A beautiful report on bad data is a fast way to be wrong. Two questions reveal whether your Fabric and Power BI foundation is solid, and the right size for you.

The unglamorous layer decides everything

Everybody wants to talk about the report. Now everybody wants to talk about the AI agent perched on top of it. Almost nobody wants to talk about the pipeline, the model and the state of the data underneath, which is a pity, because that is the layer that quietly decides whether any of the impressive stuff works at all.

We cover the practical side of this in our data strategy and leadership work, and Microsoft Fabric IQ: Sort Your Definitions First looks at a closely related question.

A beautiful report built on bad data is a fast route to being confidently wrong. An agent on the same data is faster still. None of the layers people get excited about can rescue a foundation that was never laid properly. They inherit its every flaw, magnified, and present them with a straight face.

The work nobody photographs for the keynote

The engineering that matters never makes the slide. Ingestion that does not fall over when a source system renames a column overnight. A sensible path from raw to usable, the medallion the Fabric world talks about. An agreed grain, so everyone means the same thing by a single row. Measures that were tested rather than eyeballed once and trusted ever after. It is plumbing, and it is also the difference between a platform people lean on and one they quietly abandon. The truest sign your foundation has failed is not an error message. It is a finance team that has gone back to Excel and not bothered to tell you.

The trap the mid-market falls into

There is a second failure, and it is the one nobody owns up to: over-engineering. You watch the big platform talks, all built for enterprises with hundreds of source systems and a data team the size of your whole company, and you build that for a finance team of twelve. A two-hundred-table lakehouse and three layers of abstraction to report on a business that closes its books in a morning.

Over-engineering is worse than its opposite, because it fails slowly and expensively. The under-built thing breaks early and cheaply, and you fix it. The over-built thing works beautifully right up until the contractor who designed it leaves, and then nobody on your team can change it, maintain it or explain it, and you are paying a premium to be frightened of your own platform.

Two questions for a foundation that fits

First, the maintenance question. Could a competent new analyst understand your model and make a safe change within a day. If the honest answer is no, it is too complex, too undocumented, or both, and each is a bill that arrives at the worst possible moment.

Second, the fit question. Is the architecture sized to your business or to a conference stage. Most mid-market estates are carrying complexity they will never use, bought as insurance against a scale they are nowhere near.

Run both on your current platform this week. If either one stings, that is where the work is, not in the next dashboard. Getting the foundation right and right-sized is the spine of our Data Platform and Warehouse Build and Microsoft Fabric work, because once it is solid the report and the agent largely look after themselves, and when it is not, no amount of polish on top will save you, however good the demo looked.

If any of this sounds familiar, talk to us about your data.

Related reading

• Are your people happy? The first question in any Data or AI project
• Query Where Your Data Lives
• The Spreadsheet That Became A System Nobody Maintains