Data analyst interview questions that actually get asked in 2026
The four rounds a data analyst interview is really made of, what each one grades, and the answers that get past a SQL screen into an offer.
Jordan Reyes · · 8 min read
Four questions do most of the work in a data analyst interview: write a query against two tables, explain why a number moved, walk through a project you owned, and describe a result to somebody who will never open your notebook. Everything else is scheduling. This post covers what each round is grading, what a passing answer sounds like for a Data Analyst and for a Business Intelligence Analyst, and which round you are probably losing without anyone telling you.
No federal agency publishes a list of interview questions, and nobody else has a real one either. What does exist is the task data underneath the job. O*NET surveys the people doing the work and rates each task and skill on a 0–100 importance scale, and hiring panels build their questions out of tasks — so the ratings tell you where the weight sits before a single question gets asked. Every O*NET rating below is the figure the site carried at its 2026 update.
What the panel is sampling from
Two O*NET occupations cover the ground that most Data Analyst postings describe: Data Scientists (15-2051.00) and Business Intelligence Analysts (15-2051.01). Their task lists are not the same, and the difference is worth ten minutes of your preparation.
For Business Intelligence Analysts, the highest-rated task is "generate standard or custom reports summarizing business, financial, or economic data" at 91, followed by "maintain or update business intelligence tools, databases, dashboards, systems, or methods" at 84 and "manage timely flow of business intelligence information to users" at 80. That is a reporting job with a service obligation attached to it. For Data Scientists, the two top tasks are "analyze, manipulate, or process large sets of data using statistical software" and "create graphs, charts, or other visualizations to convey analysis results", both rated 83.
So if the posting says Business Intelligence Analyst, expect more questions about stakeholders, refresh schedules and who owns a report when it breaks. If it says Data Analyst on a product team, expect more about defining a metric and testing whether a change did anything at all. Both sit inside the same federal occupation code, 15-2051, which is why BLS publishes one wage figure covering the pair of them. Same wage line, different interview.
The tool question does not split the same way, and this is where candidates over-prepare in the wrong direction. O*NET flags a subset of each occupation's software as hot or in-demand, and the two lists overlap far more than the job titles imply: Microsoft Power BI, Alteryx, Google Looker, Snowflake, Apache Spark, Excel and PowerPoint are flagged on both pages. The shared middle is most of the stack; the interview lives in the tails. The Business Intelligence Analyst page adds a reporting-and-enterprise edge — Salesforce, SAP, Google Analytics, SQL Server Reporting Services. The Data Scientist page adds a modelling-and-engineering edge — TensorFlow, PyTorch, SAS, MATLAB, Docker, Kubernetes, Git, Apache Airflow. So revising TensorFlow for a Business Intelligence Analyst interview is studying for the wrong exam. Treating a Data Analyst screen as though Power BI were somebody else's tool is the same mistake pointed the other way, and it is the more common one.
The SQL round
You get two or three tables and a question in English. Write the query, out loud, while somebody watches. The syntax is not the point.
| What you are handed | What is actually being graded |
|---|---|
| Two tables, "how many users ordered twice in March?" | Whether you ask what counts as a user before you type |
| A query returning a plausible but wrong number | Whether you notice the join duplicated rows |
| "Now do it per country, ranked" | Whether window functions are muscle memory or a struggle |
| "This query runs for nine minutes" | Whether you can name one likely cause before guessing |
The trap in the first row is the one that costs people offers. A Data Analyst asked for "users who ordered twice in March" who starts typing immediately has skipped three questions: does a guest checkout count as a user, does March mean the order date or the ship date, and does a refunded order still count. Ask first. Then write. O*NET rates "identify business problems that can be addressed through data analysis" at 78 for Data Scientists, which is the formal way of saying that turning a vague request into a precise one is part of the job rather than a preamble to it.
One more habit worth building for this round: say what you would check next. "This returns 4,812, and the first thing I'd do is compare it against the raw order count to see whether the join inflated it." That sentence has ended more SQL rounds well than any elegant common table expression.
