which makes a good answer easy to recognize.
Set the level before you start
A candidate for a first job can answer from coursework, an internship or a personal project, and should be judged on reasoning. An experienced hire should bring examples with consequences: a decision made, money saved, an error caught. Decide beforehand what would count as a good answer at the level you are hiring, or the most confident talker will set the bar.
Consider a short exercise
Twenty or thirty minutes with a small made-up data set or a spreadsheet containing a planted error shows things conversation cannot. Give every candidate the same task, tell them in advance that it is coming, and say which tools they may use. If you are thinking of a longer take-home task, ask your HR team how those are handled where you are before you send one.
During the interview
Turn 'how do you' into 'what did you'
'How do you check your work?' gets a principle. 'What did you check on the last thing you sent out?' gets a description of an actual afternoon. Most of the 'how do you' questions on this page can be put the second way, and one 'what happened next?' afterwards usually brings out the part that was left off.
Find out which part was theirs
Analysis is usually shared work, and 'we' can hide a lot. When a story sounds impressive, ask who wrote the query, who built the model and who stood up and presented it. A candidate who did the work answers without having to think. One who watched it being done becomes vague at the second question.
Answer their questions on the scenarios
On 'Why are sales down?' and the other scenario questions, a candidate who questions you back is doing the job as it should be done. Play the stakeholder and answer as one would, including the unhelpful answers. How they cope with 'I am not sure, just find out' tells you a good deal about how they would cope with your colleagues.
Keep the quiz short
A definition, whether of a join or a discount rate, can be revised the night before. One or two make a fair screen, and the join question here carries a second half that asks for experience, while the depreciation one works best followed by 'where has that caught you out?' After that, spend the time on examples and scenarios, where preparation helps less.
Allow thinking time
Someone who pauses before a scenario question, or asks for a moment and a sheet of paper, is showing the habit you are hiring for. Say at the start that pauses are welcome. Fast, fluent answers are pleasant to listen to and are not the same thing as correct ones.
Judging what you heard
What strong answers have in common
They are specific: a named decision, an actual figure, the check that caught the problem, the thing that went wrong. They end with what happened after the work was handed over. And they include some doubt about the candidate's own results, which in an analyst is a working habit and not a lack of confidence.
Warning signs
A list of tools with nothing hard built in any of them. No mistake they can remember. Stakeholders described as the obstacle in every story. Results stated with total certainty. An explanation meant for a beginner that only another analyst could follow. One of these is worth a follow-up question, and several together are a pattern.
Weigh thinking above tools
A capable person can usually pick up a new tool on the job. Asking what a number is for, and distrusting a result that looks too neat, are habits that take much longer to form. The exception is a role that needs someone productive in one particular tool from the first week, in which case test that tool directly with the exercise.
Score before you compare notes
Write a mark and a line of evidence against each question as soon as the interview ends, and before you talk to the other interviewers. It is the same discipline you are asking of the candidate: decide what you are measuring first, then record what you saw, then discuss.
What to leave alone
Personal questions
Age, family plans, health, religion and where someone comes from have nothing to do with whether they can analyze. The rules on what may be asked are not the same everywhere, so find out from your HR team, or whoever handles hiring for you, where the lines fall for this role. If the real concern is long hours at quarter end, describe the hours and ask whether that works for them.
Brain teasers
Estimating the number of piano tuners in a city mostly measures practice at that kind of puzzle. A vague question from your own team tests the same skills, breaking a problem down and stating assumptions, on work the person would really be doing, and it is fairer to candidates who have never met the genre.
Their current employer's numbers
Do not press for real figures, client names or anything else that belongs to the place they work now. A candidate who declines is showing the discretion you will want once they have access to your own data. Invite them to change the numbers and keep the shape of the story.
Live work for free
An exercise is there to show how someone thinks, and should not double as a way to get a real problem solved. Use invented or old data, keep it short, and say what will happen to what they produce. Candidates talk to each other, and a test that looks like unpaid work can cost you the good ones.