OneLake doesn't ask you to migrate anything, and that's the whole difference

For fifteen years, every data project started the same way: before answering the first business question, you had to move the information. Copy it from the ERP, from the SQL database at headquarters, from the sales team's Excel files, and put it in a new repository. That upfront migration was half the budget and almost all the risk, and it happened before anyone had seen a single dashboard.

OneLake, the storage layer of Microsoft Fabric, changes the starting point: it connects sources where they already are.

What a shortcut is, in plain English

A shortcut is a link. From Fabric, a folder or table that lives somewhere else—another storage location, another account, another service—appears as if it were inside your data lake, and you query it the same way. The information is not copied or duplicated: it is referenced.

Mirroring is its close relative: an operational database is continuously mirrored into OneLake, without you building or maintaining the copy process, and without penalizing the database that is serving day-to-day operations. And for everything else—business systems, files, on-premises services—there are the usual connectors.

Three paths, and each source comes in through the one that fits it.

Why this changes the cost of entry

When you don't have to migrate to get started, the whole project is reordered:

• The first dashboard arrives in weeks, not quarters, because it isn't waiting for a migration to finish.
• Risk goes down: if a link isn't useful, you remove it. There's no need to roll back a copy of a million rows.
• You test before you commit. You can validate whether a source adds value with the real data, not with a sample.
• Governance comes first, not last: permissions and lineage are defined on the linked source from day one.

What still takes work

It's worth saying, because product messaging tends to leave it out: linking is not modeling. OneLake solves access, not meaning. You still need to agree on what an active customer is, what counts as a sale and at what point; you still need a model people understand; and you still need to clean the data, because data quality doesn't improve just by changing storage.

The difference is when that work gets done: on data you can already see, with the business team watching, instead of blindly during the three months the copy used to take.

Where the dashboards live

On your platform. Pipelines, the model, and the dashboards are built inside your tenant, with your identity and your permissions; the information doesn't move to any service of ours. And when you want to apply artificial intelligence to it, Azure AI Foundry works on the same platform, with your real data and not a demo dataset.

If your information is spread across an ERP, a few databases at headquarters, and too many Excel files, tell us which question you want to answer first. That's where you start, not with the migration.