Microsoft Fabric
Power BI Visualization in Microsoft Fabric: From Semantic Model to Effective Report
How to turn a governed Fabric semantic model into a Power BI report people act on: page hierarchy, visual choice, KPIs, drillthrough, accessibility and performance.
On this page
Most of a Fabric platform is invisible to the people it serves. Pipelines, notebooks, Delta tables and semantic models exist so that someone, somewhere, can look at a page and decide what to do next. For most organizations that page is a Power BI report.
This guide is for data engineers, analytics engineers and developers who understand data well but have spent less time on the visualization side. It is not a tour of every Power BI feature. It follows the analytical serving path to its end:
Fabric data → Gold layer → semantic model → Power BI report → a user’s decision.
The goal is a report that answers a question quickly, on top of logic that lives in the right place.
Where Power BI fits in a Fabric architecture
In a medallion architecture, Power BI sits at the very end of the chain:
- BronzeRaw, as ingested
- SilverCleaned and conformed
- GoldBusiness-ready tables
- Semantic modelMeasures, relationships, security
- Power BI reportPages, visuals, interactions
- Business usersA decision
The most important architectural rule is the simplest: visualization is the consumption layer, not the data-processing layer. A report should not clean data, join raw tables, fix duplicates or redefine what revenue means. Each of those belongs upstream: data quality in Silver, business-ready shapes in Gold, metric definitions in the semantic model. A report that compensates for missing upstream work becomes slow, fragile and impossible to keep consistent with the next report.
When the layers below are healthy, report design becomes what it should be: deciding what to show, in what order, to whom.
Start with the question, not the visual
The most common way to build a report is also the least effective:
“We have these fields. What chart can we make?”
That approach produces pages full of visuals that are technically correct and practically useless. A better starting point is:
“What decision does this person need to make, and what do they need to see to make it?”
A few realistic examples:
| Decision | What the user needs to see |
|---|---|
| Are we on track this month? | Revenue against target, with the trend so far |
| Is demand changing? | Order volume over time, compared with the same period last year |
| Where do we need to act? | Variance against target by region, largest gaps first |
| Which region is underperforming, and why? | Regional performance, then the products or customers driving it |
| What needs attention today? | Operational exceptions: late orders, failed payments, stock-outs |
Write these questions down before you open Power BI Desktop. Each one becomes a candidate for a page, a visual or a drillthrough path. If a visual does not help answer one of them, it probably does not belong on the page.
Semantic model before report design
A report is only as consistent as the model underneath it. In Fabric, each Power BI report connects to a single semantic model. That model should already contain:
- governed measures, such as Total Sales, Order Count and Average Order Value, defined once
- relationships between facts and dimensions, so report authors never have to know which keys join which tables
- business names, formats and descriptions, so fields are self-explanatory
- hierarchies, such as Year → Quarter → Month or Category → Product
- security: row-level and object-level rules applied to every query
When report authors reconstruct business logic inside visuals, through visual-level calculations, ad hoc filters or local measures duplicated across reports, two reports eventually disagree about the same number. When they consume governed measures, every report inherits the same definition, and a correction in the model reaches every page.
The division of work is clear: the semantic model defines what a number means; the report decides how to show it. The semantic model guide covers modeling, Direct Lake, Import and DirectQuery in depth.
Choosing the right visual
Every visual answers a type of question. Choose by the question, not by what looks interesting.
| Visual | The question it answers | Use it when |
|---|---|---|
| KPI card | “Where are we now?” | One headline value, ideally with a target or comparison |
| Line chart | “How is this changing over time?” | Trends across continuous time: days, weeks, months |
| Bar chart (horizontal) | “Which items are biggest or smallest?” | Ranking categories, especially with long names |
| Column chart (vertical) | “How do periods or a few categories compare?” | Discrete periods or a small number of categories |
| Table / matrix | “What are the exact values?” | Detail, lookup and exceptions, usually after filtering |
| Scatter chart | “Is there a relationship between two measures?” | Correlation or outliers, such as discount vs margin |
| Map | “Where is it happening?” | Only when geography itself matters to the decision |
Three practical rules cover most cases:
- Time goes on a line. If the x-axis is a date, a line chart is almost always clearer than columns or a table.
