Key takeaways
- Ask for specific decisions and follow up on what the candidate would change.
- Use ambiguous scenarios to test structure, not a single correct answer.
- Listen for reconciliation methods and honest handling of uncertainty.
- Score evidence with a consistent rubric and a realistic work sample.
On this page
Ask for a decision, not a definition
Any candidate can define a p-value or explain a join. Judgement appears when they describe a decision they influenced, the information they had, and what they did when it was incomplete. Ask for a specific example, then ask what they would do differently.
Prefer questions with no single right answer. The aim is to hear how the candidate structures the problem, not whether they guess the answer the interviewer has in mind. A confident wrong answer is informative, provided the rubric captures the reasoning.
Test how they handle an ambiguous request
Scenario: a head of operations asks for a dashboard showing warehouse productivity by the end of the week. There is no agreed definition of productivity, two source systems, and no analyst time for a discovery phase. Ask what they would do first, what they would deliver in week one, and how they would prevent the dashboard from becoming a misleading scoreboard.
Strong answers separate the request from the decision, ask who will use the output and what they will do differently, propose a small first version with an explicit definition, and name the trade-off. Weak answers start with tooling or accept the deadline without surfacing the definitional problem.
Probe data quality and disagreement
Scenario: finance reports one revenue figure, the dashboard another, and both teams are certain. Ask how they would investigate, in what order, and when they would stop. Listen for grain, filters, timing, currency, refunds, and duplicated joins before anything about the tool.
A strong candidate will also ask who owns each number and what decision depends on it. That separates a reconciliation exercise from an argument about whose report is better.
- Confirm the period, grain, and whether both figures include the same entities.
- Reconcile a single day or a single account before comparing the full month.
- Check the source of each number and whether one is derived from the other.
- Document the difference and agree which number is authoritative for which purpose.
Ask how they communicate uncertainty
Scenario: an analysis suggests a campaign increased conversion, but the sample is small and the result is not statistically significant. The marketing lead wants to present it as a success. Ask what the candidate would say, in writing, and what they would recommend.
Look for a clear separation between what was observed and what is inferred, a plain-language explanation of the uncertainty, and a recommendation that does not overclaim. The best answers offer a way to present the result honestly without dismissing the team's work.
Test ownership and handover
Ask about a report they built that someone else had to maintain. What documentation existed, how did they handle a change in source data, and what would they improve? A candidate who has never considered handover may build something that only works while they are in the room.
Follow up with a small practical exercise: a messy extract and a business question. Review the assumptions they write down as much as the final number. A short exercise reveals habits that a conversation can hide.
Score evidence, not confidence
Use the same questions across candidates and record notes against the signals. Confidence is not evidence of competence; a calm answer can still be vague. Check references for the examples that matter, and avoid hiring on the strength of a single impressive story.
| Question | What it tests | Strong signal |
|---|---|---|
| When did you change a recommendation? | Willingness to update | Names the evidence, not just the outcome |
| A stakeholder wants a metric you distrust | Integrity and influence | Offers an alternative and explains the risk |
| Two sources disagree on revenue | Method | Proposes a reconciliation path before tooling |
| How would you hand over this report | Ownership | Documentation, validation, and a named owner |
| Estimate impact from incomplete data | Reasoning | States assumptions and a range, not false precision |
Frequently asked questions
Should I ask brainteasers in an analyst interview?+
Rarely. They test familiarity with puzzles rather than reporting judgement. A messy dataset and a short business brief produce stronger evidence of how the candidate actually works.
How long should a work sample take?+
Around sixty to ninety minutes for a realistic extract and a short written recommendation. If the task takes much longer, pay the candidate or use a smaller brief; unpaid extensive work discourages good applicants.
How do I avoid bias when scoring?+
Use the same questions and brief for every candidate, score against a written rubric, and have interviewers record notes before discussing. Compare evidence rather than impressions of confidence or cultural similarity.
What if the candidate has no work experience?+
Ask about a course project, volunteering, or a self-directed analysis. Look for the same reasoning: a clear question, an explicit method, an honest limitation, and a recommendation tied to evidence.
Need help applying this?
Improve your bi dashboards & reporting.
Review the service scope or tell us what you are trying to improve. We will reply with a clear next step.
Free KPI definition templateRelated reading
A data quality checklist before starting a BI project
The checks that prevent a dashboard from failing after launch: ownership, definitions, completeness, duplicates, and reconciliation.
When to move from spreadsheets to a database
The signals that a spreadsheet has become a database, and the simplest path to a more reliable system.