The round that decides more offers than SQL does
Most data analyst candidates prepare for the technical screen and lose the job in the communication question, and the task data says why. Look at what O*NET rates highest among daily work activities for Data Scientists: "analyzing data or information" at 96, "working with computers" at 94, "processing information" at 93 — and then "interpreting the meaning of information for others" at 87, which sits above "getting information" and above "making decisions and solving problems", both at 83. For Business Intelligence Analysts the same activity rates 89, and "communicating with supervisors, peers, or subordinates" rates 81.
The skills list makes it starker still.
Reading comprehension at 75, speaking at 72, writing at 69, and mathematics at 63 behind all of them. The Bureau of Labor Statistics puts the same point in plain words in its description of the occupation: data scientists "must be able to convey the results of their analysis to technical and nontechnical audiences to make business recommendations."
The question that tests this usually arrives sounding harmless. "How would you explain your last analysis to someone in finance?" Or the compressed version: the VP has ninety seconds, go. A Business Intelligence Analyst who answers by describing the pipeline — the sources, the joins, the refresh cadence — has failed it politely and will not be told. Start with the finding and the decision it implies; the method comes only if somebody asks for it. Ninety seconds is three sentences. Spend all three on the conclusion.
"A number moved. What do you do?"
Here is a full answer worth memorising the shape of, for a Data Analyst asked why weekly signups fell 12 percent:
"First I'd check whether it's real — is the tracking still firing, did a release change the event name, is the week actually complete? If it's real, I'd cut it by the dimensions that could isolate it: channel, platform, country, new versus returning. If one segment carries the whole drop, that's usually a bug or a paused campaign. If it's spread evenly across every segment, it's usually seasonality or something upstream in traffic. Then I'd say which of the two it looks like, and what I'd need to be sure."
Notice what that does. Instrumentation first, splitting before theorising, and a stated confidence level at the end instead of a verdict. Panels ask this because it is the actual work: the second-highest-rated task for Business Intelligence Analysts is keeping the tools and dashboards correct, at 84, and half of that job is deciding whether a strange number is the world or the pipeline.
"Tell me about a project"
The version that lands starts with the decision and not the stack. "Support wanted to know which ticket categories drove repeat contacts, so I joined ninety days of tickets to the account table, found two categories accounted for most of the repeats, and they reworded the help article for one of them." Tools go in a clause at the end, if they go in at all.
A project that went nowhere is still a good answer, and a better one than a polished story you cannot defend. A Business Intelligence Analyst who says "I built a churn dashboard and nobody opened it — I'd built it from what was easy to query rather than asking the two people who'd use it what they decide each week" has just demonstrated the task O*NET rates at 80 for this occupation, managing the timely flow of business intelligence information to users. That is a better signal than a dashboard with a screenshot.
Will they ask about your degree?
Rarely, and not the way candidates fear. O*NET places Data Scientists in Job Zone Four, "considerable preparation needed", where 48 percent of surveyed workers reported a bachelor's degree and 44 percent a master's — so a Data Analyst without a graduate degree is in the larger group, not the exception.
"What are your salary expectations?"
Answer with a range and a location attached to it. The federal wage data for this occupation is unusually wide, and that width is your argument for refusing to name a single number.
Data scientists: wages and outlook, May 2025
BLS puts the median at $120,230 as of May 2025, with the lowest 10 percent under $67,240 and the highest 10 percent above $199,130, across 275,600 jobs. Employment is projected to grow 35 percent from 2025 to 2035, with about 24,800 openings a year on average over the decade. A range that sits inside that spread and names your metro is defensible. A single number is a guess you will be held to for two years.
