What are the most asked Power BI interview questions for freshers?

What are the most asked Power BI interview questions for freshers?

Power BI fresher interviews are not a tour of every button. Panels ask you to explain a model you built: which tables you loaded, how they relate, what a measure calculates, and why a number on the page is correct. The concepts that carry those answers are Power Query versus DAX, a star schema, relationships and filter direction, and the difference between a measure and a calculated column. If you can open your file and change a slicer without breaking the total, you are close. If you can only name visuals, you are not.

SQL sits underneath a serious Power BI answer. A wrong grain in the table becomes a wrong chart. Use the SQL interview questions guide for joins and GROUP BY, and the data analyst interview questions guide for the wider analyst round. This page stays on the Power BI conversation.

What is the difference between Power Query and DAX?

Power Query is the preparation step. You connect to a source, clean columns, change types, remove duplicates, and shape tables before the model loads them. DAX is the calculation language inside the model. You write measures and, when you must, calculated columns that the report evaluates. A fresher who cleans text with a DAX calculated column on a huge table has often done work that belonged in Power Query, where it runs on refresh rather than in every visual. Say that split in one sentence: “I shape data in Power Query. I calculate results in DAX.”

Be ready to show one Power Query step you actually used. Changing a date type, splitting a name, or filtering test rows are all valid. Be ready to show one measure, not a screenshot of a pie chart. The measure is the interview. The pie chart is the decoration.

What is the difference between a measure and a calculated column?

A calculated column is computed row by row and stored in the table. It uses memory. A measure is computed when a visual asks, in the filter context of that visual. Totals, ratios, and anything that must change when a slicer changes belong in measures. A column is reasonable when you need a value on each row for a relationship, a slicer, or a grouping that truly lives on the row, such as a price band you will filter by. Freshers who write a calculated column for “total sales” and then drop it in a card will get a sum of a sum, or a number that ignores the slicer in a way they cannot explain.

The spoken test is: “If the user clicks Chennai, should this number change?” If yes, it is a measure. If it is a property of the row that stays the same, it might be a column, or better a column you created in Power Query. Say the question out loud in the interview before you name the feature. It shows you start from the user’s click.

What is a star schema, and why do interviewers want one?

A star schema puts a fact table in the middle and dimension tables around it. The fact holds the numbers and the keys: order amount, order date key, customer key. The dimensions hold the descriptions: customer name, product category, date attributes. You relate them on keys. You do not build a web of many-to-many links between every table because the export arrived that way. The star keeps filters predictable. A slicer on customer city filters the customer dimension, which filters the fact, which changes the measure.

The grain is the sentence you must be able to say. “One row in this fact is one order line.” If one row is sometimes an order and sometimes a payment, your sums will lie. Interviewers ask “what is one row?” more often than they ask you to recite Kimball. Answer with your file. If your college project was a single flat table, say you would split customer and date out before you trusted it with more data. That is a stronger answer than pretending you designed a warehouse.

How do relationships and filter direction work?

A relationship connects a column on one table to a column on another, usually a key. The filter usually flows from the one side, the dimension, to the many side, the fact. A city slicer then reduces the fact rows. Both-direction filtering can solve a model you painted into a corner, and it can also create ambiguous filters that are hard to explain. Prefer single direction and a star. If you used both-direction, say why, and say what broke when you tried the simpler model.

Inactive relationships appear when two date columns, such as order date and ship date, both want the calendar table. Only one relationship is active. The other is used inside a measure with USERELATIONSHIP. You may not have done this. If the interviewer describes it, the idea is enough: one active path, and an explicit function when you need the other date. Do not claim you use USERELATIONSHIP in every report.

What is filter context, in fresher language?

Filter context is the set of filters that apply when a measure runs: the slicers, the visual’s row, and any filters in the measure itself. A card with no slicer might show all sales. The same measure in a table visual, on the Chennai row, shows Chennai sales. You did not write two measures. The context changed. Row context is different. It appears when a calculated column or an iterator such as SUMX walks rows. CALCULATE is the function that changes filter context. You do not need every CALCULATE pattern. You need one: a measure that computes sales, and a second measure that computes last year’s sales by changing the date filter.

