Power BI Interview Questions

You can build a working report and still stall on DAX questions, because interviews probe the evaluation model rather than the function list. This section covers the DAX questions that appear in nearly every Power BI interview, and what a strong answer contains.

• What is the difference between a measure and a calculated column? A calculated column is computed row by row during refresh and stored in the model, consuming memory. A measure is computed at query time against the current filter context and stores nothing. Say the memory and the timing, not just "one is dynamic".
• Explain row context and filter context. Row context is the current row during an iteration, filter context is the set of filters applied by slicers, visuals and DAX. Most confusing results come from expecting one and getting the other.
• What does CALCULATE do? It evaluates an expression in a modified filter context, and it converts row context into filter context. That second half is what interviewers listen for.
• When do you need an iterator such as SUMX? When the calculation must happen per row before aggregating, for example price times quantity where no stored column holds the product.
• What does ALL do, and why pair it with CALCULATE? It removes filters, which is how percent-of-total measures are built.
• Explain RELATED versus RELATEDTABLE. Direction matters: one crosses to the one side, the other returns rows from the many side.

A useful habit for the whole round: when a DAX answer starts to wander, return to context. Almost every follow-up resolves to which context is active and what changed it.

Modeling questions separate report builders from people who can own a dataset. The expected vocabulary is dimensional.

• Star schema or snowflake, and why? Star, for query performance and simpler DAX, with denormalized dimensions. Snowflake only where a dimension is genuinely large or shared. A flat single table is the answer that ends the round early.
• What is a fact table and what is a dimension table? Facts hold measurable events at a stated grain, dimensions hold the attributes you slice by.
• Why does relationship cardinality and cross-filter direction matter? One to many with single direction is the default for a reason; bidirectional filtering introduces ambiguity and can produce wrong totals.
• How do you handle a many-to-many relationship? A bridge table, and an explanation of what the totals mean afterwards.
• Why do you need a dedicated date table? Time intelligence functions require a contiguous marked date table; without one, year-to-date and prior-year measures misbehave.
• Power Query or DAX for this transformation? Shape in Power Query where possible, because transformations there happen once at refresh rather than on every query.

A BI analyst interviewing for a retail reporting team was asked why a sales total looked correct by product but wrong by region. The answer the panel wanted was a modeling one, a bidirectional relationship creating an ambiguous filter path, not a DAX rewrite. Diagnosing at the model layer first is the instinct being tested.

Other role-specific banks are collected under interview questions by role.

These questions appear in BI developer and analytics engineer rounds, and they are where candidates who only build in Power BI Desktop run out of material.

• Import, DirectQuery or Composite? Import is fastest to query and limited by memory and refresh windows. DirectQuery keeps data current and pushes load onto the source, with DAX limitations. Composite mixes them. Give the tradeoff and one condition that decides it.
• What is an on-premises data gateway for? Connecting cloud services to sources behind a corporate network, for scheduled refresh and DirectQuery.
• How does incremental refresh work? Partitioning by a date range so only recent partitions reload, configured with the range parameters the feature expects.
• How do you implement row-level security? Roles with DAX filter expressions on dimension tables, with dynamic security using the logged-in user function to look the user up in a mapping table.
• Workspace, app, dataset, report: how do they relate? Basic service literacy, and an easy question to fail by having only ever worked in Desktop.
• How do you improve a slow report? Reduce visuals per page, prefer measures over calculated columns, cut cardinality on relationship columns, and inspect with Performance Analyzer.

If your experience is entirely in Desktop, say so plainly and describe what you would need to learn. Interviewers accept a stated gap; they do not accept invented deployment experience that unravels one follow-up later.

Power BI answers fail out loud in a specific way: the candidate knows the distinction but explains it in circles, because they have only ever demonstrated it by clicking. Reading a question bank produces recognition, and recognition collapses when a stranger asks and waits.

Pick the five questions above that you would least like to be asked and answer each aloud in under ninety seconds, without a screen. If you cannot say what CALCULATE does without drawing, you are not ready to say it in a panel. Running the same prompts against an AI interviewer that asks follow-ups is closer to the real conversation than rereading notes, which is what mock interview mode is built for.

Aaron Cao, founder of SubcueAI, designed the practice mode around rehearsal rather than content delivery, because question lists are abundant and repetitions of speaking are not. In a live round the desktop app and the browser extension Side Panel can surface structure while the interviewer is talking, which helps most on material you already know. Setup steps are on the tutorial page.