What Power BI is and what it does
Power BI is Microsoft’s business analytics platform. It brings together tools for connecting data, defining models, building reports and distributing them. Its main output is not a record list that every reader edits, but an analytical representation built on identified sources. It is part of the Microsoft Fabric experience while retaining its own applications and concepts.
Its value becomes apparent when several people need to answer questions using the same definitions: what counts as a transaction, which population is included or which date determines the reporting period. Interactive reports let readers explore those definitions through selections and filters. In OSH, this can support regular monitoring if the model preserves the distinction between recorded activity, completed action and verified closure.
Reference documentation: Microsoft Learn: What is Power BI?.
Origins, development and architectural changes
Power BI grew from analytical tools in Excel and the SQL Server environment. In July 2013, Microsoft introduced Power BI for Office 365, using Power Query for preparation, Power Pivot for modelling and Power View for visualisation. That offering was closely tied to Excel workbooks and their distribution through the cloud.
24 July 2015 marked general availability of the new Power BI service and Power BI Desktop, previously called Power BI Designer. Preparing reports in an application and distributing them through a service became a central distinction. In April of the same year, Microsoft had announced its acquisition of Datazen to expand its mobile analytics strategy.
Subsequent development places it within the broader Microsoft Fabric data ecosystem. Historical names should not be treated as a current product catalogue: Power BI datamarts were retired on 1 October 2025, with migration guidance towards Fabric Data Warehouse. An old manual may explain a feature’s origins without describing an option available today.
Reference documentation: Microsoft (2013): What powers Power BI in Office 365? · Microsoft (2015): Power BI general availability · Microsoft (2015): Acquisition of Datazen · Microsoft Japan CSS Support: Datamart retirement and migration.
Data models, measures and reports
Power BI turns connected data into a model for analysis through reports. Distinguish three layers: the source where an event is recorded, the semantic model that defines relationships and calculations, and the visualisations the reader uses. A correctly designed chart can still show the wrong result if its model relates tables incorrectly.
Microsoft documents imported models, models querying external sources, and composite models. The choice affects refresh, performance and administration. For an OSH organisation, the first question is who owns each definition: what an open action means, which site owns it and which date assigns it to a reporting period. Those decisions should not depend on each reader’s interpretation. A shared report is most useful when its underlying terms are equally shared and documented.
Reference documentation: Microsoft Learn — Semantic models in the Power BI service.
Desktop, the service, applications and deployment
Power BI Desktop is the Windows application for creating models and reports. The Power BI service provides a web environment for publishing, organising and sharing content. Mobile applications provide device-adapted viewing experiences; they do not transfer every development feature to a phone. Accessing the service in a Mac browser does not make Desktop a native macOS application.
Workspaces organise content and responsibilities. An app distributed from this environment presents selected reports to its audience. Separate the workspace where a model is built and changed from the content readers use, allowing changes to be reviewed before wider release.
A gateway connects certain on-premises sources to cloud services. It needs administration, credentials and availability; it does not correct source data. Power BI Embedded integrates analytics into applications, while Report Server supports an organisation-managed deployment. These are architectural alternatives with their own requirements, not interchangeable names for sharing a link.
Reference documentation: Microsoft Learn: What is Power BI? · Microsoft Learn: On-premises data gateway · Microsoft Learn: Embedded analytics overview · Microsoft Learn: Power BI Report Server.
Preparation, storage and calculation context
Power Query arranges data preparation into steps and uses the M language. DAX defines calculations over the model, including measures evaluated in context. These are not two names for the same task: cleaning a column during import and calculating a percentage when a reader filters a site happen at different stages.
In Import mode, the model holds data that needs refreshing. DirectQuery queries the source to resolve the analysis, so performance also depends on that source and the generated queries. Direct Lake is another Fabric semantic-model option associated with OneLake data. Composite models combine approaches under specific conditions. Choosing a mode requires testing actual volumes and interactions.
A percentage measure illustrates context: its numerator and denominator must refer to the same population and period. Dividing verified actions by all historical actions answers a different question from dividing them by actions planned for the quarter. Relationships, filters and active dates can change that population. An indicator dictionary should explain the decision in understandable language as well as retaining the formula.
Reference documentation: Microsoft Learn — Semantic models in the Power BI service · Microsoft Learn: Direct Lake overview · Microsoft Learn: DAX basics in Power BI Desktop · Microsoft Learn: What is Power Query?.
Interactive reports, paginated output and distribution
An interactive report supports exploration: selecting a site can change other views and expose relevant detail. A dashboard brings indicators together for an overview, but its components may not share the same source or refresh date. Custom visuals should also be assessed for accessibility, maintenance and their fit with the analytical question.
Paginated reports address a different problem: producing documents with controlled page layout and tables that continue across successive pages. Power BI Report Builder is the specific authoring tool and should not be confused with Desktop. This format can suit a complete action listing or documentary output, whereas an interactive report supports exploration.
Publish to web is public access without authentication, not a shortcut to private sharing. Microsoft warns that underlying detail can become accessible. An internal OSH dashboard needs a distribution route appropriate to its audience. Embedding content in an application also requires decisions about who authenticates the reader and which permissions govern access to the model.
Reference documentation: Microsoft Learn: Paginated reports · Microsoft Learn: Publish to web · Microsoft Learn: Embedded analytics overview.
