Data Analyst Interview Questions to Ask Candidates
Data analyst interview questions to ask candidates: 33 employer-side questions with what a strong answer sounds like, a 20-minute screen, and a scorecard.
Data Analyst Interview Questions to Ask Candidates
Six question sets written for the person doing the hiring: 33 questions with the reason each one is worth asking and what a strong answer sounds like, plus a 20-minute screen with an answer key and an eight-area scorecard. Download as DOCX.
The first time I interviewed a data analyst, I asked about SQL for ten minutes and had no idea whether the answers were good. I could tell the candidate was confident. I could not tell whether they were right. That is the position most owners are in, and it is the reason so many analyst hires come down to who interviewed most smoothly.
The fix is not to learn enough SQL to grade a query by Thursday. It is to ask questions whose answers you can judge without the vocabulary, and to run one short exercise where you score the questions the candidate asks rather than the answer they produce. At FirstHR we build for companies hiring without an HR department, and this page is written for the person on the hiring side of the table.
Below are 33 questions in six downloadable sets, each with the reason it is worth asking and what a strong answer sounds like, plus a 20-minute live screen with an answer key and an eight-area scorecard. The matching data analyst job description covers the posting side.
TL;DR
Interview a data analyst on five things: request intake, SQL and spreadsheet fluency, messy-data judgment, presentation, and behavioral evidence. Grade the explanation rather than the syntax, run one 20-minute live screen on a real export with a planted flaw, and score eight areas from 1 to 5 with written evidence before anyone discusses. Download 33 questions and the scorecard as DOCX.
The Four Things This Interview Must Settle
A data analyst interview has to settle four things: whether the candidate can turn a vague request into an answerable question, whether they can get a correct number out of the systems you actually run, whether anyone will understand the result, and whether they tell you when they were wrong. Technical depth matters, but it is the fourth of four.
That ranking surprises people who expect a coding screen. It holds because the common failure of a first analyst hire is not incompetence. It is an analyst who is busy every week producing work nobody uses, because the request was never clarified and the output was never readable.
The other reason to rank it this way is practical. You can grade clarification, presentation, and candor yourself. Grading a query, you cannot, unless you write queries. Weighting the interview toward what you can actually judge produces a better decision than pretending otherwise.
Analyst, Scientist, or Engineer?
Decide which of the three roles you are hiring before you write the questions, because they fail in different ways and the interviews barely overlap. An analyst answers what happened and why. A scientist models and predicts. An engineer builds and maintains the plumbing that both rely on.
Most small businesses need the analyst, because the immediate problem is that nobody trusts the numbers rather than that nobody has built a model. If your problem is that the data never arrives anywhere reliably, that is an engineering hire, and the data engineer questions fit better than these. If it is forecasting or experimentation, use the data scientist set instead.
What the hire does
Data analyst
Data scientist
Data engineer
Answers business questions from existing data
Owns recurring reports and dashboards
Reconciles systems that disagree
Builds statistical models and experiments
Builds and maintains pipelines
Usually the right first data hire
Titles are unreliable here, so interview for the work rather than the word on the resume. A candidate whose last job was titled analyst may have spent it building pipelines, and one titled scientist may have spent it in dashboards. Ask what they did on a normal Tuesday.
The Spreadsheet Reality Nobody Interviews For
Your data almost certainly lives in exports, a billing tool, a CRM, and a shared drive of spreadsheets, and no two of them agree on how many customers you have. That is the environment the hire is walking into, and it is the single biggest source of mismatch when a candidate arrives from a larger company.
Analysts trained inside a mature data team often worked downstream of infrastructure somebody else maintained: clean tables, a defined metric layer, a documented catalog. Drop that person into forty spreadsheets and three definitions of customer, and the skill is real but the reflexes are wrong. This is a fit question, not a competence question, and it is much cheaper to ask in the interview than to discover in month two.
Ask this
Why it separates candidates
What does your first month look like with no warehouse?
Strong answers inventory sources and owners; weak ones open with a migration
What is the most complicated thing you built in a spreadsheet?
Reveals whether spreadsheets are a tool or an insult to them
Which of our tools have you pulled data out of before?
Names the real integration work rather than a tidy table they inherited
When should a report stop living in a spreadsheet?
Tests judgment about triggers, not a preference for better tooling
Who decided what a customer meant at your last job?
Shows whether they have ever owned a definition or only applied one
None of this argues for hiring someone unfamiliar with modern tooling. It argues for asking directly how they work when the tooling is not there yet, and for weighting that answer heavily when yours is not.
