Power BI Interview Questions 2026: 12 Real Questions, Model Answers and the DAX They Expect

Your interviewer shares their screen, opens a flat 400,000 row sales extract, and asks how you would model it. Three minutes later they ask why your margin shows 25% when finance says 10.3%. That question has ended more Power BI interviews than every syntax quiz put together, and it is not really about DAX.

Below are 12 questions in the order a real round runs, with the answer to give and why it works. One example runs throughout: a two person analytics team at a mid-size Indian insurer, 14 million policy rows in Azure SQL, one Power BI Pro workspace, and a sales head who opens the report at 9:05 AM and complains when it is stale.

What Power BI Interview Questions Actually Test in 2026

Nobody is checking whether you memorised the function list. They want to know whether you can build a model that stays correct when someone adds a slicer, explain why a report is slow, and say no to a bad request.

One honest caveat. This interview is a weak predictor of whether you will enjoy the job, because much of the real work is chasing a column that got renamed upstream. If you would rather write code than defend numbers, a live Microsoft Fabric data engineering program covering DP-700 points at the engineering side instead.

Power Query and Data Prep Questions (Q1 to Q3)

Q1. What is query folding, and how do you check it is happening?

Folding is Power Query pushing your steps back to the source as one native query, so the database filters instead of your laptop. Right click the last applied step and choose View Native Query; greyed out means folding broke at or before that step. Usual culprits: Table.Buffer, an added index column, a custom column with no SQL equivalent.

Q2. Where should a transformation live: the source, Power Query, or DAX?

In that order of preference. Power Query runs once per refresh and compresses into the model, a calculated column also materialises but cannot fold, and a measure recalculates per visual. For the insurer, policy_year belongs in the SQL view. Use DAX only when the result must react to what the user selected.

Q3. What does incremental refresh need before you can configure it?

Two parameters named exactly RangeStart and RangeEnd, both of type Date/Time, used in a filter step on a date or date/time column, against a source that folds. Microsoft's documentation on configuring incremental refresh states you cannot configure the policy without them.

// Power Query (M): the filter that makes partitioning possible
= Table.SelectRows(
    Source,
    each [OrderDateTime] >= RangeStart
     and [OrderDateTime] < RangeEnd
  )

Point at the half-open interval deliberately: >= on one side and < on the other is what stops a row landing in two partitions and being counted twice.

Data Modelling Questions: Star Schema, Relationships and Filter Direction (Q4 to Q6)

Q4. Why a star schema rather than one wide table?

Because filters have to travel somewhere. A star gives every filter one unambiguous path from dimension to fact, which keeps measures correct and lets the columnar store compress repeated dimension values hard. One wide table makes every slicer scan the largest object in the model. If that is shaky, our star schema walkthrough is the fastest fix.

Q5. Explain filter context as if I were the sales head.

This is the question. Answer with the mechanism, not a definition.

Out loud: the slicer filters the dimension table, the relationship carries that filter down to the fact table, the measure is evaluated against only the surviving rows, and CALCULATE is the one function that can change which rows survive before aggregation.

Q6. When do you enable a bidirectional relationship, and why do you need a date table?

Bidirectional: almost never, and say it that plainly. It creates ambiguous paths the moment a third table joins, and the symptom is a number that changes depending on which visual it sits in. Keep the relationship single direction and open the gate inside one measure with CROSSFILTER, so the side effect is scoped; the honest exception is a bridge table for a genuine many-to-many. The date table is separate because time intelligence needs a contiguous date column marked as the model's date table, and your fact table's order date has gaps on every weekend and holiday. Generate it with CALENDAR, extend it to the end of the fiscal year, and mark it, or SAMEPERIODLASTYEAR quietly returns blank.

Power BI DAX Interview Questions: Filter Context, CALCULATE and Time Intelligence (Q7 to Q8)

The Power BI DAX interview questions that separate candidates are never about syntax. They are about which number is correct.

Q7. Finance says margin is 10.3%, the report says 25%. Find the bug.

Reproduce it in two minutes with two rows, where the source already carries a MarginPct column:

AVERAGE over the percentage column returns 25%, because it weights a one thousand rupee policy exactly as heavily as a one lakh policy. Aggregate the parts and divide once:

-- Wrong: averages percentages, ignores the weight of each row
Margin % (wrong) = AVERAGE( factSales[MarginPct] )

-- Right: sum the numerator, sum the denominator, divide once
Margin % =
DIVIDE(
    SUM( factSales[Margin] ),
    SUM( factSales[Revenue] )
)

Extra marks for DIVIDE over the forward slash, and for saying why: it returns blank on a zero denominator instead of an error that breaks the visual.

Q8. Write a year-on-year measure, explain CALCULATE, then tell me how it fails.

CALCULATE evaluates its first argument in a filter context you have modified, and its filter arguments replace existing filters on the same column rather than adding to them. That is why ALL and ALLSELECTED exist as escape hatches, and why you wrap a filter in KEEPFILTERS when you want to add rather than replace.

Sales YoY % =
VAR CurrentSales = [Total Sales]
VAR PriorSales =
    CALCULATE(
        [Total Sales],
        SAMEPERIODLASTYEAR( dimDate[Date] )
    )
RETURN
    DIVIDE( CurrentSales - PriorSales, PriorSales )

It returns blank if the date table is not marked, it flatters or damns a partial current month because the prior period is complete, and it breaks on a fiscal year that does not start in January unless you use DATESINPERIOD against your own fiscal columns. Name the partial period problem before they ask: that is the bug that reaches the sales head. Filter context and context transition reward live practice with someone correcting you, which is what the Power BI and PL-300 blocks of 360DT's live Data Analyst course drill in the weekend sessions rather than leave to video.