What this data does not tell you
O*NET surveys the people doing the job, not the people hiring for it. Nothing in it says how many rounds a company runs, what share of candidates clear the SQL screen, how long a take-home is meant to take, or whether a particular employer will ask about machine learning at all. The ratings tell you what the work weighs. They do not tell you what any one panel will do with an hour of your time, and anyone selling you a ranked list of "the 50 questions they will ask" is inventing the ranking.
The take-home, if there is one
Treat the written summary as the assignment and the analysis as the appendix. Somebody will skim one page before they open a notebook, and that page is where "interpreting the meaning of information for others" — 89 for Business Intelligence Analysts, 87 for Data Scientists — actually gets marked. Two paragraphs will do it: what you found, and what you would do about it. Then add the caveat you would raise if this were real work, because naming the weakness in your own analysis is the most reliable signal available that you have done this before.
Keep the code readable and stop when the question is answered. A Data Analyst who returns four charts and one recommendation reads as senior. Fourteen charts and no recommendation reads as somebody who could not decide what mattered, and deciding what matters is the job.
What to do this week
Three things, in this order. Write the ninety-second version of your last project — finding first, method last — and say it out loud until it fits inside ninety seconds without rushing. Then take one query you wrote recently and practise narrating it while you type, ending with the sentence about what you would check next. Finally, look up the wage range for your own metro before the recruiter call, so the money question gets a range and not a pause.
If the harder constraint is getting to the interview at all rather than passing it, HeroApply fills the applications from your resume, so the hours go into the four rounds above instead of into forms.
Sources
- BLS Occupational Outlook Handbook — Data Scientists, May 2025 wages and 2025–35 projections
- O*NET OnLine — Data Scientists 15-2051.00, tasks, work activities and technologies (updated 2026)
- O*NET OnLine — Business Intelligence Analysts 15-2051.01, tasks, skills, activities and technologies (updated 2026)
Frequently asked questions
What questions are asked in a data analyst interview?
Almost all of them fall into four groups — a SQL or query exercise, a diagnostic case about a metric that moved, a walkthrough of a project you owned, and a question about explaining a result to someone non-technical. O*NET rates "deliver oral or written presentations of mathematical modeling results" at 80 out of 100 in importance for the Data Scientists occupation, above "clean and manipulate raw data using statistical software" at 76. The order of your preparation should match that.
How much SQL do you need for a data analyst interview?
Enough to join two tables, aggregate with a GROUP BY, filter on a date range, and use a window function without stalling. Interviewers are usually checking whether you can turn a business question into a query, not whether you have memorised syntax. If you can write the query and then say out loud what could be wrong with the result, you are past the bar for most postings.
How do I answer "tell me about a data project" as a data analyst?
Lead with the decision the analysis changed, not the tooling. Name the question you were asked, the data you had, the thing you found, and what somebody did differently afterwards. A project that ended in a dashboard nobody opened is still a usable answer if you say why it went unused and what you would change.
What should a data analyst say about salary expectations?
Give a range anchored to public wage data for the occupation and the metro, then say the total package is negotiable. BLS puts the median wage for data scientists at $120,230 as of May 2025, with the lowest 10 percent under $67,240 and the highest 10 percent over $199,130. That spread is wide enough that a number with no location attached tells the recruiter almost nothing.
Do data analyst interviews still use take-home assignments?
Many do, usually as a small dataset with an open question attached. Treat the written summary as the graded part, because O*NET rates "interpreting the meaning of information for others" at 89 out of 100 for business intelligence analysts — the highest-rated work activity on that list that is not handling the data itself, and well above "communicating with supervisors, peers, or subordinates" at 81. A clean two-paragraph answer beats a notebook with forty cells and no conclusion.
Let HeroApply do the applying
It finds the jobs that match your resume, answers the screening questions in your voice, and submits the application — desktop, phone or cloud.
Jordan Reyes · Careers Editor
Jordan writes the resume, cover letter and interview guides — the parts of the search that happen before anyone presses Apply. Every example comes from a real job title.
The best of the blog, once a week
One email with the week’s guides. No spam, unsubscribe any time.
