BI engineer behavioural interview questions

As asked

Tell me about a time when two stakeholders had different definitions of the same KPI and both believed their version was correct. How did you resolve it?

Sample answer outline

Strong answer shows the candidate facilitated a meeting with both stakeholders, traced each definition back to its business purpose (e.g., finance includes refunds, marketing does not), documented the difference, and proposed two certified metrics (one per use case) rather than forcing one winner. They should mention communicating the decision to a wider audience and versioning the metric definitions in a tool like dbt or a wiki.

Expect these follow-ups

• What did you do when one stakeholder refused to accept the agreed definition?
• How did you prevent the same problem from recurring with future metrics?

As asked

Describe a time when incorrect data made it into a dashboard that senior stakeholders had already seen or acted on. Walk me through what happened and how you handled it.

Sample answer outline

A strong candidate describes discovering the issue quickly (or being informed), immediately flagging the affected dashboard as under review, notifying the stakeholders who saw it with a clear explanation of what was wrong and by how much, providing a corrected figure, and doing a root-cause analysis. They follow up with a post-mortem and a preventive control (dbt test, monitoring alert) to stop a recurrence. They communicate calmly and own the problem rather than deflecting.

Expect these follow-ups

• How long did it take to detect the error, and what would have caught it sooner?
• Did any stakeholder lose trust in the data, and how did you rebuild it?

As asked

In the same week, the CEO asks for a revenue breakdown dashboard, the head of operations wants a logistics KPI tracker, and the head of engineering requests a pipeline health monitor. You have one week and one developer. How do you decide what to build?

Sample answer outline

The candidate should ask for urgency and business impact context before deciding. They should communicate transparently with all three requestors about the constraint and the order they have chosen and why. A strong answer discusses how they would assess which request has the highest business value this specific week (e.g., board meeting next week means CEO request is time-sensitive). They also consider whether any request can be partially satisfied quickly (5-minute query vs full dashboard).

Expect these follow-ups

• How do you communicate the delay to the requestors who are not getting their work done first?
• What process would you put in place to prevent this situation from recurring every week?

As asked

Tell me about a time a leader made a significant business decision based on a misinterpretation of a dashboard you built. How did you find out, and what did you do?

Sample answer outline

A strong answer shows the candidate took responsibility without throwing the leader under the bus, investigated whether the misinterpretation came from a design flaw (e.g., ambiguous label, missing context tooltip) or from the user not reading it correctly. They fixed the design, added a brief guide or tooltip, and followed up with the leader directly. They also describe whether the business impact was reversible and what they learned about dashboard literacy.

Expect these follow-ups

• What design change did you make to prevent the same misinterpretation?
• Do you think it was primarily a communication failure or a design failure?

As asked

Tell me about a time a business stakeholder asked you for a very complex report with 20 metrics across 5 tabs, and you pushed back. What happened?

Sample answer outline

The candidate should show they asked the stakeholder what decision the report was supposed to help them make, then proposed a simpler version focused on 3 to 4 key metrics. They should explain how they handled any pushback and describe the outcome. A strong answer shows both technical discipline (fewer metrics = better performance and less maintenance) and consultative skill (probing for the real need).

Expect these follow-ups

• How did you decide which metrics to keep and which to cut?
• How did you prevent the same stakeholder from coming back three months later asking for the full 20 metrics?

As asked

You joined a team and discovered the main revenue dashboard is powered by a single 500-line SQL query with no documentation, no tests, and hardcoded date ranges. How did you approach fixing it without breaking production?

Sample answer outline

First: understand what the query does by reverse-engineering it and speaking to the person who built it. Do not touch it until you understand it. Second: write integration tests against the current output so any refactor can be validated. Third: migrate incrementally using dbt models, one CTE at a time, validating each step. Finally: run old and new side by side in parallel for one reporting cycle before switching. A strong answer also discusses how they communicated the risk and timeline to stakeholders.

Expect these follow-ups

• How did you prioritize this refactor against new feature requests from the business?
• What was the first test you wrote before touching anything?

As asked

You built a dashboard you thought was valuable, but 3 months after launch the usage stats show only 2 of the intended 30 users open it regularly. What do you do?