The Six Sets and What Each One Catches
The questions are grouped into five sets by what they test, plus a sixth containing the scorecard, the red-flag checklist, and the live screen. A strong candidate performs across all five, not only on the behavioral questions that every candidate has rehearsed.
Request Intake and Priorities
7 questions
Turning a hallway question into an answerable one, triaging a queue with no manager to shield it, and offering a rough number now instead of a perfect one too late.
SQL, Spreadsheets, Your Stack
7 questions
Fluency judged in plain English, honest spreadsheet depth, and what their first month looks like when there is no warehouse and forty exports.
Messy Data and Reconciliation
6 questions
Two systems that disagree, duplicates, blank fields, and the moment to stop cleaning. The judgment that decides whether your reports get trusted.
Charts and Reporting People Use
6 questions
Explaining a result to you in two minutes, choosing a chart on purpose, killing a dashboard nobody opens, and handing a recurring report back to the team.
Judgment and Behavioral
7 questions
A decision that changed, a number that was wrong, correlation talked about carefully, and concrete habits around payroll and customer data.
Scorecard and Live Screen
Score, do not guess
An eight-area rubric with a required evidence line, an eight-item red-flag list, and a 20-minute screen with an answer key a non-technical interviewer can mark.
Ask Two From Every Set, Not Ten From One
The temptation is to spend the whole hour on the technical set, because it feels like the part that matters. Resist it. Two questions from each of the five sets, with real follow-ups, tells you far more than ten variations on one theme, and it keeps a candidate who rehearsed one area from clearing the whole interview on it. Use the same selection for every candidate so the evaluation form compares like with like.
33 Questions and a Scorecard to Download
Download all six sets as a single Word document, or copy them individually. Each set lists the questions, the reason each one is worth asking, what a strong answer sounds like, and space for notes. The sixth file holds the scorecard, the red flags, and the live screen with its answer key.
Download All 6 Question Sets and the Scorecard
Request intake, tools, messy data, reporting, judgment, plus a scorecard, red flags, and a 20-minute screen. All in one DOCX.
Set 1: Request Intake and Prioritization
Turning a hallway question into an answerable one, triaging a queue nobody else manages, and offering a rough number now rather than a perfect one too late. Ask this set first.
Request Intake and Prioritization Questions
REQUEST INTAKE AND PRIORITIZATION QUESTIONS
Candidate: __
Company: __
Interviewer: __
Date: _
WHY THIS SET COMES FIRST
At a company without a data team, the analyst is a service desk. Requests arrive
by message, in hallways, and in meetings, most of them vague, several of them
urgent, and a few of them not worth answering at all. The candidate who cannot
manage that queue will be busy every week and useful in none of them. Ask this
set before anything technical.
QUESTIONS
1. Someone asks you "why were sales down last month". What are the first three
questions you ask before you open anything?
Why ask it: separates people who answer requests from people who answer
questions. The literal request is almost never the real one.
Strong answer: asks which product or segment, over what comparison period,
and what decision the answer will drive. Confirms the definition of "sales"
before pulling a number.
2. You have five requests and time for two this week. How do you decide?
Why ask it: an analyst with no prioritization method becomes whoever shouted
loudest.
Strong answer: weighs the decision at stake and the deadline behind it, asks
requesters what changes if they get the answer, and tells the other three
when to expect a reply rather than going quiet.
3. A manager needs a number by tomorrow, but doing it properly takes three days.
What do you do?
Why ask it: this is the most common real situation in a small company.
Strong answer: offers a rough figure with stated caveats now and the exact
one later, and is explicit about which is which. A candidate who only offers
the three-day answer will be routed around.
4. How do you record what a request actually asked for, so nobody relitigates it
two weeks later?
5. Tell me about a request you pushed back on. What did you say?
Why ask it: tests whether they can decline work without damaging a
relationship. An analyst who never says no produces low-value output.
6. What do you do when the data you have cannot answer the question you were
asked?
Strong answer: says so plainly, explains what could be answered instead, and
names what would need to be collected. Does not quietly deliver a proxy and
let the requester assume it is the real thing.
7. How do you know whether last quarter’s work mattered?
Strong answer: points to a decision that changed, a process that was dropped,
or a report that people now run themselves. Points to volume of tickets
closed only if pushed.
NOTES
__
__
Set 2: SQL, Spreadsheets, and Your Actual Stack
Fluency judged in plain English, honest spreadsheet depth, and what their first month looks like when there is no warehouse and the numbers live in exports.