If you have not written CALCULATE, do not invent a formula. Say filter context is why the same measure shows different numbers in different visuals, and say you would learn CALCULATE to compare periods. Then show the measure you did write. Honesty plus one real measure beats a formula you cannot evaluate when they change the slicer.

Which DAX functions should a fresher be able to explain?

SUM and COUNT are the start, and COUNTROWS is often the clearer count of table rows. DISTINCTCOUNT counts unique keys, which is how you answer “how many customers” after a join multiplied rows. DIVIDE is safer than the slash when the denominator can be zero, because a blank is kinder than an error on a visual. CALCULATE, as above, changes filters. RELATED walks from the many side to the one side to fetch a column, and it is a sign your relationship works. If RELATED fails, the relationship is missing or the direction is wrong. That failure is a useful story.

Avoid a long list of time-intelligence functions you have never called. TOTALYTD and a date table are worth learning if your project has dates. A date table marked as a date table, with no missing days, is the honest prerequisite. Saying “time intelligence” without a continuous date table is how year-to-date numbers skip months and nobody notices until the panel asks.

How does refresh work, and what should you say about the gateway?

Import mode loads data into the model on refresh. The report is then fast, and it is only as fresh as the last refresh. DirectQuery leaves the data at the source and asks a query when the visual needs it. It is more current and easier to make slow. Freshers almost always use Import. Say so. Scheduled refresh in the service runs that import on a clock. A gateway is the bridge when the source sits inside a private network the service cannot see. If your project used a local Excel file, you did not need a gateway story. Say the file was local and a real deployment would need a shared source and a refresh plan.

Also say who owns failed refresh. A report that broke three days ago and still looks confident is worse than a report that shows the last success time. You can mention a simple habit: check the refresh history, and do not email a screenshot of a stale total.

What is row-level security, at the level freshers need?

Row-level security limits which rows a person can see, using a role and a filter. A regional manager sees one region. You test it with “view as role” before you publish. It is not the same as hiding a visual. A hidden visual is still in the model. A role filter removes rows for that user. If your college project had one user, say you did not implement roles, and say where you would add one if the sales file had regions. Do not claim a security design you never tested.

The follow-up is a measure that must still be correct under the role. If the role filters the fact, your measure should not need a second copy of the filter unless you are doing something special. Duplicate filters are how people get zeros they cannot explain. Trust the role, then test with a user who should see less.

How do you defend a visual and a dashboard?

A report page answers a question. A dashboard in the service can pin tiles from reports, but in fresher speech people say dashboard for the page. Clarify if they ask. Then defend one page: the question, the measure, the slicers, and the comparison. “Sales by category this month versus last month, with a city slicer” is a defence. “I used a gauge because gauges look executive” is not. Name one chart you refused. A pie with twenty categories fails. A bar sorted descending usually works. Say you labelled the units and you checked the total against the source query.

That last check is the analyst habit. Take the measure’s number, run the SQL or the Excel equivalent, and reconcile. If you never reconciled, the panel will not trust the colour scheme. Do it before the interview on one number. Bring the difference if you found one, and say whether the grain or the filter caused it.

How should you present the Power BI file?

Open the model view, not only the pretty page. Name the fact, the dimensions, and the grain. Show one relationship. Show one measure in the formula bar and read it in words. Change a slicer and say what the filter context just did. Then mention one mistake: a many-to-many you removed, a calculated column you moved into Power Query, or a total that double-counted because the join multiplied rows. The mistake is the proof you understand the tool.

Students building this skill can use Power BI training in Pondicherry and keep SQL in the same week. Placement conversations still include the HR questions on the most common interview questions page. A clean model will not answer “tell me about yourself.” Prepare both. After the offer, read the pay on the data analyst fresher salary page as a planning band, and read the annexure as the real number.

The day before, refresh the file, reconcile one total, and practise a three-minute model tour without touching a decorative visual. If you run out of time, spend it on the measure and the grain. Those two sentences carry the round.

What do one-to-many and many-to-many mean in a Power BI model?

One-to-many is the normal star relationship: one customer, many orders. The filter travels from the one to the many. Many-to-many appears when both columns repeat, such as a bridge between students and courses. It can be modelled with a bridge table so each side is one-to-many into the bridge. Dropping a many-to-many relationship because the click path allowed it will produce totals that depend on a filter path you cannot explain. If your file has one, draw it. If it does not, say you kept the model to one-to-many on purpose. Interviewers prefer a small clean model over a large ambiguous one you cannot narrate.

