Interview guide for data analysts: a full loop with a SQL session and an analysis case

A data analyst is useful when the numbers are right and the people using them understand what they mean. The interview loop has to test both: whether the candidate can get correct answers out of messy data, and whether they can turn a vague business question into a clear finding someone acts on. This guide gives hiring managers a complete loop for a mid-level analyst supporting a business team: five stages, who owns which competency, core questions with what to listen for, a live SQL session and an analysis case, the scorecard weights, and the debrief decision rule.

The recruiter's first call is covered in data analyst phone screen questions. This page starts there and runs to the decision.

The loop at a glance

The hiring manager interview can come before or after the case; running it after means the manager can probe anything the case left unclear.

Stage 1: the screen (30 minutes)

• "In a normal week, what tools do you use, and roughly how much of your time is SQL, spreadsheets, a BI tool, or Python or R?" Listen for: a realistic split and tools at the level the role needs.
• "Tell me about a report or dashboard you built that someone made a decision from." Listen for: who used it, what decision, and whether the candidate knows if it was the right one.

Stage 2: the live SQL session (60 minutes)

A senior analyst shares a small invented schema: customers, orders, order items and a products table, with a few hundred rows loaded in a database the candidate can query. The data has three planted problems: duplicate order rows from a retry, a handful of orders with no customer, and timestamps stored in UTC while the business reports in local time.

1. Warm-up (10 minutes): "How many orders did we take last month, and what was the revenue?"
2. Join and aggregate (20 minutes): "Which ten customers spent the most in their first 90 days?"
3. Window or cohort question (20 minutes): "What share of customers who bought in January bought again within 60 days?"
4. Wrap-up (10 minutes): "If finance said your revenue number was 4% higher than theirs, where would you look first?"

Candidates may look up syntax. Listen for: clarifying the definition before writing ("last month by order date or ship date?"), checking row counts after each join, noticing the duplicates or orphaned orders, and sanity-checking results against a rough expectation. A query that runs but double-counts because of the duplicates, without the candidate checking, scores below one that is slower but verified. The attention to detail questions pair well with this stage if you want a behavioral check too.

Stage 3: the analysis case

Send an invented dataset of about a year of subscription data (sign-ups, plan, channel, cancellations, support tickets) and one question from an invented head of marketing: "Cancellations went up in the last quarter. What happened, and what should we do?" Cap the work at two hours, say so in writing, and ask for no more than five slides or a one-page memo.

The 45-minute discussion

• 15 minutes: the candidate presents to the hiring manager and a business stakeholder.
• 20 minutes of questions: "How confident are you in this? What else could explain it? What would you need to be sure?"
• 10 minutes: the stakeholder asks one question a non-analyst would ask, such as "So should we cut the discount plan or not?"

Listen for: a finding stated in one sentence at the start, a check on whether the rise is real or a change in how cancellations were recorded, a comparison by segment rather than one overall number, and a clear separation between what the data shows and what the candidate is guessing. Strong candidates say what they would not conclude. How to set fair limits and score take-home work is in how to evaluate a take-home assignment.

Stage 4: stakeholder interview (30 minutes)

• "Tell me about a request where what the person asked for was not what they needed." Listen for: asking what decision the analysis would support, proposing a different cut, and agreeing it before building.
• "Tell me about a time your analysis gave an answer the stakeholder did not want." Listen for: presenting it plainly, explaining the method, and not softening the finding to keep someone happy.

Stage 5: hiring manager interview (45 minutes)

1. "Tell me about a number you published that turned out to be wrong." Listen for: how it was found, who they told and how fast, the correction, and a check they added afterward. An analyst who has never shipped a wrong number either has not shipped much or did not notice.
2. "You have five requests from three teams and time for two this week. How do you decide?" Listen for: asking about impact and deadlines, telling the others when to expect theirs, and involving the manager when priorities conflict across teams.
3. "Tell me about a metric your team used that you think measured the wrong thing." Listen for: a specific flaw, such as a definition that rewarded the wrong behavior, and what they proposed instead.

Scorecard competencies and weights

Example weights for a mid-level analyst supporting business teams. Set yours at intake.

A reporting-heavy role might move weight to data manipulation and a BI tool; a role closer to experimentation would add statistics and move toward the data scientist loop. You can build and print the sheet in the scorecard builder.

The debrief decision rule

1. Scorecards and the candidate's case output are shared before the debrief; everyone scores before reading others.
2. Gate: data manipulation at 3 or above. The role cannot be done without it, and it is hard to learn on the job at the speed the team needs.
3. Floor: no 1 on data quality skepticism or on rigor. An analyst who trusts every table will eventually publish a wrong number with confidence.
4. Weighted total: in this example, 2.8 or higher on the 1–4 scale is an offer.
5. The stakeholder's view counts on communication: if the hiring manager scored communicating findings a 3 and the stakeholder a 2, take the stakeholder's notes seriously; they are the audience.

Adjusting by level and type

Common mistakes in analyst loops

• Syntax quizzes. Asking for exact function names tests memory, which the job does not need. Test whether they get a correct, verified answer.
• Clean data in the exercise. Real data is never clean. If the exercise has no traps, it cannot show skepticism.
• No business stakeholder in the loop. The analyst's customers are the best judges of whether the work is useful.
• Unlimited take-homes. Candidates with more free time look better. State the cap and score within it.

After the hire, the 30-60-90 day plan for data analysts covers onboarding into your data and stakeholders.