BI Tools interview questions
BI Tools questions a Data analyst / BI can be asked, from first job to senior. Open one to see the strongest answer and why it works.
A stakeholder pastes a screenshot of a number and asks you to “refresh it.” What do you need?
The strongest answer: The query, the grain, and the time cut — a picture is not a metric
Why it works: Exactly. Source or it is archaeology. Memory is how you ship a cousin of the number.
- Get the query and the grain.
- Get the time cut.
- A screenshot is not a source.
- Do not recreate from memory.
A director wants to click a region on your chart and see the countries inside it. How do you build that?
The strongest answer: Define the hierarchy in the model — region, country, store — and put it on one visual, so drilling down keeps the same measure and the same filters, with a detail view for row-level questions
Why it works: Right — one visual, one measure, a hierarchy underneath. The numbers then agree at every level by construction.
- One visual, one measure, one hierarchy.
- Define the levels in the model.
- Drilling keeps the filters.
- Offer a detail view, not raw exports.
A user clicks a slicer and one chart on the page changes while the chart next to it does not move.
What do you check?
The strongest answer: The scope and the model — whether that filter applies to the visual, the page or the whole report, and whether the second chart’s table is actually related to the slicer’s table
Why it works: Exactly — a filter reaches only what it is scoped to and what the model connects it to. An unrelated table never responds.
- Filters have a scope.
- Visual, page, or whole report.
- The model decides what responds.
- Never fix it by removing the control.
A Tableau view duplicates marks after you add a dimension. What happened?
The strongest answer: You changed the visual grain — each mark is a group; I will check the pill and the table calc, not “Tableau is adding rows to the warehouse”
Why it works: Exactly. Viz grain ≠ warehouse grain. Blending another copy is how it gets worse.
- Check the visual grain.
- Pills and table calcs first.
- Do not blend a second copy by habit.
- Do not hide a lying tooltip.
A “debug” page lists emails and phone numbers for anyone on the VPN. What do you do?
The strongest answer: Take it down, restrict to a need-to-know role, and ask whether those fields must be there — VPN is not a purpose
Why it works: Exactly. Purpose and access. Exports multiply the copy. VPN is a network, not a lawful basis.
- Need-to-know, not VPN-wide.
- Ask if the field must exist.
- Do not export to “fix” access.
- VPN ≠ purpose.
Twelve dashboards compute “active” differently. What do you build?
The strongest answer: A semantic layer or certified model those twelve read — I do not add a thirteenth SQL dialect
Why it works: Exactly. One certified computation. Meetings and sheets are how the thirteenth appears.
- One certified definition.
- Dashboards consume it.
- No thirteenth dialect.
- Meetings are not a layer.
A margin percent field in the BI tool reads correctly on a row of detail but is wrong on every subtotal.
What happened, and how do you write the field?
The strongest answer: It is computed per row and then aggregated — usually averaged — by the tool; I write it as an aggregate, sum of margin over sum of revenue, so it recomputes at whatever grain the view shows
Why it works: Right — where the division happens decides the answer. Divide the sums; never average the ratios.
- Aggregate, then divide.
- Row-level calculations get averaged.
- Write ratios over the sums.
- Check the field at every grain.
Self-serve BI is producing 30 versions of revenue. What do you change?
The strongest answer: Certify a few sources and make the rest look unofficial — self-serve without a paved road is a copy machine
Why it works: Exactly. Paved and dirt. Off and careful are how shadows or chaos win.
- Certify the few.
- Mark the rest unofficial.
- Do not kill self-serve.
- Do not hide the badge.
Half the workspace is unused dashboards. What do you do?
The strongest answer: Expire them with an owner ping — unused certified numbers still confuse search — I do not keep a museum “just in case”
Why it works: Exactly. Search is the scarce thing. Surprise deletes and eternal museums are the failure modes.
- Expire with a ping.
- Search is scarce.
- No surprise Friday deletes.
- “Legacy” is not a lifecycle.
A new joiner asks for warehouse admin “to explore.” What do you give?
The strongest answer: A role that matches the job, plus a certified sandbox — explore is not admin, and PII is not a playground
Why it works: Exactly. Least privilege plus a safe sandbox. Hazing and prod-write are the other sins.
- Role matches the job.
- Sandbox for explore.
- No hazing lockouts.
- No prod write to learn.
Extract the data into the BI tool or query the source live — how do you decide?
The strongest answer: On how fresh the answer must be and what the source can take: an in-memory extract is fast and predictable but as old as its last refresh, a live connection is current and puts every click on the warehouse
Why it works: Exactly — it is freshness against source load, and the refresh schedule is part of the answer either way.
- Trade freshness against source load.
- Extracts are fast and dated.
- Live is current and costly.
- Publish the refresh schedule.
Power BI and Tableau both exist. How do you stop a third?
The strongest answer: A written paved road: who uses which, and a sunset for the rest — two is already expensive; a third is a hobby
Why it works: Exactly. One paper, a sunset. Lunch and freezes are how you get four.
- Write who uses which.
- Sunset the extras.
- Do not add a third hobby.
- Do not freeze the company.
Product wants event-level replay of named customers in a shared tool. What do you require?
The strongest answer: A purpose, access, retention, and a less-identifying default — replay is surveillance unless it is scoped
Why it works: Exactly. Purpose-limited replay. Speed, unread wrap, and someday-hash are not a design.
- Purpose, access, retention.
- Default to less identifying.
- Replay is scoped, not a toy.
- No someday-hash.
Power BI, Tableau, Looker — what actually differs, and what decides it for a company?
The strongest answer: Mainly where the model lives — Power BI keeps an in-memory model and DAX, Looker defines metrics in code and queries the warehouse, Tableau leads on visual exploration. Decide on that layer, your licensing and your skills
Why it works: Right — the modelling layer is the part you live with; the chart library is the part people argue about.
- The model matters more than the charts.
- Ask where metrics are defined.
- Weigh identity, licensing and skills.
- Say what you are giving up.
What does an analytics-platform team ship?
The strongest answer: Certified models, access paths, and quality pages a squad can use on Tuesday — not a rewrite of every local query
Why it works: Exactly. The road. Centralising every chart and annual tools are bottlenecks and taxes.
- Ship certified models and pages.
- Squads add decisions, not dialects.
- Do not steal every chart.
- Do not annual-rewrite the BI.
Give the one sentence you want every squad to say about analytics platform.
The strongest answer: “I can find the certified metric, see its freshness, and read the definition before lunch” — if they cannot, you built a ticket queue, not a road
Why it works: Exactly. Self-serve meaning before lunch. Tickets and hope and sheets are the failed platform.
- Find, freshness, definition — before lunch.
- Not a ticket.
- Not copy-and-hope.
- Not a shadow sheet.