Clarifying questions and stakeholders

1. junior

   1. You receive a messy export and a one-line request: "clean this". What do you ask before starting?

   What it tests: Clarifying questions.

   Model answer and scoring

   What the data will be used for and what one row should represent; which columns are identifiers; how missing values and duplicates should be handled; which formats dates and amounts use; what to do with invalid rows (delete, fix, quarantine); and who decides when a rule is unclear.

   Strong answers

   • Purpose, grain, rules for invalid rows, decision owner.

   Red flags

   • Starts cleaning immediately.

2. mid

   2. A business owner asks why 120 rows are missing from your clean table. How do you explain it?

   What it tests: Communicating exceptions.

   Model answer and scoring

   They aren't lost; they're in the exception report, each with a reason. I'd show the counts by reason (for example 70 missing_offset, 50 unknown_customer), explain what each means, and say what would fix it at the source. The reconciliation shows input rows equal clean plus exceptions.

   Strong answers

   • Counts by reason and a path to fix.

   Red flags

   • "They were bad data."

3. mid

   3. How do you tell a stakeholder that a date column can't be reliably parsed?

   What it tests: Explaining ambiguity without jargon.

   Model answer and scoring

   "Some dates like 03/04/2026 could be 3 April or 4 March, and the file doesn't say which convention it uses. Rather than guess and risk wrong dates in the report, I've set those rows aside. If the source can tell us the convention, or send ISO dates, I can include them."

   Strong answers

   • Concrete example, risk of guessing, a clear ask.

   Red flags

   • Silently picks a convention.

4. senior

   4. A stakeholder asks you to "just delete the outliers" before a board report. What do you do?

   What it tests: Pushing back with options.

   Model answer and scoring

   I'd ask what makes them outliers and check whether they're legitimate (large orders, refunds). Then I'd offer options: flag them and report results with and without, or exclude a defined population with the rule written into the report. Deleting real data silently changes the population the board is looking at.

   Strong answers

   • Checks legitimacy and offers transparent options.

   Red flags

   • Deletes them without a rule.

   Practise it: Flag large transactions, keep refunds

5. mid

   5. You find ten data problems and have one day. How do you prioritise?

   What it tests: Impact-based prioritisation.

   Model answer and scoring

   I estimate each problem's effect on the numbers the stakeholder uses, how many rows it touches and how hard it is to fix, and start with the high-impact, cheap ones.

   I tell the stakeholder which issues remain and how they could affect the results.

   Strong answers

   • Impact on decisions first.
   • Communicates leftovers.

   Red flags

   • Fixes the easiest ones only.

6. mid

   6. Your cleaning shows last quarter's report overstated revenue. How do you tell the stakeholder?

   What it tests: Communicating uncomfortable findings.

   Model answer and scoring

   Privately first, with the size of the difference, the cause, the evidence and what you recommend, without blaming people.

   I offer a corrected figure and a check to prevent it next time.

   Strong answers

   • Quantified, evidence-based, constructive.

   Red flags

   • Announces it publicly without checking.

7. junior

   7. A manager asks for "customers". What do you ask before starting?

   What it tests: Clarifying requirements.

   Model answer and scoring

   Which customers count (active, paying, any account), over which period, at what grain (person, household, account), and what decision the number supports.

   I write the definition down and confirm it.

   Strong answers

   • Asks about scope, period and grain.

   Red flags

   • Starts querying immediately.

8. junior

   8. Who should decide whether an invalid record is corrected or excluded?

   What it tests: Data ownership.

   Model answer and scoring

   The data owner or the business requester, not the analyst alone. I quantify the impact, propose an option and record their decision.

   I can apply a temporary rule if it's clearly documented and reversible.

   Strong answers

   • Escalates to the owner with evidence.

   Red flags

   • Decides silently.

9. mid

   9. What do you include when handing a cleaned dataset to another analyst?

   What it tests: Documentation.

   Model answer and scoring

   The rules applied in order, assumptions, exception counts by reason, open questions, how to rerun the pipeline and where raw data lives.

   A short reconciliation proving nothing was lost.

   Strong answers

   • Rules, assumptions, exceptions, rerun steps.

   Red flags

   • Just the file.

10. senior

    10. A stakeholder wants the numbers today, but a key field is unreliable. What do you do?

    What it tests: Trade-offs under pressure.

    Model answer and scoring

    I deliver what can be trusted today, clearly label what can't, and give the expected effect of the unreliable field with a date for the corrected version.

    I avoid presenting a number with false precision.

    Strong answers

    • Transparent partial delivery.

    Red flags

    • Ships everything without caveats.

Model answers are original study material for practice. They are not employer-endorsed and don't guarantee an outcome.