Start a Power BI model answer by stating what one fact row represents. Relationships are easier to reason about when the grain is clear. A sales line, an order and a monthly total are different grains, even if all three tables contain an amount.

These Power BI interview questions use a fictional sales model. It is a learning fixture, not an actual business dataset. The expected totals below come from the listed rows and make no performance claim.

What does the small model contain?

Product
ProductId | Name
P1        | Notebook
P2        | Pen

Customer
CustomerId | Name
C1         | Ada
C2         | Ben

Sales
LineId | ProductId | CustomerId | Amount
L1     | P1        | C1         | 20
L2     | P1        | C2         | 30
L3     | P2        | C1         | 10

One Sales row is one sales line. ProductId is unique in Product, and CustomerId is unique in Customer. Create one-to-many relationships from each dimension to Sales, initially with single-direction filtering from dimension to fact.

What should a product filter do?

Selecting P1 should leave L1 and L2, giving an amount of 50. Selecting customer C1 should leave L1 and L3, giving 30. Selecting both P1 and C1 should leave only L1, giving 20. The unfiltered total is 60.

Use these small checks before building a complicated visual. If the results differ, inspect the keys, relationship activity and filter path before changing the measure.

Why must the one side be unique?

The relationship's one side represents a unique key. Duplicate product keys mean the data does not match that intended cardinality. Do not immediately change the relationship to many-to-many just to make an error disappear.

Find out whether the duplicates are accidental, historical versions or a genuine many-to-many requirement. A slowly changing dimension or a bridge design needs a deliberate key and grain, not a silent compromise.

When might a bridge be appropriate?

A customer can belong to several groups, and a group can contain several customers. A bridge can represent those memberships. Then define how group selection reaches the fact and what totals mean when memberships overlap.

A bridge does not automatically allocate one sale across groups. If a sale belongs to two groups, adding group subtotals may count it twice. Explain whether that overlap is intended or whether the business needs allocation weights.

Why not turn on bidirectional filtering everywhere?

It can create unexpected propagation or ambiguous paths. Start with the simplest direction that serves the reporting requirement. Add a more complex path only when you can explain and test its effect.

Also inspect inactive relationships, unmatched keys and data types. A measure can be correct while the relationship design sends it the wrong filter context.

Model checks to practice

  • Verify unique dimension keys.
  • State the fact grain in one sentence.
  • Check the four totals from the fixture.
  • Add one unmatched key and explain the resulting behavior.
  • Compare a dimension filter with a fact-column filter.
  • Test an actual many-to-many requirement separately.

A strong interview answer draws the filter path and predicts a result from a few rows. That gives the interviewer something concrete to inspect instead of a memorized definition of a star schema.

References and next practice

learn.microsoft.com — bi model

Browse interview topics · Practise a question in Interview Challenge · How to Answer Technical Interview Questions Out Loud