- Rankings go on bars. Sort descending, keep the categories readable, and show the top N rather than everything.
- Maps need a geographic reason. A map of sales by country is attractive, but if the question is “which region is lowest?”, a sorted bar chart answers it faster and more precisely.
Avoid visuals that look rich but hide the answer: pie charts with many slices, 3D effects, dual axes that suggest a false relationship, and gauges that use a lot of space to show a single number.
Designing an effective report page
A good report page has a reading order. Users scan from the top left, so place information by priority:
- Primary KPIs: the handful of numbers that answer “where are we?”
- Trend: how those numbers got here.
- Drivers or breakdown: which regions, products or customers explain the movement.
- Detail: the exceptions or rows that need action.
- Primary KPIsWhere are we now?
- Revenue
- Orders
- Margin
- TrendHow did we get here?
- Revenue trend vs target
- DriversWhat explains the movement?
- By region
- By product
- DetailWhat needs action?
- Exceptions
This structure also limits density. Microsoft’s own optimization guidance recommends limiting the number of visuals on a report page to what is necessary, and using drillthrough pages and report page tooltips to provide more detail instead of crowding the page. Every extra visual competes for attention and sends its own queries.
Keep a consistent layout across pages: the same position for slicers, the same title style, the same color meaning. Users learn a report once and reuse that knowledge on every page.
Slicers and filters
Slicers let users focus the page on the context they care about. The useful ones are usually few:
- Date: a period selector or relative date range, almost always needed
- Region: for organizations that manage by geography
- Product or category: when the audience owns a product line
- Account or customer: on pages designed for account managers, not on executive summaries
Common mistakes:
- Dozens of slicers. If the page needs ten controls, it is probably several pages.
- Duplicated controls. The same filter as a slicer, a page filter and a visual filter confuses users about what is applied.
- Configure-before-use pages. Users should learn something the moment the page opens. Set sensible defaults, such as the current month or all regions, so the first view is already meaningful.
Remember that slicers, page filters and report filters are a user-experience tool, not a security mechanism. Data restrictions belong in the semantic model and the data layer.
Drillthrough
Drillthrough lets users move from a summary to a detail page that is automatically filtered to the item they selected. They right-click a data point or use a drillthrough button, and Power BI opens the target page with a back button.
Drillthrough is better than adding more information to the primary page when:
- the detail is needed only for some items, not all of them
- the detail has a different shape, such as order lines instead of monthly totals
- adding it to the summary would push the primary answer out of view
- Executive summaryKPIs, trend, regions
- drillthrough on a regionRegional detailProducts and customers in that region
- drillthrough on a customerCustomer or order detailOrders, exceptions, history
Design each drillthrough page around the selected item: show its name prominently, so users always know which context they are in, and include only visuals that make sense for a single region, product or customer. Cross-report drillthrough can link reports in the same workspace when the field names and data types match.
Tooltips
Default tooltips show the values behind a data point. Report page tooltips go further: a small report page, set as a tooltip, appears on hover and is filtered to the data point under the cursor. A tooltip on a regional bar might show that region’s trend and top products.
Use them for supporting context that helps interpret a value: a mini trend, a comparison or the drivers behind it. Keep them small and readable at a glance.
Two limits matter. Report tooltips are not interactive: users cannot click, scroll or use slicers inside them, so use drillthrough when users need to explore. And Microsoft notes that screen readers cannot read report tooltip content, so never put essential information only in a tooltip.
Bookmarks
A bookmark captures the state of a report page: the current page, filters and slicers, sort order, drill location, cross-highlighting and object visibility. Report authors create report bookmarks; viewers in the Power BI service can also save personal bookmarks.
Appropriate uses:
- Alternate analytical views: switch between “by region” and “by product” on the same page.
- Guided storytelling: step through a sequence of states in a presentation or review meeting.
- State changes: show or hide a details panel, or reset filters to a known starting point.