SQL, Spreadsheets, and Your Actual Stack
SQL, SPREADSHEET, AND STACK QUESTIONS
Candidate: __
Company: __
Interviewer: __
HOW TO USE THIS SET
You are not testing whether the candidate can pass a technical screen at a large
technology company. You are testing whether they can produce a correct number
from the systems you actually run, which for most small businesses means
exported spreadsheets, a billing tool, and one or two apps with a reporting tab.
Every question below has a note that lets a non-technical interviewer judge the
answer.
QUESTIONS
1. Walk me through the tools you used at your last job, from raw data to the
thing a manager read.
Why ask it: reveals whether they worked at the start of the chain or only at
the end of one somebody else built.
Strong answer: names each step and says which parts they built themselves.
2. We do not have a data warehouse. Our numbers live in a billing tool, a CRM,
and about forty spreadsheets. What does your first month look like?
Why ask it: the single best predictor of fit at a small company.
Strong answer: starts by listing sources and owners, picks the two or three
numbers the business runs on, and gets those right before proposing tooling.
A weak answer opens with a warehouse migration.
3. In plain English, what does a JOIN do, and what goes wrong when you get one
wrong?
Why ask it: you do not need SQL to grade this. You need to hear whether they
can explain a technical idea to you.
Strong answer: combines rows from two tables on a matching value; the classic
failure is a row multiplying, so a total comes out too high.
4. What is the difference between a GROUP BY total and a COUNT DISTINCT, and
when has that difference bitten you?
Strong answer: gives a concrete case, usually counting orders and thinking
they were customers.
5. How good are you in Excel or Google Sheets, honestly, and what is the most
complicated thing you have built in one?
Why ask it: candidates from large companies sometimes treat spreadsheets as
beneath them, and at your size that is a real problem.
Strong answer: comfortable and unembarrassed, names pivot tables, lookups,
and cleanup work, and knows the point at which a spreadsheet should stop.
6. When would you say a spreadsheet is no longer the right home for a report?
Strong answer: names the triggers, several people editing it, the same
numbers rebuilt by hand every month, or a file too slow to open, rather than
an abstract preference for a better tool.
7. Which reporting or dashboard tools have you set up from scratch, as opposed
to used?
WHAT TO LISTEN FOR
•Explains technical work in plain language without being asked twice
•Comfortable with spreadsheets and honest about their limits
•Has built something end to end, not just consumed a finished pipeline
•Names tools with specifics about what they did in them
NOTES
__
Still Using Spreadsheets for Onboarding?
Automate documents, training assignments, task management, and track onboarding progress in real time.
Two systems that disagree, duplicate records, blank fields, and knowing when to stop cleaning. The judgment that decides whether anyone ends up trusting your reports.
Messy Data and Reconciliation Questions
MESSY DATA AND RECONCILIATION QUESTIONS
Candidate: __
Company: __
Interviewer: __
WHY THIS SET MATTERS MORE THAN IT LOOKS
Most of the job is not analysis. It is deciding what a record means when two
systems disagree, when a customer appears three times under slightly different
names, and when a field has been empty since a form changed. Judgment here is
what makes your reports trustworthy, and it is invisible on a resume.
QUESTIONS
1. Our billing tool says we had 412 customers last month. Our CRM says 389. What
do you do?
Why ask it: the exact situation every small business is in.
Strong answer: does not pick a winner. Finds out what each system counts,
when each one records a customer, and whether the gap is a definition problem
or a data problem. Then writes the definition down.
2. How do you define an active customer, and who decides?
Strong answer: the business decides, the analyst writes it down and applies
it consistently everywhere. A candidate who defines it alone will produce
reports nobody agrees with.
3. A field is blank in 30 percent of rows. Walk me through your options.
Strong answer: finds out why it is blank before deciding anything, because a
field blank by design and a field blank by breakage need opposite treatment.
4. How do you find and handle duplicate records?
5. How do you know when you have cleaned enough and should start analyzing?
Why ask it: perfectionism is a real failure mode in this role.
Strong answer: cleans to the precision the decision needs, states what is
still imperfect, and moves on.
6. You deliver a number. Three weeks later you realize it was wrong. What
happens next?
Why ask it: this will happen, and how they handle it is the whole answer.
Strong answer: tells the people who used it, immediately, explains what
changed and why, and adds a check so the same error is caught next time. Any
answer that involves quietly correcting the file is a red flag.
