6.3Best practices & pitfalls
The essentials in 30 seconds
- Structure: star schema (facts at the center, dimensions 1:*), cleaning at the source in Power Query, a dedicated and marked date table.
- Measure & secure: explicit measures (DIVIDE, VAR/RETURN) stored in « _Measures », tested RLS, sensitive data outside the report.
- Deliver: accent-color design, timed performance, publication in a team workspace and refresh that doesn't depend on your laptop.
Zineb's thread — the rule stuck on the wall
After a quarter, Zineb stuck her rule on the wall in Algiers: clean and proven first, fast afterwards. What mattered most wasn't one more brick, but three habits — star schema, explicit measures, tested RLS — and her biggest pitfall narrowly avoided: a report distributed to everyone while per-department security had not been verified.
Pick one single habit to anchor in your ritual, then move to the next.
The do's / don'ts of a reliable report
The difference between a makeshift build and a professional practice isn't in the visuals, but in the reflexes: a clean star schema, explicit measures, a color that means something, a secured, high-performing and truly refreshed report.
The model first
A well-built star schema makes DAX simple and performance correct. Conversely, staying in « single table » degrades everything: the star is not optional.
Meaningful rather than decorated
An accent color, neutrals, titles as statements: design serves the decision. Coloring every category in a rainbow makes the report pretty but mute.
✓ To do
- Model as a star: facts at the center, dimensions around, 1:* relationships — the foundation of everything.
- Clean at the source (Power Query) and create a dedicated date table, sorted and marked.
- Prefer explicit measures (DIVIDE, VAR/RETURN), stored in « _Measures » and formatted.
- Secure (tested RLS) and optimize on evidence (Performance Analyzer, in the order model → measures → visuals).
✗ To avoid
- Staying in « single table »: DAX and performance degrade — the star is not optional.
- Multiplying calculated columns when a measure would do; forgetting to format numbers.
- Coloring every category (rainbow) instead of a meaningful accent; neutral titles.
- Publishing an RLS without testing it, or distributing a sensitive report to too broad a workspace.
Ready-to-say script — checking your report before sending (30s)
« Clean star? Marked date table? Measures in « _Measures »? Zero useless columns? Meaningful accent? Tested RLS? Timed performance? If any no, I fix it before sending. »
To reread just before distributing a report.
Ready-to-say script — explaining your design (30s)
« One accent color, neutrals, titles as statements, two to four slicers: my page serves the decision, it doesn't decorate. »
One sentence to recall that design serves the decision.
Copy-paste template — healthy measures
Total Revenue = SUMX(Sales, Sales[Quantity] * Sales[UnitPrice])
Average Basket = DIVIDE([Total Revenue], [Sales Count])
Growth % = VAR py = [Revenue PY] RETURN DIVIDE([Total Revenue] - py, py)
Share = DIVIDE([Total Revenue], CALCULATE([Total Revenue], ALL(Products)))Copy-paste template — bad track to fix
-- BEFORE: frozen column, does not react to slicers
Column Revenue = Sales[Quantity] * Sales[UnitPrice]
-- AFTER: explicit reusable measure
Total Revenue = SUMX(Sales, Sales[Quantity] * Sales[UnitPrice])SME / small budget
One habit per quarter is enough: star schema + clean measures first, the rest afterwards.
SME Algiers
Adapt the rituals to the closings: a slot for formatted measures + tested RLS held every month.
Remote / hybrid
Discipline #1 at a distance: named source, tested RLS and refresh date written down, even for small reports.
Intercultural
Check local usages rather than assuming: an unusual date or number format should raise a flag everywhere.
48h challenge — apply tomorrow
Run your report through the sieve: zero single table, zero useless column, zero untested RLS. Fix the first gap found.
- I know the reflexes to adopt and the pitfalls to avoid.
- I checked my critical points (star, measures, tested RLS).
- I did the 48h challenge or scheduled its date.
Before going to production, which reflex best protects Zineb?
Golden rule: a star schema, explicit measures and a tested RLS — the rest (visuals, theme) only reveals this base.