A warning: bookmarks combined with buttons and hidden objects can be used to build application-like navigation, with menus, toggles and pop-ups. A little of this helps. A lot of it becomes hard to maintain, because every new visual must be added to the right bookmarks and visibility states. It is also hard for keyboard users, and Microsoft’s accessibility checklist explicitly asks you to test bookmark navigation with a keyboard. Prefer report pages and drillthrough for navigation; reserve bookmarks for genuinely different views of the same content.
KPI design
A giant number alone is not a KPI. “Revenue: 4.2M” does not tell anyone whether to celebrate or worry. A useful KPI shows:
- the current value: revenue this month
- a target or benchmark: the plan, budget or same period last year
- the variance: the gap to target, in absolute and percentage terms
- the trend: the direction of travel, often as a small sparkline or a trend line below
- context: the period and scope, such as “October, all regions”
The variance should come from the semantic model, not be calculated in each visual:
Revenue vs Target = [Total Sales] - [Sales Target]
Revenue vs Target % = DIVIDE ( [Revenue vs Target], [Sales Target] )
When you use color to signal good or bad, pair it with something that does not depend on color: a sign (+ / −), an arrow or a label such as “Below target”. That makes the KPI readable for everyone, including in a printed or exported copy.
Accessibility
Power BI includes accessibility features such as keyboard navigation, screen reader support, high contrast support and a screen-reader-friendly data table view. Some of them need the report author to configure them:
- Color contrast: Microsoft’s guidance follows WCAG and asks for a contrast ratio of at least 4.5:1 between text and background.
- Color is never the only signal: use labels, icons, markers or patterns as well. Turn on markers with different shapes for multi-series line charts.
- Alt text: add alt text to every visual that conveys meaningful information. It can be dynamic, driven by a DAX measure.
- Logical reading order: set the tab order in the Selection pane to match the visual order, and remove decorative shapes from it.
- Keyboard usability: test the report with a keyboard only, especially pages that rely on bookmarks or buttons.
- Restrained visual density: fewer visuals, clear titles without jargon, and data labels only where they help.
Accessibility is not a separate task at the end. Most of it follows naturally from the same principles that make a report clear for everyone.
Performance
Report performance is an engineering problem with a visual symptom. Every visual sends at least one query to the semantic model, so slow reports usually combine too many visuals, expensive measures and a model that makes them harder than necessary.
From an engineering perspective:
- Limit the visual count. Fewer visuals means fewer queries and less rendering. Move secondary detail to drillthrough pages and tooltips.
- Avoid overly granular visuals. A table showing millions of rows by default is expensive and unreadable. Filter it, for example with a Top N filter, or show it only on a drillthrough page.
- Keep measures efficient. Complex iterators over large tables, deeply nested logic and measures that recalculate the same expression repeatedly all add up. Optimize them once, in the model.
- Respect the model design. A star schema with single-direction relationships is easier and faster to query than wide flattened tables and bidirectional filters.
- Understand the storage mode. Import and Direct Lake answer from the in-memory engine; DirectQuery sends queries to the source. For Direct Lake, healthy Delta tables matter as much as the report itself.
- Measure, don’t guess. Performance Analyzer, available in Power BI Desktop and when editing a report in the Power BI service, records how long each visual takes and splits the time into DAX query, DirectQuery, visual display and other work. You can copy a visual’s DAX query and run it in DAX query view.
A useful habit: record a Performance Analyzer session before every significant release, and treat a sudden jump in DAX query time as a model problem, not a visual one.
Report vs dashboard vs semantic model
These three terms are often confused. Current Microsoft terminology defines them clearly:
| Item | What it is | Key characteristics |
|---|---|---|
| Semantic model | The analytical model (formerly called a dataset) | Tables, relationships, measures and security; the single source for reports |
| Report | One or more pages of interactive visuals | Built on one semantic model; supports filtering, slicing, drillthrough and bookmarks; created in Power BI Desktop or the Power BI service |
| Dashboard | A single-page canvas of pinned tiles | A feature of the Power BI service only; tiles can come from several reports and semantic models; cannot be filtered or sliced |
In a Fabric architecture, the semantic model is the contract, the report is the analytical experience, and a dashboard is an optional summary that points back to reports. Most teams can start with semantic models and reports alone.
