Fix the foundation
Move off the warehouse without a Friday night
Snowflake, Synapse, Teradata or SQL Server onto Databricks, in validated waves, with a cutover that has a tested way back.
The conversion is the part everybody plans for
It is also the part that goes fine. What sinks a migration is the six months afterwards.
Nobody can say what was in scope
The inventory was a spreadsheet taken in March by somebody who has since left, and three objects nobody counted are still running.
The cutover has no reverse
Which is why it is scheduled for a Friday night. A plan with no way back can only run in a window where being wrong is survivable.
The old platform stays on
Decommissioning is nobody's funded phase, so both bills run for a year and the migration has not saved anything yet.
None of that is conversion. It is record-keeping, and a migration that keeps its own records answers all three without anybody being asked to remember.
The shift
Three ways to move a warehouse
Two of them lose the receipts.
Hand conversion
Accurate, and it does not finish
A tool and a spreadsheet
Fast, and unprovable
A governed migration
Provable, and reversible
How the engagement runs
Assessment first, at a fixed price, so the number you take to a board is grounded in your estate rather than in an average.
Two weeks, fixed fee
A Migration Readiness Sprint. What the estate contains, what converts cleanly, what will fight back, and a sequence with a number against it.
A pilot subject area
One real workload moved end to end and validated, so the estimate for everything else stops being a guess.
Waves, not a big bang
The estate moves in sequenced waves, each validated and each reversible, and the business keeps running throughout.
Evidence per object
Signed migration certificates recording what converted, by which tool version, and what validation said.
Governance carried across
Grants and lineage designed in Unity Catalog as part of the move rather than reconstructed afterwards from what the old system implied.
Decommissioning
Turning the old platform off is in the plan, because it is the phase that decides whether any of this saved money.
27
Source platforms with executable playbooks
Twenty-seven source platforms
Airlift exists because we ran these migrations by hand first, and the parts that hurt are the parts it now handles. It is free and open source and it runs in your own workspace.
Clients we have done this for
B2B ecommerce
AmTab
AmTab grew fast enough to outrun the systems underneath it, and modernised the estate to match the demand rather than throttle it.
Automotive / Fintech
Location Services
Location Services moved off Synapse and SQL Server onto Databricks Lakehouse and Lakebase, converted with TechFabric Airlift and cut over in reversible waves. The rules engine that validates a repossession assignment on arrival now runs against governed lakehouse data.
FAQ
Warehouse migration: common questions
How long does it take?
Nobody can tell you before the assessment, and a firm that quotes a duration on a first call is guessing. What is fixed is the assessment: two weeks, fixed fee, ending in a sequence and a number.
What happens if the cutover goes wrong?
You go back. The reverse path has been tested rather than documented, which is why these do not have to happen on a Friday night.
Can we keep running the old platform in parallel?
For a period, and it is usually the right call. What matters is that turning it off is a dated phase rather than an intention, because otherwise it does not happen.
If your problem is shaped differently
Build an agentic software factory
Launch a product in a regulated vertical
Take the AI demo to production
Make the estate answer for itself
Modernise the estate that outgrew itself
Keep people in charge of the agents you already run
Leave your team able to do it without us
See what production is doing, and be allowed to act