Sample answer outline

Candidate should start with user research: talk to the 28 non-users. Common reasons include not knowing the dashboard exists, it does not answer their actual question, it is too slow, or they have a different tool they prefer. The answer should show a willingness to change the product rather than blame users, and to measure success by outcomes (decisions made, hours saved) not by dashboard views. A strong answer also distinguishes between a discoverability problem and a design problem.

Expect these follow-ups

• What metric would you track to know if the dashboard is genuinely valuable vs just being opened?
• How would you decide whether to rebuild the dashboard vs retire it?

As asked

The CFO needs a board-level revenue figure in 2 hours but the clean version of the data will not be ready for 8 hours. Tell me about a time you faced this kind of tradeoff and what you decided.

Sample answer outline

The candidate should describe being transparent with the CFO about the tradeoff: the fast number will be an estimate with a stated confidence range, and the clean number will follow by end of day. They should not fabricate a precise number from dirty data and call it clean. A strong answer includes: how they estimated the confidence interval, how they labeled the estimate in the presentation, and how they handled the follow-up when the final number differed slightly.

Expect these follow-ups

• If the estimate was off by more than 5% from the final number, how did you explain that?
• What would you do differently next time to reduce the lead time for the clean number?

As asked

Marketing wants dashboards optimised around leads, sales wants pipeline, and customer success wants retention. How do you design a BI reporting suite that supports all three without fragmenting truth?

Sample answer outline

Separate shared company metrics from team-specific operating metrics. Build a small set of certified cross-functional measures, then let each team have dashboards that answer their workflow questions against the same governed dimensions. Facilitate metric definition sessions where tradeoffs are explicit, especially around attribution, ownership, and time windows. Document which dashboard is authoritative for which question, and retire duplicates that disagree. Interviewers want evidence that the BI engineer can mediate ambiguity rather than simply build every requested chart.

Expect these follow-ups

• How do you push back on a VP asking for a misleading metric?
• What makes a dashboard worth retiring?
• How do you measure whether the reporting suite is actually being used well?

As asked

Tell me about a time you had to get finance and marketing to agree on a shared definition of a metric (like customer acquisition cost) when both teams had fundamentally different incentives.

Sample answer outline

Strong answer shows the candidate first listened to each team's perspective separately, then facilitated a joint session grounded in data rather than opinion. They proposed a shared base definition with optional add-back or adjustment fields that each team could include for their own reporting. They got sign-off from a finance or data governance owner and documented the final definition in the data catalog. The answer should show diplomacy without hiding behind indecision.

Expect these follow-ups

• Who ultimately had the final say on the definition, and how was that decision made?
• What would you have done if the teams remained deadlocked?

As asked

A senior product manager uses direct SQL access to Snowflake and sends an analysis to the VP of Product with numbers that are wrong because they did not apply a filter for cancelled orders. You discover this in a meeting. How do you handle it in the moment and after?

Sample answer outline

In the moment: raise the issue factually and without embarrassing the PM, provide the corrected number. After the meeting: follow up privately with the PM to explain the filter they missed, not as a criticism but as a knowledge-share. Then address the root cause: either the data model should exclude cancelled orders by default (put it in the certified mart table), or add documentation/a reminder comment in the warehouse that analysts must filter on order_status. Avoid gatekeeping self-serve access as the solution.

Expect these follow-ups

• What change would prevent this specific class of error for all self-serve users?
• How do you balance enabling self-serve SQL access with ensuring data accuracy?

As asked

You joined a company where every analyst had write access to production tables and there were no naming conventions or documentation standards. How did you introduce governance without alienating the team?

Sample answer outline

The candidate should show they started by understanding why the current state existed (speed, small team) and did not assume the old way was purely careless. They introduced governance incrementally: first naming conventions and a project-level .yml file, then read-only production access with a dev schema for everyone, then dbt tests on critical models. They brought the team along by framing each change as making their own work easier. They should mention what they did NOT try to enforce (avoiding governance for governance's sake).

Expect these follow-ups

• What was the first governance rule you introduced and why that one first?
• How did you handle a senior engineer who resisted the new standards?