- Power BI turns tables into a model and a report. In the round, separate Power Query, the relationships, and the DAX measures, then show that one total matches the source.
Short answer: Power BI is Microsoft's tool for turning tables into a model and a report people can filter. In an interview, split the work into three parts. Power Query cleans and shapes the data. The model holds the tables and the relationships. DAX measures calculate the numbers on the report. If you mix those three up, the rest of the round gets harder.
This page is the spoken version of that round for a fresher or a junior analyst. It is not a copy of Microsoft's documentation. When a function name is in doubt, check DAX reference and the Power BI documentation. When the source file is still a workbook, the neighbour skill is Advanced Excel. When the source is a database, revise SQL interview questions as well.
What does a Power BI interview test?
The interviewer wants a report they could trust on Monday morning. They ask where the data came from, how you cleaned it, how the tables join, why a number is a measure and not a column, and what a user can filter. A pretty dashboard with a wrong total fails the round.
Have one report you can narrate. Name the fact table, two dimension tables, one relationship, one measure, and one visual that uses that measure. Then name one thing you refused to do, such as a bidirectional relationship on every join, or a calculated column that divides two numbers the user will filter later.
- Say import or DirectQuery, and why.
- Say the grain of the fact table: one row is one order line, one row is one attendance day, or whatever is true.
- Say who the report is for. A class teacher and a finance manager do not need the same page.
- If the file was a college dataset, say so. Do not describe a gateway and a workspace you have never opened.
What is Power BI?
Power BI is a set of tools. Power BI Desktop is where you build the model and the report on your machine. The Power BI service is where you publish that report so other people can open it in a browser. A gateway is the bridge when the service must refresh data that still lives on a machine or a network the cloud cannot see.
You do not need all three on day one. A fresher interview can be entirely about Desktop: Power Query, the model, and the canvas. Mention the service when they ask how someone else would open the report. Mention the gateway only if you have used a scheduled refresh against a source that is not already in the cloud.
What is the difference between Power BI and Excel?
Excel is a workbook. It is excellent for a table you clean, a formula you can point at, and a chart for one reader. Power BI is a model plus a report, built so many people can filter the same official numbers without copying the file. The preparation step is often still Excel or Power Query. The shared report is Power BI when the audience has outgrown one workbook.
| Question | Excel | Power BI |
|---|---|---|
| Where the work starts | A grid and formulas | A query, a model, and a report |
| Who refreshes it | The person who has the file | A published dataset on a schedule, when you have set that up |
| How a total respects a filter | SUMIFS, a PivotTable, or a formula you must design | A measure, evaluated in the filter context of the visual |
| When to stay | One analyst, one file, a table you can audit cell by cell | A report more than one team must filter |
Do not tell a hiring manager that Excel is finished. Many Power BI models start as a clean sheet. The skills of an Advanced Excel user are still the skills that make the source table honest.
What is Power Query?
Power Query is the step that runs before the model. You connect to a file, a folder, or a database. You remove junk rows, set types, split columns, and append months. Each action is a step you can refresh. You do not retype the cleanup when next month's file arrives.
In the interview, open the Applied Steps list in your head and say four of them in order. Example: promote the first row to headers, filter out a total row that was typed at the bottom, set the date type, append the branch files. That list proves you used Power Query rather than a one-time cleanup in the grid.
What is a star schema?
A star schema is a fact table in the middle and dimension tables around it. The fact holds the numbers and the keys: order amount, quantity, product id, customer id, date. A dimension holds the description you filter by: product name, customer city, month name. You relate the keys. You do not merge every description back onto the fact if you can leave it on the dimension.
Say the grain before you say the picture. "One row in Sales is one invoice line." If two rows could be the same line, the model is already wrong and no visual will save it. Interviewers who build reports listen for grain first.
How do relationships work?
A relationship connects a key on the dimension to the same key on the fact. The usual cardinality for a fresher model is one-to-many: one product, many sales lines. The filter direction is usually single, from the dimension to the fact. The product slicer then filters sales. Sales does not need to filter the product table for a normal report.
Both-direction filters feel powerful and cause totals that change for a reason you cannot explain. Use them only when you can name the visual that is wrong without them. Many-to-many relationships exist and they are a design you should justify, not a default. If the interviewer asks why a slicer does nothing, the first checks are: is there a relationship, is the cardinality right, and is the column on the visual the column you related.