Licences, capacities and permissions: three separate decisions
Microsoft distinguishes individual Fabric Free, Power BI Pro and Premium Per User (PPU) licences as well as capacity licences. An individual licence defines user capabilities; capacity and content location affect distribution and processing. Saying that an organisation has Power BI is not enough to establish who can open a particular report.
A free user can create content for personal use in My workspace and consume certain content hosted on qualifying capacity. Content in a PPU workspace requires PPU for its consumers; model reading and Build permissions form another layer. These distinctions explain why a link can work for its author but not for someone else.
Before implementing recurring reporting, identify four roles: the person maintaining the source, the model developer, the person approving distribution and the reader. Then test their actual permissions. A broader licence does not justify wider access to personal information, and an editing role can have a different scope from row-level controls intended for viewers.
Reference documentation: Microsoft Learn: Power BI service features by license type · Microsoft Learn: Features available to free users · Microsoft Learn — Row-level security with Power BI.
Maintaining and using analytical reports
Recurring reports need an operating process: detect a failed refresh, investigate its cause and decide whether the last version remains usable. The technical refresh date should be accompanied by the date of the data itself; they differ when a source delivers information late.
Dataflows can reuse preparation across models, while a shared semantic model can prevent each report from inventing its own measure. Reuse creates dependencies: changing a classification may affect several consumers. A change record should explain what changed, when it took effect and which historical comparisons are no longer consistent.
Beyond OSH, Power BI is used for sales, operations, budgets and service monitoring. Across these fields, distinguish description from explanation: observing a change starts a question. A dashboard can indicate where to focus a review, but explaining the cause requires examining the process, population and way the events were recorded.
Reference documentation: Microsoft Learn — Semantic models in the Power BI service · Microsoft Learn: On-premises data gateway · Microsoft Learn: What is Power Query?.
Safety questions it can help explore
- Action progress: comparing age, due dates and verified closure separates work performed from work that has not yet been checked.
- Activity coverage: relating a plan to completion records helps identify outstanding sites or periods. A missing record must be distinguished from an activity that did not apply.
- Data quality: showing duplicates, incomplete dates and unclassified records makes the dashboard’s own uncertainty visible.
These are editorial use cases proposed by Sabentis. Power BI presents results from rules defined by the organisation; it does not independently determine whether a risk is acceptable or demonstrate causation between an intervention and an observed change.
Worked example: a verifiable action dashboard
Hypothetical example. An organisation wants to compare action follow-up across four sites without duplicating records appearing in several reports.
- Prepare an action table with unique identifiers and a separate site table. Where an action has several evidence items, keep those items separate so they do not multiply the action count.
- Define the reporting period and verified-closure rule. Separate creation, due and final-validation dates.
- Build measures for open, overdue and verified actions, alongside a detail table explaining each total.
- Reconcile a sample with the source register. Test an action with two evidence items, one without a date and another reassigned to a different site.
- Publish for authorised recipients and check what each role sees. Record the refresh date and model owner.
The acceptance criterion is that a reader can trace an indicator back to its records and that totals do not change because of an accidental table relationship. Incomplete records should remain visible as a review category, rather than disappearing from the denominator without explanation.
A numeric test makes the criterion concrete: if the source contains 40 unique actions, 30 with one piece of evidence and ten with two each, the indicator must still show 40 actions, not 50. If eight of 32 open actions are overdue, the rate among open actions is 25%. Dividing eight by all 40 produces 20% and answers a different question. Both calculations can be arithmetically correct; the label must identify the population used.
Reference documentation: Microsoft Learn — Semantic models in the Power BI service.
Security, filters and interpretation
Row-level security restricts data through roles. Microsoft specifies that, in the service, it applies to users with Viewer permissions, not workspace Admin, Member or Contributor roles. A visual filter is not equivalent to that access control. Test permissions with representative user roles before sharing sensitive information.
A percentage needs a defined, comparable population. If the number of planned actions changes, the completion proportion may change even when actual effort is similar. Show the reporting date and refresh status: an accessible dashboard does not guarantee fresh data. Changes to column names, sources or credentials require renewed checking. Clearly separate a missing value from a genuine zero and explain any records excluded from calculations.
Reference documentation: Microsoft Learn — Row-level security with Power BI.
When it adds value beyond a spreadsheet
Power BI fits recurring reporting where several readers need consistent definitions. Before adoption, confirm licensing, the access model, connector availability and who will resolve a failed refresh; no single configuration suits every organisation.
Excel can support reconciliation and one-off analysis. Tableau provides another visual-analysis environment to assess against available sources, skills and architecture. Avoid maintaining conflicting definitions of the same indicator in different tools. Decide where each calculation is authoritative before distributing the output widely.
Frequently asked questions about Power BI in OSH
Does Power BI replace the action-recording system?
The report analyses information. The operational register and its evidence still need an accountable owner and an identifiable source.
Is a dashboard validated if refreshing succeeds?
No. Technical execution does not establish that the denominator, relationships and calculation rules are appropriate.
Can each site see only its own data?
Row-level access controls can be designed for that purpose, but they must be configured and tested with the relevant roles. A site-selection filter alone is insufficient.
Reference documentation: Microsoft Learn — Row-level security with Power BI.