Also read: SQL interview questions for data analyst roles, with 20 model answers, because most Power BI rounds have a SQL round attached.

Performance, Refresh and Licensing Questions (Q9 to Q11)

Q9. A report takes 20 seconds to load. How do you diagnose it?

Record an interaction in Performance Analyzer and read the three numbers per visual: DAX query, visual display, other. If DAX query dominates, copy it into DAX Studio and check whether the storage engine or the formula engine is working; formula engine heavy usually means an iterator over a large table, or a FILTER you could have passed as a CALCULATE argument. If visual display dominates, the page has too many visuals and the fix is editorial. On the model side, drop the high cardinality identifier you never aggregate: a policy number with 14 million distinct values can cost more than every numeric column combined.

Q10. Which limits do you keep in your head?

Two carry nuance worth volunteering. Microsoft's refresh documentation notes the eight slot limit covers REST API refreshes too, while a manual refresh you click in the service is not counted; and the PPU FAQ is explicit that 100 GB is an upper bound rather than a guarantee, because memory is reserved for refresh and queries at once.

Back to the insurer: eight slots cover a 9:00 AM refresh, until the team wants afternoon top-ups on three models in one workspace. The right answer is not "upgrade to Premium" but "move the heavy model to incremental refresh so the morning run finishes in four minutes instead of fifty." Cost discipline is an interview answer too, as in our guide to the AWS mistakes that quietly double a bill.

Q11. How does row-level security work, and what does it not protect?

You define roles with a DAX filter on a table, usually comparing a column to USERPRINCIPALNAME, then assign users or security groups to that role in the service. What it does not do: it filters rows, not columns, so a measure aggregating a hidden column is still readable; it does not apply to anyone with Admin, Member or Contributor rights on the workspace; and it does not survive a user downloading the .pbix. Volunteering that last part usually ends the security questioning in your favour. Governance of this kind now sits beside Copilot administration, the territory of the AB-900 Copilot and agent administrator course.

Power BI Developer Interview Questions: Q12, the Scenario That Decides the Offer

Senior Power BI developer interview questions almost always end here. The sales head wants every field from the policy system on one page, refreshed hourly. The wrong answer is to build it. The second wrong answer is to refuse.

Ask what decision the page supports, because "every field" is a symptom of not trusting the numbers. Propose one summary page with three or four measures plus a drill-through detail page. Then handle the hourly ask with arithmetic rather than opinion: eight slots a day on Pro means hourly needs Premium or a capacity, so either the business case justifies the licence or the refresh lands at the two or three times people actually look. Finally, offer the test: if nobody opens the drill-through page in three weeks, you delete it. That move gets remembered, because most candidates never volunteer a way for their own work to be judged.

Power BI Interview Questions for Freshers, and the Certification Question

Under a year of experience means fewer limit questions and more "show me your work". The Power BI interview questions for freshers run: walk me through a report you built, why did you model it that way, what was hardest, what would you do differently now. The last one is the only one that matters. "I built 12 calculated columns because I did not understand measures, and I would replace nine of them now" beats any polished success story.

Two 2026 specifics are worth knowing. The Power BI October 2026 feature summary on the Microsoft Fabric community blog added custom visual creation through natural language prompts in the report authoring skill, so some interviewers now hand you DAX that Copilot wrote and ask you to explain it. Reading measures you did not write is the skill the tool shifted demand toward, and building the systems that generate them is a different job again, which is what the AI Engineer course covers through RAG and agents. The same summary carries a deadline: Desktop builds from March 2026 or earlier can no longer save and share files to OneDrive and SharePoint, because the old file picker was deprecated. Update your practice machine before an assessment round.

On certification, have a position. The PL-300 skills outline dated April 2026 splits into four domains, with Prepare, Model, and Visualize and analyze each weighted 25 to 30% and Manage and secure Power BI at 15 to 20%; the pass mark is 700 on a 1,000 point scale. My read: it does not get you hired on its own and is not worth the fee with no report to show, but it earns its money once you have one project and need a credential that survives a resume screen. Structured prep for it sits inside the 10 week Data Analyst program, and the certifications overview shows where it sits beside DP-700 and AI-200.

A 14 Day Prep Plan Before Your Power BI Interview

With two weeks and about 20 hours, spend them in this order rather than rewatching a course.

If nobody is available to interrupt you, a free 360DT webinar gets a trainer to look at your model and a demo class forces you to answer out loud. Also read: the Power BI tutorial that builds a first dashboard from a CSV in eight steps, if you need the project before the prep.

In your position I would pick one dataset I cared about, model it properly, and get so fluent on it that any question could be answered from that one model. Interviewers do not want breadth from someone with two years of experience; they want one example you have thought about harder than they have. Spend the fortnight on filter context and the limits table, skip the function trivia, and walk in ready to be wrong out loud about a measure. For that practice with someone correcting your DAX live, start with the Data Analyst course and its PL-300 track, or sit in on a free webinar first.

Related guides

• PL-300 Exam Prep 2026: A 6-Week Study Plan if the certification question landed and you want the structured route.
• Power BI Developer Salary in India 2026 for the band to quote when they ask about expectations.
• Power BI vs Tableau in 2026 for the "why not Tableau?" question that shows up in product companies.
• Microsoft Fabric Tutorial for Beginners 2026 because more Power BI roles expect you to know where the lakehouse under the report lives.
• Data Analyst vs Data Engineer in 2026 if the caveat earlier landed and you are rethinking the role.

Frequently asked questions

About this guide. 360 Digital Transformation is an Anthropic Partner and a Microsoft Authorized Training Partner. Other certification bodies, vendors and employers named here are not affiliated with us. Tools and versions change quickly; commands and figures cited were checked on 25 October 2026.