Define what a row represents

A visual can look excellent while telling the wrong story. Before choosing colours, define your data grain: does a row represent an order, an order line or a monthly total? Mixing grains can lead to double counting.

Then write down metric definitions. Does revenue include returns? Which date determines the reporting month? Agree on these business rules before creating measures.

Separate events from their context

A star schema separates facts, such as sales lines, from descriptive dimensions: product, customer and calendar. Relationships should reflect available keys and table granularity.

Check uniqueness on the dimension side, unmatched values and filter direction. Adding a bidirectional relationship to remove a symptom can introduce ambiguity elsewhere. Understand the filter path first.

Test measures under different filters

Do not validate a measure only at grand-total level. Test one product, a period, a combination of segments and a no-data case. For a margin ratio, verify the calculation from totals rather than averaging row-level percentages.

Keep a small dataset whose expected results can be calculated manually. It becomes a reference when the model changes and prevents a plausible chart from being mistaken for proof of correctness.

Distribution is part of the project

A shared report also needs an access, refresh and maintenance strategy. Document its source, connection identity and what to do when refresh fails. Sharing options depend on licences and the environment.

If you use row-level security, test the actual roles. In a Power BI workspace, RLS applies to Viewers, not to Admin, Member or Contributor roles. Testing only with an editing account is therefore insufficient.

References

Back to the blog →