Move the warehouse to Fabric without breaking a single report.
On-prem SQL, aging warehouses, file shares full of extracts — we migrate them into Microsoft Fabric in stages, with the old and new estate running side by side until the numbers reconcile.
Four phases, and the cutover is the boring part
Migrations fail at the end — when reports point at the new platform and the numbers don't match. So we make reconciliation a phase of its own, not an afterthought.
Assess & sequence
Inventory what exists, what's actually used, and what can be retired instead of moved. The migration order follows business dependency, not table size.
- Source inventory with usage evidence
- Retire list — data that doesn't make the trip
- Wave plan by business area
Land in OneLake
Data flows into OneLake via mirroring, shortcuts, or pipelines — whichever fits each source. Where Fabric can mirror a database, we avoid building pipelines at all.
- Mirroring / shortcuts before custom pipelines
- Incremental loads, not big-bang copies
- Source systems untouched throughout
Model & reconcile
Landed data is shaped into governed lakehouse layers, and every migrated number is reconciled against the legacy system while both still run.
- Medallion-style layered model
- Row-count and value reconciliation reports
- Both estates live until sign-off
Cut over & retire
Reports repoint to Fabric one wave at a time. The legacy estate is decommissioned only after its consumers are gone — which is also when the old licence and hardware costs stop.
- Wave-by-wave repointing
- Rollback path per wave
- Decommission checklist, signed off
Three things that shape a Fabric migration
OneLake stores data once in open Delta format; SQL, Spark, and Power BI all read the same copy. Migrating well means you stop maintaining duplicates — that's where the cost falls out.
Data in ADLS, S3, or Dataverse can be shortcut into OneLake without copying. We move data only when there's a reason to — less pipeline, less to go wrong.
Workspaces, domains, and Purview policies are set before data lands, not retrofitted after. A migration is the one chance to start governed — we don't waste it.
Start with the inventory.
A short assessment of your current estate: what moves, what retires, and what order makes the migration safe.