Production architecture
In production, the report is one link in a governed chain:
Fabric data
- WarehouseGold tables
- LakehouseGold Delta tables
Semantic model
- Semantic modelMeasures, relationships, RLS, OLS
Report
- Power BI reportPages, drillthrough, tooltips
Distribution
- WorkspaceOwned by the team
- AppWorkspace app or org app
Consumers
- Business users
- Executives
- Operations
Cross-cutting concerns
- Identity
- Row-level security
- Sensitivity labels
- Endorsement
- Lineage
- Monitoring
- Deployment
A few production principles follow from this picture:
- Separate authoring from consumption. Reports and semantic models are developed in Dev workspaces and promoted through Test to Prod with Git and CI/CD, as described in Git integration in Microsoft Fabric and Fabric CI/CD with fabric-cicd.
- Distribute through apps. A workspace app packages reports for broad audiences, with audience groups that show different content to different people. Org apps are a newer Fabric item that coexists with workspace apps. Consumers then never need a role in the authoring workspace.
- Secure the data, not the page. Row-level security in the semantic model, and access control in the data layer, decide what each user can see. Hiding a visual is not security.
- Make trust visible. Endorse the production semantic model and report, apply sensitivity labels, and keep lineage clean back to Gold. The Fabric governance guide explains how these controls fit together.
Common mistakes
These patterns undermine reports far more often than any missing feature:
- Too many visuals. The page loads slowly and nobody knows where to look first.
- Every chart in a different color. Color should carry meaning, such as one brand color, one highlight and one warning, not decoration.
- Dashboards that need a manual. If users need an explanation before they can read the page, the layout is not doing its job.
- Business logic only in report visuals. Visual-level calculations and filters that define a metric will drift from the next report. Put definitions in the semantic model.
- Giant tables instead of analysis. Exporting a million rows is not insight. Show the answer, then let users drill to the rows.
- No mobile consideration. Executives often open reports on a phone; a mobile layout for key pages is cheap insurance.
- No performance testing. A report that is fast on a developer’s sample becomes slow on production data and concurrency.
- No accessibility testing. Contrast, keyboard navigation and alt text are rarely checked unless someone makes them part of the release checklist.
- Copying an operational application’s UI. Analytical reports are for understanding and deciding. Forms, workflows and transactional screens belong in applications.
Closing
A good Power BI report is not a collection of charts. It is a decision interface built on trusted data and a governed semantic model.
In Fabric, the hard work happens upstream: reliable Gold tables, a well-designed semantic model and clear governance. The report’s job is to respect that work and keep it visible. Start with the decision, show the headline first, reveal detail on demand, keep business logic in the model, and test for performance and accessibility before the first user does.
References
- Microsoft Learn — Use Performance Analyzer to examine report performance
- Microsoft Learn — Drillthrough in Power BI reports
- Microsoft Learn — Create report tooltips in Power BI
- Microsoft Learn — Create report bookmarks in Power BI
- Microsoft Learn — Design Power BI reports for accessibility
- Microsoft Learn — Optimization guide for Power BI
- Microsoft Learn — Introduction to dashboards for Power BI designers
- Microsoft Learn — Publish a workspace app in Power BI
Was this helpful?
What could be better?
Thanks — that helps me improve Javi Builds.
Continue learning
- Cheat sheetMicrosoft Fabric Workspace Settings Cheat SheetThe Microsoft Fabric workspace settings that change how a data engineering workspace behaves: Spark pools, environments and jobs, domain assignment, OneLake shortcut caching and diagnostics, and data workflow (Apache Airflow job) runtime pools, with what to verify for each.Open cheat sheet
- ArticleSQL Server to Microsoft Fabric: What Maps to What?A mental model for SQL Server professionals and DBAs moving into Microsoft Fabric: the closest Fabric equivalent of an instance, database, table, SQL Agent, SSIS, Query Store, tempdb and backups, where each DBA task goes, and where the mapping breaks down.Read article
What should I build next?
I'm building Javi Builds around practical problems data engineers actually face. Tell me what you'd like to see next.
Stay in the loop
Get the Javi Builds updates you actually care about.
Check your inbox
I sent you a confirmation link. Your subscription will be active after you confirm it.