4) Technical and professional questions (what separates prepared candidates)
This is where NZ employers quietly filter. They don’t need you to recite textbook definitions—they need to know you can ship reliable insights with the tools they use, and that you understand the local expectations around privacy and reporting.
Q: Walk me through how you’d build a KPI dashboard that executives will trust.
Why they ask it: They’re testing end-to-end thinking: data lineage, definitions, refresh, and adoption.
Answer framework: Pipeline narrative (Source → Transform → Validate → Visualize → Govern)
Example answer: “I’d start with KPI definitions and owners—what counts as revenue, active customer, churn, and when they’re ‘final.’ Then I’d map sources and build a clean model in SQL or dbt with documented logic. I’d add validation checks: row counts, null thresholds, and reconciliation to finance totals where relevant. In Power BI, I’d keep visuals simple and add drill-through for trust. Finally, I’d set refresh schedules, access controls, and a change log so stakeholders know when logic changes.”
Common mistake: Talking only about charts and ignoring data validation and governance.
Q: Write a SQL approach to find duplicate customers when identifiers aren’t consistent.
Why they ask it: Real-world data is messy; they want pragmatic matching logic.
Answer framework: Heuristic matching (standardize → block → score → review)
Example answer: “I’d standardize fields first—lowercase names, trim spaces, normalize phone formats. Then I’d ‘block’ on something like postcode + first letter of surname to reduce comparisons. Within blocks, I’d use fuzzy matching or simple similarity rules—same DOB and similar name, or same email domain + phone match. I’d output a candidate pairs table with a match score and sample it for manual review before merging.”
Common mistake: Claiming you can solve it perfectly with one SQL query and no review step.
Q: How do you choose between Power BI measures (DAX) and doing logic in SQL?
Why they ask it: They want maintainability and performance, not a fragile dashboard.
Answer framework: “Model first” rule (SQL for reusable logic, DAX for presentation-layer calculations)
Example answer: “If the logic is a business definition—like ‘net revenue’ or ‘active customer’—I prefer SQL/dbt so it’s centralized and testable. I use DAX for things that are truly report-layer: time intelligence, dynamic slicer behavior, or small calculations that don’t belong in the warehouse. The goal is one definition of truth, not five slightly different measures across reports.”
Common mistake: Doing everything in DAX because it’s faster in the moment.
Q: Explain how you’d validate a metric against a source of record (like finance).
Why they ask it: In NZ, finance reconciliation is a credibility gate for analysts.
Answer framework: Reconciliation checklist (scope → timing → mapping → variance explanation)
Example answer: “First I’d confirm scope: are we comparing invoiced revenue, recognized revenue, or cash received? Then I’d align timing—transaction date vs posting date. I’d map product codes and exclusions so categories match. Finally, I’d quantify variances and explain them—refund timing, FX, write-offs—until we’re within an agreed tolerance. I’ll document the reconciliation so it’s repeatable each month.”
Common mistake: Treating any mismatch as ‘finance is wrong’ or ‘data is wrong’ without investigating definitions.
Q: What’s your approach to experiment design or A/B testing in a product or marketing context?
Why they ask it: Many NZ companies want analysts who can measure change, not just report.
Answer framework: Hypothesis → Metric → Randomization → Power/Duration → Decision rule
Example answer: “I start with a clear hypothesis and a primary metric that reflects the decision. I check whether we can randomize at the right unit—user, session, store—and avoid contamination. Then I estimate baseline rate and minimum detectable effect to set duration. I predefine the decision rule and guardrails like churn or support tickets. After the test, I interpret results with confidence intervals and practical impact, not just p-values.”
Common mistake: Running tests without a primary metric or stopping early when results ‘look good.’
Q: How do you handle missing data and outliers in business reporting?
Why they ask it: They want to know if you’ll hide problems or surface them responsibly.
Answer framework: Triage (cause → impact → treatment → disclosure)
Example answer: “I first identify the cause—pipeline failure, late-arriving data, or genuine business change. Then I assess impact: does it change decisions or just add noise? For reporting, I’ll use clear rules—impute only when defensible, cap outliers if they’re errors, and always annotate dashboards when data is incomplete. If it’s a pipeline issue, I’ll add monitoring so it doesn’t repeat.”
Common mistake: Quietly deleting outliers to make charts look clean.
Q: In New Zealand, what privacy considerations matter when working with customer data?
Why they ask it: They need to trust you with personal information and compliance.
Answer framework: Principle-based answer (collect → access → use → retain)
Example answer: “I work from the Privacy Act 2020 principles: only use data for a legitimate purpose, minimize what you extract, and control access. Practically, I’ll use role-based permissions, avoid exporting raw personal data to spreadsheets, and anonymize or aggregate wherever possible. If we’re sharing insights externally, I’ll check re-identification risk and document the basis for use. And if there’s a suspected breach, I escalate immediately through the company process.”
Common mistake: Treating privacy as ‘IT’s job’ instead of part of analytics practice.
Q: What would you do if the BI tool fails an hour before a monthly performance meeting?
Why they ask it: They’re testing resilience and stakeholder management under pressure.
Answer framework: Stabilize → Communicate → Fallback → Post-mortem
Example answer: “First I’d confirm whether it’s a refresh failure, gateway issue, or access outage and capture error details. I’d message the meeting owner with a realistic ETA and a fallback plan. Then I’d produce a minimal set of critical KPIs from the warehouse via SQL and export a controlled snapshot with definitions noted. After the meeting, I’d run a post-mortem: root cause, monitoring, and a documented runbook so we’re not improvising next month.”
Common mistake: Going silent while you troubleshoot, leaving stakeholders blindsided.
Q: How do you document and test transformations so other people can trust them?
Why they ask it: NZ teams often have small data functions; handover matters.
Answer framework: “Docs + tests + lineage” (lightweight but consistent)
Example answer: “I document models at the level someone can maintain: what the table represents, grain, key joins, and known limitations. I add tests—unique keys, not-null fields, accepted values—and I monitor freshness. I also keep lineage visible, so a dashboard metric can be traced back to source tables. That’s how you scale trust without needing a big governance team.”
Common mistake: Relying on tribal knowledge or comments buried in SQL files.
Q: What’s the difference between a Reporting Analyst and a Data Analytics Specialist in practice?
Why they ask it: They want to place you correctly: operational reporting vs deeper analytics.
Answer framework: Scope comparison (cadence → ambiguity → methods → stakeholders)
Example answer: “A Reporting Analyst role is often about reliable recurring outputs—KPI packs, dashboards, reconciliations—with tight definitions and deadlines. A Data Analytics Specialist usually has more ambiguity: diagnosing drivers, designing experiments, building forecasting or segmentation, and influencing product or commercial strategy. I’m comfortable in both, but I’m strongest when I can improve the reporting foundation and then use it to answer higher-value questions.”
Common mistake: Dismissing reporting work as ‘basic’—in NZ, reliability is respected.