What is the difference between a calculated column and a measure?
A calculated column is computed row by row and stored in the table. It is useful when you need a value on the row, such as a band you will slice by: "High", "Medium", "Low". A measure is a formula computed when the visual asks, in the filter context of that visual. Sum of amount, average of score, and year-to-date sales are measures.
The mistake is to store a ratio as a column. A column that divides profit by revenue on each row does not become the company ratio when you sum it. The measure DIVIDE of the two sums does. Say that sentence. It is the answer that separates someone who has dragged fields onto a canvas from someone who has checked a total.
What is filter context?
Filter context is the set of filters in force when a measure runs. A card with no slicer sees the whole table. The same measure in a matrix row for April, with a slicer on one city, sees April and that city. You do not pass the city into the measure as an argument. The visual and the slicers pass it.
Row context is different. It is the current row, which a calculated column has automatically and a measure does not, unless an iterator such as SUMX creates it. You do not need the full theory in a fresher round. You need to say why the same measure shows a different number in two visuals: the filters around it changed.
What does CALCULATE do?
CALCULATE evaluates a measure under a filter you change. You use it to remove a filter, to replace a filter, or to add one. A year-to-date measure and a measure that ignores a slicer both end up here. ALL removes filters from a table or a column. FILTER returns a table of rows that pass a test, and it is easy to write badly on a large fact table.
In the room, do not recite ten functions. Write one measure you used. Example: revenue year to date is CALCULATE of the revenue measure, with a date pattern such as DATESYTD, if your date table is marked as a date table. If you have not used time intelligence, do not invent the function names. Describe the card you did build.
Which DAX functions should a fresher be able to explain?
SUM adds a column. DIVIDE divides and lets you define the result when the denominator is zero, which is safer than the slash operator on a report. COUNTROWS counts the rows still in filter context. RELATED pulls a column from the one-side of a relationship into a row context, which is why it shows up in calculated columns more than in simple measures. DISTINCTCOUNT counts unique keys, such as customers who ordered, not order lines.
Know the difference between COUNT and COUNTA in the way the interviewer means "how many": rows, non-blank values, or distinct ids. Ask which one they want if the question is vague. That question is a sign you have shipped a report, not a sign you are stuck.
What is a date table?
A date table is a dimension with one row per day, a date key, and the labels you filter by: month, quarter, year. You relate it to the date on the fact. You mark it as a date table when you want time-intelligence functions to use it. A fact table's own date column is a poor substitute once you need a month with no sales to still appear, or a year-to-date that does not depend on whichever dates happened to be in the sales file.
Build the date table in Power Query or in DAX. Either is acceptable. Say that it is continuous, that it covers the years in the model, and that you did not type the months by hand on the visual.
What is row-level security?
Row-level security, RLS, is a filter on the model that depends on who is reading. A branch manager sees their branch. You define a role, you write a filter, and you test it with View as role before you publish. Publishing without that test is how a manager sees another branch's revenue.
RLS is not a visual-level filter you hide on the page. A user who can edit or who can reach the dataset another way is not protected by a hidden slicer. Say that plainly. If your college report had no roles, say you would add a role when the same report serves more than one audience.
What is the difference between Import and DirectQuery?
Import copies the data into the Power BI model on refresh. Reports are fast because they read the copy. The numbers are only as fresh as the last refresh. DirectQuery leaves the data in the source and sends a query when the visual needs it. The report is current and the source must answer quickly. A composite model mixes the two and is more than a first interview needs.
Choose Import for a fresher project unless the table is too large to copy or the business cannot wait for a refresh. Say the trade-off in one line: Import is fast and slightly stale. DirectQuery is current and only as fast as the database. Then say which one your file uses.
How does refresh work?
In Desktop, refresh reruns Power Query and reloads the model. In the service, a scheduled refresh does that in the workspace on a clock you set. If the file or the database is on a private network, a gateway must be online or the refresh fails. If the source is a file you uploaded by hand, there is nothing to refresh until you publish a new file.
Be honest about which of those you have done. "I clicked Refresh in Desktop after I replaced the CSV" is a correct fresher answer. Pretending you operated a gateway is not.
How do you keep a model fast?