WHAT TO LISTEN FOR
•Investigates causes rather than choosing the more convenient number
•Writes definitions down instead of holding them in their head
•Proportionate: knows when good enough is good enough
•Owns errors out loud
NOTES
__
Set 4: Charts, Dashboards, and Reporting People Use
Explaining a result to you in two minutes, choosing a chart on purpose, retiring a dashboard nobody opens, and handing a recurring report back to the team.
Charts, Dashboards, and Reporting People Use
CHARTS, DASHBOARDS, AND REPORTING QUESTIONS
Candidate: __
Company: __
Interviewer: __
WHEN TO USE THIS SET
An analyst whose work is never read is an expensive hobby. This set tests the
half of the job that has nothing to do with querying: choosing a presentation,
writing an interpretation a busy owner can act on, and killing reports that have
stopped earning their place.
QUESTIONS
1. Explain a result you are proud of to me as if I have never seen a
spreadsheet. You have two minutes.
Why ask it: the most predictive question on this page for a small company,
because you are the audience.
Strong answer: leads with the finding and what it means, holds the method for
the end, and stops. Jargon that survives a request for plain language is a
warning.
2. When do you use a line chart, a bar chart, and a table? What would you never
use a pie chart for?
Strong answer: connects the choice to the question, comparison over time,
comparison between categories, exact values to be read off. Confident
opinions here usually mean real reps.
3. Show me a dashboard you built. What is on it, and what did you deliberately
leave off?
Why ask it: what they excluded says more than what they included.
4. You built a dashboard and nobody opens it. What went wrong and what do you
do?
Strong answer: goes and asks the intended users, expects the cause to be a
mismatch with a real decision rather than a design flaw, and is willing to
delete it.
5. How do you present a result you know the recipient will not like?
Strong answer: leads with the number and the confidence in it, separates the
finding from the recommendation, and does not soften the figure itself.
6. How do you make a recurring report something the team can run without you?
Why ask it: at your size this is what buys back the analyst’s time.
Strong answer: documents the definitions, simplifies the output, and trains
one person, rather than treating the report as job security.
WHAT TO LISTEN FOR
•Leads with the finding, not the method
•Has opinions about presentation and can defend them
•Willing to delete their own work
•Builds so that other people can self-serve
NOTES
__
Companies Using FirstHR Onboard 3x Faster
Join hundreds of small businesses who transformed their new hire experience.
Set 5: Judgment, Skepticism, and Behavioral Evidence
A decision that changed because of their work, a number they got wrong, careful language about correlation, and concrete habits around payroll and customer data.
Judgment, Skepticism, and Behavioral Evidence
JUDGMENT AND BEHAVIORAL QUESTIONS
Candidate: __
Company: __
Interviewer: __
HOW TO SCORE THIS SET
Ask for real events, not opinions. Use the STAR pattern: Situation, Task,
Action, Result. Push every answer until you get a specific number, a named
consequence, or an admission that there was not one. Vagueness that survives two
follow-ups is itself the finding.
QUESTIONS
1. Tell me about a time your analysis changed what the business did.
Why ask it: the difference between an analyst and a report generator.
Strong answer: names the decision, the person who made it, and what happened
afterward, including if it did not work.
2. Tell me about a time you were wrong about a number. How did you find out?
Strong answer: has a real example and diagnoses the cause. A candidate with
an unbroken record has either not done much or is not telling you.
3. Two things move together in our data. How would you talk about that to
someone who wants to act on it?
Why ask it: tests statistical care without a statistics exam.
Strong answer: distinguishes what moved together from what caused what, names
an alternative explanation, and suggests what would be needed to tell.
4. You made a recommendation and nobody acted on it. What did you conclude?
Strong answer: considers that the recommendation may have been unusable or
badly timed, not only that the audience was wrong.
5. What would change your mind about a finding you currently believe?
Why ask it: a candidate who cannot answer this treats their output as fact.
6. This role sees payroll, customer, and performance data. How have you handled
information that should not be shared?
Strong answer: has a concrete habit, limited access, aggregated outputs,
asking before pulling sensitive fields. Not a general assurance of
discretion.
7. What is the most tedious part of analyst work, and how do you deal with it?
Why ask it: cheap honesty test. The tedious part is data cleaning, and every
experienced analyst knows it.
WHAT TO LISTEN FOR
•Specific events with named outcomes
•Owns errors and can diagnose them
•Careful with causal language without being paralyzed by it
•Concrete habits around confidential data
NOTES
__
Set 6: Scorecard, Red Flags, and the 20-Minute Screen
An eight-area rubric with a required evidence line per score, an eight-item red-flag checklist, and the live screen with an answer key written for a non-technical interviewer.