Cardinality also shows up as “why did my total double?” A relationship at the wrong grain, or a fact joined twice, duplicates rows before the measure sums them. Go back to SQL, count the rows, and fix the grain. The visual is the last place to look, not the first.

Why do data types and a date table decide whether the report is trusted?

A date stored as text will not sort in calendar order and will not support real time intelligence. A number stored as text will not sum. Fix types in Power Query, not with a visual workaround. A date table has one row per day, no gaps, and is marked as a date table. Your fact stores a date key that matches it. Without that, “last month” can mean a broken set of text values. If your college file was a single Excel sheet with dates as text, say you would convert them and introduce a date table before you compared periods. That plan is a valid fresher answer.

Also name one type you got wrong and what the visual did. Scientific notation on a phone number, or a sum of student ids, is the kind of mistake that teaches the lesson. Bring that story. It is more convincing than a list of data-type icons.

What changes when you publish, and what stays a desktop habit?

Power BI Desktop is where you model and design. The service is where you share, refresh on a schedule, and apply roles. A report that only works on your laptop is a college file, not a team report. Say what would need to change: a shared data source, a refresh, and a workspace other people can open. Row-level security, if you need it, is tested in the service or with view-as, not assumed. Apps and dashboards are sharing features. You can define them simply and admit you have only published a report, if that is the truth.

Performance at fresher scale is usually a model problem: too many columns imported, a calculated column that should have been a measure, or a visual that dumps a detail table of a lakh rows onto a page. Remove columns you do not use. Prefer a measure. Put detail in a drill-through page. You do not need DAX Studio in the interview unless you have used it. You need the habit of asking which column the page actually requires.

A three-minute model tour to rehearse

Set a timer. Minute one: fact table, grain, and the dimension you filter by. Minute two: one measure, read in words, then a slicer change and what the number did. Minute three: one mistake you fixed, or one type you would fix before you trusted a period comparison. Stop when the timer ends, even if you wanted to show another chart. The panel remembers a short tour that survived a question. They do not remember a twelve-minute click path through every page.

Then reconcile one number with the source. If the source is Excel, a pivot on the same grain should match. If the source is SQL, the grouped query should match. Write the two numbers down. If they differ, you have found the interview story and you should not publish the prettier wrong number. Bring the file on the laptop you will use, and open it once that morning so a missing gateway or a moved spreadsheet does not eat the first five minutes of the round.

A three-minute model tour to rehearse

Set a timer. Minute one: fact table, grain, and the dimension you filter by. Minute two: one measure, read in words, then a slicer change and what the number did. Minute three: one mistake you fixed, or one type you would fix before you trusted a period comparison. Stop when the timer ends, even if you wanted to show another chart. The panel remembers a short tour that survived a question. They do not remember a twelve-minute click path through every page.

Then reconcile one number with the source. If the source is Excel, a pivot on the same grain should match. If the source is SQL, the grouped query should match. Write the two numbers down. If they differ, you have found the interview story and you should not publish the prettier wrong number. Bring the file on the laptop you will use, and open it once that morning so a missing gateway or a moved spreadsheet does not eat the first five minutes of the round.

What to do if they ask for a measure you have not written

Talk through the filter context instead of inventing a formula you cannot evaluate. The card is all rows that survive the slicer. The matrix row is an extra filter. CALCULATE would change that filter if you needed last year. Then offer to write the simpler SUM you do know. A correct simple measure, tied to the grain, beats a time-intelligence function you cannot defend when they clear the slicer. End by telling them the one number you reconciled and the date of the file. Currency of the data is part of the answer. If the file is a week old, say so. A fresh wrong total is worse than an old total you have checked. Offer the check, not a promise that the visual is magically current.

Write the grain sentence on the first slide of your notes, next to the reconciled total, so you can say both before you touch a chart.

Related reading: SQL interview questions, Data analyst interview questions, Most common interview questions, Data analyst fresher salary.

These answers are a preparation guide for students and freshers, including learners at Asmorix in Pondicherry. A company may ask a different follow-up. This page is not a leaked paper and it is not a promise of a result. Say the short answer, show one example, and revise the same day any question you could not code or click.