Your dashboard reports revenue up 4%, finance says 7%. Both cite “the data,” and now they’re in an unwanted meeting trying to decide which to believe.

That argument almost never starts with the report. It starts upstream, in the plumbing, where data is pulled, cleaned, and shuffled by a patchwork of jobs and scripts that grow one workaround at a time.

Microsoft Fabric is Microsoft’s answer to that patchwork. Storage, movement, transformation, reporting- all dragged onto one platform instead of six tools held together with hope. Which sounds great in a keynote.

The interesting question is what a modern data pipeline looks like inside it, and how you build one that doesn’t fold the moment your data doubles.

That’s what we’re getting into. What Fabric hands you, how the bits fit together, how to structure the thing properly, and the deep hole worth steering round.

What Microsoft Fabric Actually Changes

So, what is it? Microsoft’s unified analytics platform, covering data engineering, data integration, warehousing, real-time analytics and Power BI.

All of it perched on one foundation: OneLake.

People skim past that last word, which is a shame, because it’s the whole trick. OneLake is a single storage layer for your entire organisation.

You stop shuffling copies between a lake, a warehouse and a reporting model, each one drifting a little further from the truth every week. The data lands once. Everything else just reads it.

Spend years wiring Azure services together by hand? Then you’ll know why that matters. Fewer moving parts means fewer places for things to break in silence.

The Building Blocks of a Fabric Data Pipeline

Before you design a thing, know which tool does what. Fabric gives you several, and grabbing the wrong one for a job is where an awful lot of data pipeline work quietly goes wrong.

Data Factory and Pipelines

This is your movement layer. Fabric data pipelines orchestrate the whole flow, pulling data in from wherever it lives, whether that’s Dynamics 365, SQL, an API or a pile of flat files, and marshalling the steps that follow. Think of it as the conductor rather than the musician.

Dataflows Gen2

Where Data Factory moves, Dataflows Gen2 shapes. It’s Power Query at cloud scale, so if your team already knows Power Query in Power BI, they’ll be at home within an afternoon. Ideal for transformations that don’t need code.

Lakehouse and Warehouse

Two destinations, two personalities. The Lakehouse suits raw, varied, semi-structured data and code-first work with Spark. The Warehouse suits structured, SQL-driven analytics. Plenty of setups use both, and there’s no prize for purity here.

Notebooks and Spark

For the heavier transformations, or anything with machine learning ambitions, notebooks give your engineers proper room to work. This is where data engineering in Fabric stops feeling like a wizard and starts feeling like software.

Designing the Pipeline the Medallion Approach

Most of the value lives here, and it has almost nothing to do with tooling. It’s the structure. The pattern that keeps proving itself is medallion architecture: three layers, one job each.

Bronze Raw and Untouched

Land it exactly as it arrived. No cleaning. No clever fixes. No opinions. Sounds wasteful, right up until someone asks a question out of nowhere and you can answer it because the original never went anywhere.

It’s also your AI insurance policy. Models are hungry for detail, including the scruffy bits a tidy-minded pipeline would have binned on sight.

Silver Cleaned and Conformed

Now you are tidy. Deduplicate, fix the types, standardise the fields, and join sources that belong together. Silver is where “customer” finally means one thing, whether the record came from your CRM or your finance system.

Gold Business-Ready

The only layer your reports touch. Aggregated, modelled, and built around how the business thinks, rather than how some source system happened to file things away. Power BI sits on Gold and nowhere else. That’s exactly why the reports stay quick and believable.

Three layers, clear jobs. Something breaks; you know which one to open. That alone saves more hours than any feature on the box.

Where ETL and ELT Fit into This

It’s worth being clear, because the terms fly around constantly. Fabric leans naturally towards ELT the raw data comes first, then it’s transformed within the platform using its own compute. The medallion layers make ELT thinking visible.

That said, ETL hasn’t retired. Sensitive fields that must be masked before storage, or small, stable feeds that were transformed upfront, still make a strong case for the old approach. Most real pipelines use a bit of both.

We’ve unpacked that whole debate properly in our guide on ETL vs ELT in Power BI, including how to decide which suits your data. Worth a read before you commit to a pattern.

Turning Pipelines into Insight

A pipeline nobody uses is just an expensive hobby. What comes out the far end is the point.

Power BI is native to Fabric, so Gold-layer data flows into your reports. No exports. No refresh gymnastics. No copies quietly drifting apart. Direct Lake mode reads directly from OneLake, delivering import-level speed without an import model to babysit.

And Copilot rides on top, so people can just ask, in plain English, instead of joining a three-week queue for a report request.

Say that part out loud because it’s the real business case. Answers come faster. People believe in numbers. Nobody spends Thursday arguing over whose figures are right. Engineering was only ever a means to that end.

The Mistakes That Cost the Most

A few patterns recur repeatedly, and they’re all avoidable.

Skipping Bronze. Teams under pressure transform on ingestion to save a step. Six months later, someone needs the original, and it’s gone. Land raw first. Always.

Treating Gold as a dumping ground. If every report builds its own logic on top of Silver, you’ll end up with four versions of “revenue” and a very awkward board meeting. Model it once, in Gold.

Ignoring governance until later. Ownership, naming, access, lineage. Sort these early, or your lakehouse quietly becomes a swamp nobody trusts.

Over-engineering from day one. Not every organisation needs Spark notebooks and real-time streaming. Start with what your data demands, then grow.

Getting It Right First Time

None of this is beyond a decent team. But there’s a gulf between a pipeline that works on Tuesday and one that still stands two years and five sources later. That gulf is designed, nearly every time.

That is the work we do. Our BI and Data Science services cover the entire journey, including architecture, medallion design, pipeline build, and the Power BI layer on top, so your data estate grows rather than being patched every quarter.

And when those pipelines need to reach Dynamics 365, Azure and whatever else you’re running, our Azure Integration services keep it all moving clean.

Stallions Solutions is a Microsoft Solutions Partner, and we work with organisations right across the public and private sector. We’re rather fond of doing the unglamorous groundwork properly. It shows later.

Final Thoughts

Fabric won’t magically fix bad data. Nothing will. What it does is strip away the excuses for a single platform, a single storage layer, and a single home for the pipeline. After that, it’s about discipline. Land raw. Clean on purpose. Model once. Govern from day one.

Nail that, and reporting stops being a monthly chore and becomes an actual advantage. If you’d like a hand designing yours, or you’re getting your data estate ready for AI, our AI Enablement Programme and data specialists can help you build it properly.

Book a free assessment, and we’ll map out the smartest path for your data.

Frequently Asked Questions

What is a modern data pipeline in Microsoft Fabric?

It’s an end-to-end flow that ingests raw data, transforms it into structured layers, and serves it to Power BI, all within a single platform rather than across several stitched-together tools.

What is medallion architecture?

A three-layer design, Bronze for raw data, Silver for cleaned and conformed data, and Gold for business-ready models. It keeps responsibilities clear and makes pipelines much easier to troubleshoot.

Should I use a Lakehouse or a Warehouse?

Lakehouse for raw, diverse data and Spark-based engineering. Warehouse for structured, SQL-driven analytics. Many organisations sensibly run both.

Does Fabric replace Azure Data Factory?

Fabric includes Data Factory capabilities natively. Standalone Azure Data Factory still has its place in some architectures, particularly where existing investment is substantial.

How does this prepare us for AI?

Keeping raw data in Bronze and governed, well-modelled data in Gold gives AI reliable data to work with. Poor data foundations are the most common reason AI projects stall.