Scorecard, Red Flags, and the 20-Minute Screen
DATA ANALYST SCORECARD, RED FLAGS, AND SCREEN
Candidate: __
Interviewer: __
Date: _
PART 1: THE 20-MINUTE SCREEN AND ITS ANSWER KEY
Give every candidate the same short exercise, ideally an export of a real report
you already run with the sensitive columns removed. Twenty minutes, screen
shared, no take-home. You are scoring the questions they ask, not the answer.
Setup:
•One spreadsheet, roughly 200 rows, with at least one deliberate flaw:
duplicated rows, a date column stored as text, or two spellings of one
customer name.
•One request written in a single sentence, for example: "Tell me which of our
customers grew last quarter."
What to say: "Take twenty minutes. Talk out loud. You will not have enough
information, and that is part of the exercise."
Answer key for a non-technical interviewer. Award a point for each:
[ ] Asked what "grew" means before starting
[ ] Asked which quarter and what it is compared against
[ ] Checked the row count or scanned for duplicates before computing anything
[ ] Noticed at least one of the deliberate flaws
[ ] Said out loud what they were assuming
[ ] Gave an answer with a stated caveat rather than a bare number
[ ] Named what they would want next if they had more time
5 or more: strong. 3 to 4: borderline, probe further. 2 or fewer: pass.
PART 2: SCORING RUBRIC
Score 1 (poor) to 5 (excellent). Every score needs one line of written evidence
from the interview. Fill this in alone, before anyone says what they thought.
Presentation and plain language Score: [ 1 2 3 4 5 ]
Evidence: __
Skepticism and honesty about error Score: [ 1 2 3 4 5 ]
Evidence: __
Works without supervision Score: [ 1 2 3 4 5 ]
Evidence: __
Handling of confidential data Score: [ 1 2 3 4 5 ]
Evidence: __
Live screen result Score: [ 1 2 3 4 5 ]
Evidence: __
Total: ______ / 40
Recommendation: [ ] Strong yes [ ] Yes [ ] No [ ] Strong no
PART 3: RED FLAGS
[ ] Cannot explain any technical idea in plain language
[ ] Every past project was a clean success
[ ] Treats spreadsheets as beneath them
[ ] Answers "why were sales down" without asking a single question first
[ ] Proposes a tooling overhaul before understanding the business
[ ] Uses causal language for a correlation and does not flinch when challenged
[ ] Has never told anyone that a number they delivered was wrong
[ ] Cannot name a report or dashboard that should have been deleted
WEIGHTING WORKSHEET
Before the first interview, write down which two areas matter most for this
version of the role, and by how much. Deciding after the interviews is how a
likeable candidate wins on the wrong criteria.
Area weighted highest: __ Why: __
Area weighted second: __ Why: __
The 20-Minute Screen You Can Grade Without SQL
The most reliable signal in an analyst interview is watching someone work for twenty minutes on a messy file, and the trick that makes it gradable for a non-technical interviewer is to score the questions they ask rather than the answer they reach. There is no correct answer. There is a correct set of questions.
Take a report you already run, strip the sensitive columns, cut it to roughly 200 rows, and plant one flaw: duplicated rows, dates stored as text, or one customer spelled two ways. Then give a deliberately underspecified request in a single sentence, and tell the candidate out loud that they do not have enough information and that this is intentional.
Use a report you already run
Export something real, strip the sensitive columns, and leave roughly 200 rows. Plant one flaw: duplicate rows, dates stored as text, or a customer name spelled two ways.
Give a one-sentence request
Something deliberately underspecified, such as which customers grew last quarter. The missing information is the point of the exercise, not an oversight.
Mark the questions, not the answer
The answer key scores whether they defined the term, checked the data before computing, noticed the flaw, and stated their assumptions out loud.
Same file for every candidate
The exercise only compares people if the input is identical. Twenty minutes, live and screen shared, so you see the thinking rather than a polished deliverable.
Why a Live Screen Beats a Take-Home
Structured, job-related assessments are the practice the federal government's own hiring guidance points to for predicting performance, and the U.S. Office of Personnel Management recommends asking every candidate the same questions and rating answers against predetermined criteria. A live exercise satisfies that and a take-home does not: you cannot verify who did a take-home, how long it took, or what was searched, and experienced candidates running several processes often decline it outright.
