What is Power BI?

Power BI is Microsoft's business analytics platform. It connects to data sources, transforms and models that data, builds interactive reports and dashboards from it, and shares those reports across the organisation through the browser, Teams, or mobile apps. It is by a wide margin the most-deployed BI tool in the world — partly on merit, partly because Pro licences ride along in Microsoft 365 E5, which means most enterprises already own it whether they planned a BI strategy or not.

That accident of licensing is Power BI's blessing and curse: adoption is effortless, and so is sprawl. This guide covers what the platform actually consists of, how the licensing works after the Fabric rebrand, and how to start in a way you won't regret in two years.

The pieces of Power BI

Power BI Desktop is the free Windows authoring tool where reports are actually built. You connect to data, clean and shape it in Power Query (see the Power Query primer — the same engine as in Excel), model tables and relationships, write measures in DAX, and design the visual report pages. Desktop is where all the real modelling work happens; the file it produces gets published to the cloud.

The Power BI Service (app.powerbi.com) is where published content lives and where everyone else experiences Power BI. Reports are organised into workspaces, refreshed on schedules, bundled into apps for wide distribution, and embedded into Teams tabs and SharePoint pages. Getting workspace structure right early matters more than most teams realise — the patterns are in workspace design.

Semantic models are the most important concept in the platform and the least visible to beginners. A semantic model (formerly "dataset") is the data model — tables, relationships, measures — published as a reusable artefact that many reports can connect to. Mature organisations build a few well-governed semantic models and let dozens of thin reports hang off them; immature ones build a new model inside every report file and end up with forty versions of "revenue". The discipline is covered in semantic model design and the calculation patterns in DAX patterns.

The rest of the family: mobile apps for iOS and Android; paginated reports for pixel-perfect, printable output (the operational-reporting workhorse); dataflows for reusable Power Query logic in the cloud (dataflows explained); embedding APIs for putting reports inside your own applications; and Power BI Report Server for the shrinking set of organisations that must keep BI on-premises.

How it handles data

Power BI ships with hundreds of connectors — SQL Server, Snowflake, Databricks, Dataverse, SharePoint, Excel, Salesforce, Google BigQuery, and the rest. The architectural choice per model is the storage mode:

• Import — data is loaded into Power BI's in-memory columnar engine. Blazingly fast to query, refreshed on a schedule. The right default for the vast majority of models.
• DirectQuery — every visual sends a live query to the source. Real-time and no data duplication, but performance now depends entirely on the source database, and slow reports follow. Use it when data volume or freshness genuinely demands it, not by reflex.
• Composite and Direct Lake — hybrids. Direct Lake, a Fabric-era mode, reads Delta tables in OneLake at near-import speed without scheduled refresh, and is the direction of travel for lakehouse-based estates.

For Dynamics 365 shops, Power BI is also the default analytics surface over Dataverse — the business-application data lands there, and the modern path pipes it through Fabric link rather than direct queries against the transactional store.

Licensing after Fabric

Power BI licensing confuses everyone because the same reports can be paid for four different ways:

• Free — full Desktop authoring, but you can't share anything with anyone. Personal analysis only, with one big exception below.
• Pro — the standard per-user licence, included in Microsoft 365 E5. Both the publisher and every viewer need Pro. This per-viewer requirement is the number-one licensing surprise.
• Premium Per User (PPU) — Pro plus bigger models, more frequent refresh, paginated reports, and deployment pipelines. A niche middle tier.
• Fabric capacity — you buy compute capacity for the tenant instead of licences per viewer. On F64 capacity and above, viewers with Free licences can consume reports — which flips the economics for large audiences. Fabric capacity also unlocks the rest of the Microsoft data platform (pipelines, lakehouses, warehouses, real-time analytics), because Power BI is now one workload inside Microsoft Fabric.

The break-even arithmetic and decision rules are in Pro vs PPU vs Fabric. The one-line version: Pro for hundreds of viewers, capacity for thousands, PPU almost never.

Governance: the part everyone skips

Because anyone with Desktop can build a report, Power BI estates grow bottom-up. Left alone, that produces the report swamp: hundreds of workspaces, conflicting numbers in every meeting, refresh failures nobody owns, and sensitive data shared further than anyone intended. The countermeasures are well-established:

• A workspace and naming structure with clear ownership, dev/test/prod separation, and deployment pipelines for the models that matter.
• Certified and promoted models — endorsement marks the blessed source of truth so people stop building on random files.
• Row-level security in the model, not in report filters — a filter is a suggestion, RLS and OLS are enforcement.
• Sensitivity labels and tenant settings — Power BI inherits Purview labelling and has its own export/sharing controls worth reviewing on day one.

None of this requires a BI centre of excellence on the org chart. It requires deciding, early, which handful of models are the truth and who owns them.

Where it fits

Inside Microsoft 365, Power BI is the analytics surface: reports embed in Teams and SharePoint, security rides on Entra ID, labels come from Purview, and Copilot can generate report pages and DAX against your models. Its centre of gravity, though, has moved to the data-platform side — Power BI is now the visualisation layer of Microsoft Fabric, and decisions about lakehouses and OneLake increasingly shape how BI gets built.

For an organisation on Microsoft 365 wanting self-service analytics, the choice is rarely whether Power BI — you already own it. The real decisions are the licensing model for your audience size, and which semantic models you'll govern as the single source of truth. Get those two right and the rest is report-building.