Remove columns you do not show and do not filter. Set types in Power Query so a number is not text. Prefer a star schema to one wide table full of repeated names. Avoid a both-direction relationship you cannot explain. Prefer measures to calculated columns that exist only to be summed. If a visual is slow, look at the query and the cardinality before you add another visual.
A practical check: if the fact table has a column of comments nobody filters, drop it. High-cardinality text is expensive and it rarely belongs on the fact. Put the label on the dimension, or leave it out of the report.
How do filters and slicers interact?
A slicer is a visual the reader uses. A filter can sit on the visual, the page, or the whole report. Filters compose. A page filter for this year and a slicer for one city both apply. Edit interactions when one visual must not filter another, for example a target card that should stay constant while a chart filters.
Bookmarks save a view: which page, which slicers, which visibility. They are useful for a "reset" button. They are a poor place to hide a number you hope the user never finds. If a total must not be seen by a role, that is RLS or a separate report, not a bookmark.
Which visuals should you defend?
A column or bar chart compares categories. A line shows change over time. A table shows the rows a manager will export. A card shows one official number. A pie with many slices is hard to read. Choose the visual from the question the page answers, then stop. A page with twelve charts and no sentence about what changed is a decoration.
Title the page with the decision, not with "Dashboard". "Branch revenue this month" tells the interviewer you know the audience. Colour is for one highlight, not for a different colour on every bar.
How do you check that a number is right?
Pick one total and match it to the source. If the sales file says 1,000 rows and a known sum, the model should show that sum with no slicer applied. Then apply one filter you can compute by hand, such as one product, and match again. If those two agree, you can talk about the report. If they do not, the relationship or the filter is wrong and the design discussion can wait.
Write the check on the report or in your notes: source file, row count, unfiltered total, one filtered total. In the interview, offer those three numbers. It is the fastest way to show you do not trust a visual just because it rendered.
How do you explain a Power BI project in two minutes?
Say the question the report answers. Say the source. Say the grain. Say one relationship. Say one measure and the visual that uses it. Say the check you ran against the source total. Then stop. If they want DAX, open with the measure text and read it as a sentence: "Revenue is the sum of line amount, and this card is that measure filtered to the year on the slicer."
Do not apologise for a college dataset. A small file you can verify is a better interview than a large file you cannot explain. The habit you are proving is the same habit a team wants on a real sales extract.
Which mistakes should you avoid in the round?
- Calling every number a column, including the ones that must respond to slicers.
- Merging every lookup into the fact table because the relationship pane felt unfamiliar.
- Setting both-direction filters on every relationship "so the slicers work."
- Using a calculated column for a ratio and then summing that column.
- Hiding a slicer and calling it security.
- Importing every column in the database because the connector offered them.
- Describing a scheduled refresh and a gateway you have only read about.
If you do not know a function, say what the measure must do and that you would look up the exact DAX. Inventing TOTALYTD syntax in the room is worse than a pause.
How should you revise before the interview?
Rebuild one small model from a CSV you already understand. Four columns are enough: date, product, city, amount. Shape it in Power Query. Split product and city onto dimensions if you want to show a star. Write three measures: total amount, number of rows, and amount for one product using a filter you can explain. Put them on one page with a slicer. Check the unfiltered total against the file.
Then practise the two-minute talk. The classroom place to have someone challenge the total is Power BI training in Chennai. The Excel habits under the source file are in Advanced Excel training. The SQL under a database source is in the SQL interview set. This page is the script for the round, after the model exists.
What should you say if they ask which tool to choose?
Choose Excel when one person must audit a table and the audience is the file itself. Choose Power BI when more than one person must filter the same model and you can refresh it. Choose a database when the workbook has become the only copy of the business. You can say that without ranking the tools as fashion. The interviewer is listening for the audience and the refresh, not for a brand.
Close the round on the number you checked. A Power BI answer that ends with "the card matches the source total" is the answer that gets a second conversation.
What do you say about sharing the report?
In Desktop the report is a file on your machine. After you publish, it lives in a workspace. People you add can open it in the browser. An app is a packaged set of reports you give to a wider group so they do not browse the workspace itself. You do not need to have published if the interview is about the model. You do need to know the difference between the file you built and the copy colleagues would open.
If they ask who owns the dataset, say the person who can change the measures and the refresh. Readers of the report should not be that person by accident. A fresher who can say "I would publish to a workspace, give the manager a read link, and keep edit rights with the analyst" has understood the last mile of the tool.