Mark the sheet immediately afterward. Seven checks, five or more is strong, two or fewer is a pass, and the marked sheets are worth keeping because they are the evidence that every candidate got the same exercise.
The Follow-Ups That Decide It
The listed questions open the conversation; the follow-ups decide it. Push for the specific number, the actual consequence, the real example. An answer that stays general after two follow-ups is itself the finding, and it is worth writing down as such.
Signals worth weighting
Asks clarifying questions before touching data
Names a decision that changed because of their work
Comfortable in spreadsheets without apologizing for it
Depth checks
Explains a join or a distinct count in plain English
Has reconciled two systems that disagreed
Has built something end to end, not just used one
Communication checks
Leads with the finding, keeps method for the end
Has deleted a report that stopped earning its place
Can teach one person to run a recurring report
Red flags
Answers the literal question with no clarification
Opens with a warehouse migration before week two
Has never had to correct a number in public
One red flag deserves singling out: a candidate whose every project was a clean success. Real analyst work produces reports nobody read and numbers that turned out wrong. A candidate who can name one and explain how they found out is more trustworthy than one with an unbroken record.
The reverse also holds. Be careful not to over-reward polish. Analysts who present beautifully and reconcile carelessly are a recognizable type, which is why the reconciliation set and the live screen sit alongside the presentation questions rather than instead of them.
Setting the Pay Band Before the First Screen
Settle the band before the first conversation, because there is no single official figure for this title and a number quoted without context either loses the candidate or overpays by a wide margin. The federal survey has no occupation called data analyst, so benchmark against the two nearest classifications and adjust for your scope.
Benchmark Against Two Official Occupations
According to the Bureau of Labor Statistics Occupational Employment and Wage Statistics survey (May 2025), data scientists reported a national median annual wage of $120,230, with the 10th percentile at $67,240 and the 90th at $199,130. Management analysts, the classification that absorbs many analyst titles, reported a median of $101,860, a 10th percentile of $60,640, and a 90th of $171,640. A general data analyst usually prices below both medians.
Version of the role
Reference point
Rough band
First analyst, one function, junior
Lower quarter of both occupations
$60,000 to $78,000
Sole analyst owning company reporting
Between the 25th percentile and the medians
$78,000 to $105,000
Senior analyst, heavy technical scope
At or above the data scientist median
$105,000 to $145,000
Part-time or fractional arrangement
Hourly against the same bands
Scope-dependent
Two practical notes. A part-time arrangement is a legitimate first step when the workload does not yet justify a full-time hire. And several states and cities now require a good-faith range in the posting, so check the pay transparency rules that apply to you before you publish rather than after. This is general information, not compensation advice.
Fair, Legal, and Structured Interviewing
Fair, legal, and structured are three descriptions of the same practice. Asking every candidate the same job-related questions, scored against criteria written down in advance, keeps you within the rules, reduces bias, and produces better hires at the same time. The EEOC publishes a plain summary of the practices that are prohibited under federal law, and it takes ten minutes to read.
Ask about the work, not the person
Federal anti-discrimination law prohibits basing a hiring decision on protected characteristics, and a question that probes one creates risk even when it is asked as small talk. Keep off age, race, religion, national origin, sex, pregnancy or family plans, disability, and genetic information. Analyst interviews drift into schools and graduation years more than most, because the conversation naturally turns to training, and a graduation year is an age proxy. Ask what they studied and what they have built, never when they finished. Every question on this page is written to stay on the job. This is general information, not legal advice.
One question set, every candidate
Ask the same core questions of everyone and score them against the same rubric. Structured interviewing is the practice with the strongest evidence behind it for predicting on-the-job performance, and it is also the simplest defense against a decision that rested on rapport. For an analyst hire it does something extra: because most interviewers cannot grade the technical half themselves, consistency is the only way to compare two candidates who each sounded competent. Write the questions before the first screen, not between interviews. This is general information, not legal advice.
The live screen counts as a selection procedure
A work sample used to decide who advances is a selection procedure, so it needs to be job-related, applied the same way to every candidate, and scored on criteria written down in advance. Practically: same file, same time limit, same answer key, and no take-home for one candidate and a live exercise for another. Keep the marked sheets. If a screen ever screens out one group at a much higher rate, you want the criteria and the records to show the exercise measured the job. This is general information, not legal advice.
Say what data the role touches, before the offer
A data analyst sees payroll figures, customer records, and performance data, often in week one. Raise that during the interview rather than at onboarding: name the systems, say what access the role gets and what it does not, and ask directly how they have handled confidential information. It is a fair, job-related line of questioning, it filters candidates who are casual about it, and it means the confidentiality agreement on day one confirms something already discussed rather than introducing a surprise.
