Why most Power BI reports fail

What you'll do

Redesign a synthetic sales-report outline around one reader decision. Use the 3-30-300 hierarchy to distinguish orientation, explanation and investigation. You need a readable semantic model and F3 or equivalent modelling knowledge. This is a design lab; no live tenant or report deployment is required.

Start with the unanswered question

A report can contain correct numbers and still fail its reader. Twelve equally prominent cards leave no obvious starting point. An impressive visual can hide the period, baseline or business action. A screenshot can look finished while a date axis is sorted alphabetically.

Write the decision before drawing a chart: "Which territories need a sales review this month?" Then identify the primary outcome, the comparisons that establish context and the detail needed for follow-up. If you cannot name an action, you may have an inventory of metrics rather than a page with a purpose.

Reader: regional sales lead
Decision: where to investigate this month
Orientation: sales result, period and comparison
Explanation: trend and ranked territory contribution
Investigation: selected territory and supporting records

These are synthetic teaching choices, not claims about a real customer or proof that territory differences cause the result.

Use 3-30-300 as a hierarchy

The approximate 3-30-300 rule describes layers of attention. It is not a promise that every reader completes a task in three, thirty or three hundred seconds. Use it to ask whether the report supports a quick orientation, a longer explanation and a deeper investigation.

A Power BI report can span these layers across pages. It is distinct from a service dashboard. Do not force every explanatory table onto the overview merely to make that one page self-contained.

Give attention a clear path

The spine's default schematic starts with a calm canvas, a clear header, contextual KPI cards and one dominant analytical visual. Supporting ranking and an action queue lead toward detail. Position and size establish priority; colour carries defined meaning rather than decorating every panel.

The illustration uses fictional sample values. Read its structure, not its numbers. Identify the primary outcome, the dominant analysis and the route to supporting detail. The actual report requires qualified measures and appropriate native mappings; a sample mock does not establish that a generated report exists.

Default KPI slots are an outcome, two actionable drivers and a guardrail. Show fewer when the model supports fewer. Repeating actual sales, YTD sales, previous-year sales and sales growth as four unrelated headline metrics exaggerates the variety of evidence. They may share the same underlying lineage.

Run a paper review

Sketch an overview and one deeper page. Label every panel with a question, a measure, a unit, a scope and a route to detail. Remove any panel that cannot explain its purpose. Keep infrequent filters in the native Filters pane; the header normally carries period and principal business scope, with at most a third slicer.

Test the sketch using a failure case: the baseline is unavailable. The page should say so and retain the current result, rather than fabricate a favourable delta. Next remove valid time history. Replace the hero trend with a suitable ranked comparison rather than inventing chronological observations.

Your expected output is a two-page rationale and a list of evidence still needed. The v2 archetype, background and theme CLI integration is Preview. This course teaches the spine's default schematic; it does not promise an automatically validated storytelling system or a generally released composition command.

Check your understanding

1. What decision does your overview help one reader make?
2. Why is 3-30-300 a hierarchy rather than a timed acceptance guarantee?
3. How should the design change when comparison or trend evidence is missing?