Bronze, Silver and Gold in a small company: what is worth it and what is surplus

The Microsoft Fabric medallion architecture, three layers Bronze, Silver and Gold on Lakehouses with notebooks for ingestion, transformation and the model, is the textbook pattern. And the textbook pattern is designed for companies that are not a small business with three data sources, one ERP, a fair amount of Excel and a management committee that wants figures that add up.

Built at that scale, the pattern splits into three parts: what is worth it as is, what is surplus from day one and what keeps chafing even when done well.

What is worth it

• Bronze as cheap insurance. Storing the data exactly as it arrives, with a load date and untouched, costs almost nothing. When someone asks "why did March show a different figure?", the answer is in Bronze, not in anyone's memory.
• Silver is where the work is. Typing, deduplicating, unifying customers the ERP calls three different things. It is the least glamorous layer and the one that decides whether the dashboard is credible.
• Gold is the semantic model, not another table. Facts, dimensions and measures: what Power BI sees. With a well-made Silver, Gold ends up thin and small, and the report responds instantly.
• Notebooks instead of drag-and-drop boxes. With code it is possible to see what happens to every piece of data, to version it and to explain it. Whoever maintains it tomorrow reads the notebook and understands it.
• One place for everything. Ingestion, transformation, model and report on the same platform. Fewer pieces, fewer things that break at three in the morning.

What is surplus

• Enterprise-grade orchestration. Pipelines with dependencies and load windows as if this were a bank. For three sources that change once a day, one notebook per layer and a simple schedule are enough and can be understood.
• Streaming "just in case". The business looks at the data in the morning. A well-made nightly load is worth more than a real time nobody checks.
• Ten tables in Silver for three entities. Over-normalising is as expensive to maintain as not normalising. In a small company, the model has to fit on a whiteboard.

The seams

• The cost of capacity. Fabric is paid by capacity, and in a small company sizing it right is the difference between it paying off or not. Measure, pause when it is not in use and distrust the "recommended" SKU.
• Excel as a source in Bronze. It works, but every column added by hand breaks something. Ingesting from Excel needs validations an ERP would not require.
• Who maintains it. A notebook is maintained by someone who can read code. If nobody in the company can, the system depends on outside help. That is why a KubiqX Core project is delivered with the model, the notebooks and the documentation, and why Loop exists as monthly maintenance: so that dependency is a decision, not a trap.

The rule

The medallion architecture is neither expensive nor difficult in itself. What is expensive is building it as if the company had a hundred sources when it has three. The rule that comes out of experience is to start with the minimum that is traceable and grow only when it hurts.

The question left open, for anyone running Fabric in a company that is not large: how much of what was built at the start is still in use.