The analyst-specific trap is the drift into schools and graduation years, since a graduation year is an age proxy. Ask what someone studied and what they have built, never when they finished. For the fuller list, see our guide to illegal interview questions. Applicant tracking is coming soon to FirstHR. This is general information, not legal advice.
Interviewing a Data Analyst Without HR
At a company with a data team, this candidate meets three people who have done the job and a recruiter who coordinates the scorecards. Everywhere else one person runs the interview between their own deadlines, with nobody in the room who can grade half the answers. Structure is what closes that gap.
Nobody in the building can grade the technical answers
At a company with a data team, an analyst candidate meets three people who have done the job. Everywhere else, the owner or an operations lead interviews alone and has to judge answers about joins and dashboards on trust. The fix is not to learn SQL before Thursday. It is to ask questions whose answers a non-specialist can grade: explain this in plain English, tell me what you would ask before starting, walk me through where this went wrong. Every set on this page is written so the quality of an answer is visible without the vocabulary.
The first analyst inherits spreadsheets, not a data platform
Candidates from larger companies have often worked downstream of infrastructure somebody else maintained: clean tables, a modeling layer, a defined metric catalog. Your version of the job starts with exports that disagree and a definition of customer that three people hold differently. That is a fit question, not a skill question, and it is worth asking directly rather than discovering in month two. The intake and messy-data sets exist to surface it in the interview, while it is still cheap.
One analyst, no manager, and a queue that never empties
A solo analyst reports to someone with no time to prioritize their work, which means they have to prioritize it themselves and say no out loud. Interview for that on purpose. Ask how they chose between competing requests, what they declined, and how they told the requester. Then set them up to succeed: name the two or three numbers the business runs on, name who owns each definition, and agree what the first ninety days must produce, so the role does not dissolve into a ticket queue.
The practical version is short. Write the questions before the first screen, use the same set for everyone, run the live exercise on the same file, and score alone before comparing. That is the whole of a structured interview, and it is worth more here than any single clever question. More hiring kits for other roles sit in the hiring templates library.
From Offer to the First Ninety Days
Once you have chosen, the work shifts from evaluating to hiring well, and an analyst has two wrinkles most roles do not. They need confidential access early, and they need a target, because an analyst without one dissolves into a request queue by week three.
Put the scope in writing
Confirm pay, the reporting line, and which numbers this person owns end to end. A verbal promise about scope is where analyst roles quietly drift.
Sign confidentiality before access
Payroll and customer data show up in week one. Get the agreement signed and the system access approved in writing before the first query, not after the first project.
Agree the metric list on day one
Name the two or three numbers the business runs on and who owns each definition, so the analyst is not relitigating what a customer is in month three.
Set one deliverable for ninety days
One report the business will actually trust, with a named stakeholder. An analyst without a target produces charts; an analyst with one produces decisions.
Handle the access wrinkle first. The confidentiality agreement gets signed before any system access is granted, not after the first project, and the approvals for each system are worth writing down while they are being decided. Retrofitting that record later is unpleasant, and for an analyst it is the difference between an audit trail and a shrug.
Then handle the target. Name the one report the first quarter must produce and the person who will use it, alongside the standard new hire paperwork. Applicant tracking is coming soon to FirstHR.
FirstHR connects that people side in one place: the offer letter, the confidentiality agreement, e-signatures, the paperwork, and the access and policy checklist, with every signed document stored on the employee profile. FirstHR is an onboarding and HR platform, not a business intelligence, warehouse, or reporting tool, so connect those separately. Applicant tracking is coming soon to FirstHR.
Key Takeaways
Settle four things: can they clarify a request, get a correct number, be understood, and admit an error. Technical depth is the fourth of four.
Grade the explanation rather than the syntax; a non-technical interviewer can score plain English reliably and a whiteboard query not at all.
Run a 20-minute live screen on a real export with a planted flaw, and score the questions the candidate asks against a written answer key.
Ask directly what their first month looks like with no warehouse; the answer is the clearest fit signal a small business gets.
Weight reconciliation heavily, because your systems disagree and the analyst decides which number the business believes.
There is no federal wage series for data analyst; benchmark against data scientists at a $120,230 median and management analysts at $101,860.
Score eight areas from 1 to 5 with one line of evidence each, alone, before anyone in the room says which candidate they liked.
Frequently Asked Questions
What questions should I ask a data analyst candidate?
Ask across five areas: how they handle incoming requests, their fluency with SQL and spreadsheets, their judgment on messy data, how they present results, and behavioral evidence from real projects. The most useful single question at a small company is what they would ask before answering a vague request such as why sales were down, because it separates people who answer questions from people who answer tickets. Add a reconciliation question, since your systems will disagree with each other, and one that makes them explain a technical result to you in plain English. Finish with a behavioral question about a number they got wrong. This page contains 33 such questions, each with the reason it is worth asking and what a strong answer sounds like.
How do I test SQL skills if I do not know SQL myself?
You do not grade the syntax, you grade the explanation and the process. Two techniques work well without any technical background. First, ask the candidate to explain a concept in plain English, for example what a join does and what goes wrong when one is written badly; anyone who has really used SQL can do this in thirty seconds, and anyone who cannot has a communication problem that will hurt you anyway. Second, run a short live exercise on a spreadsheet you already have and score the questions they ask rather than the answer they reach. The scorecard set on this page includes a 20-minute screen with an answer key written for a non-technical interviewer.
What is the difference between a data analyst and a data scientist?
A data analyst answers questions about what happened and why, using SQL, spreadsheets, and reporting tools, and their output is usually a number, a report, or a recommendation. A data scientist leans further into statistical modeling, experimentation, and prediction, and their output is often a model or an experimental result. For most small businesses the analyst is the right first hire, because the immediate problem is that nobody trusts the numbers rather than that nobody has built a model. Pay reflects the difference, and so should the interview: an analyst interview should weight reconciliation, reporting, and communication, while a data scientist interview weights statistics and modeling judgment.
Should I give a data analyst candidate a take-home assignment?
A short live exercise beats a take-home for most small businesses. Take-homes have a poor completion rate among experienced candidates, who often have several processes running, and they are impossible to police, so you may be grading someone else’s work or an unlimited amount of time. A 20-minute exercise done live on a shared screen shows you the thinking, which is the part you are actually hiring: what they ask before they start, whether they check the data before computing, and whether they say their assumptions out loud. Use a real export with the sensitive columns removed, give every candidate the same file and the same time limit, and score against a written key.
What are the red flags in a data analyst interview?
Eight recur often enough to be worth watching for. A candidate who answers a vague question without asking anything first. One who cannot explain a technical idea in plain language. One whose every past project was a clean success, which usually means limited scope or limited candor. One who treats spreadsheets as beneath them, which is a fit problem at your size. One who proposes rebuilding your stack before understanding the business. One who uses causal language about a correlation and does not flinch when challenged. One who has never told anyone a delivered number was wrong. And one who cannot name a report that should have been deleted. The scorecard on this page includes this list as a checklist.
How much does a data analyst cost to hire?
There is no separate federal wage series titled data analyst, so benchmark against the two nearest official occupations and adjust. According to the Bureau of Labor Statistics Occupational Employment and Wage Statistics survey (May 2025), data scientists reported a national median annual wage of $120,230 and management analysts $101,860. A general data analyst usually prices below the data scientist median, with a first analyst hire at a small company often landing between the 10th and 25th percentiles of those bands depending on market and experience. Add benefits and overhead on top of salary before deciding the role is full time, and publish a good-faith range where pay transparency rules apply. This is general information, not compensation advice.
How long should a data analyst interview take?
Plan a 30-minute screen, a 60-minute structured interview, and a 20-minute live exercise, which is enough for most small businesses to decide. In the structured hour, cover two or three questions from each of the five question areas and leave ten minutes for the candidate’s own questions, which are themselves informative: strong analysts ask what decisions the role supports and who owns the definitions. Resist adding rounds. Depth on a few questions with real follow-ups tells you more than a longer checklist, and every extra round costs you candidates. Score immediately after each conversation while the answers are still fresh, and have each interviewer score alone before the group compares notes.
Are these data analyst interview questions legal to ask?
Yes. Every question here is about the work: how the candidate handles requests, what tools they have used, how they reconcile conflicting data, how they present results, and what they have done in past roles. The legal caution is the general one that applies to all interviewing. Avoid questions that touch protected characteristics such as age, race, religion, national origin, sex, pregnancy or family plans, disability, or genetic information, and be careful with the analyst-specific version of that trap, which is asking when someone graduated rather than what they studied. If you use the live exercise as a screening step, apply it the same way to every candidate and keep the scored sheets. This is general information, not legal advice.
