Migration
How long does a data warehouse migration take?
What actually drives the timeline of a migration to Databricks, Snowflake or Fabric: object counts, procedures, downstream consumers and reconciliation, the phases of a typical plan, and how to estimate your own.

Hemant Walchale
· 3 min read
“It’s just copying data, why will it take until June?” is one of the most common questions in data platform migrations. The honest answer: copying data is usually the quickest part. The timeline is driven by everything that depends on the data.
What drives the timeline
| Driver | Why it takes time | What to count |
|---|---|---|
| Objects in use | Every table, view and job needs moving, testing and an owner | Tables and views actually queried in the last 90 days |
| Logic | Stored procedures, ETL jobs and scheduled SQL rarely run unchanged on the new platform | Procedures, ETL jobs and their complexity |
| Consumers | Every dashboard, extract and application needs new connections and testing | Downstream reports, extracts and systems |
| Access | Roles, grants and row-level security have to be recreated | Roles and sensitive tables |
| Proof | Each table is reconciled against the old system and signed off by its owner | Tables, and how many business owners must sign off |
| Data volume | Bulk copy, mostly parallel | Terabytes, but this rarely sets the timeline |
The three numbers that explain most timelines are tables in use, procedures and jobs, and downstream consumers. Terabytes rarely do.
The phases of a typical plan
This is how we plan migrations, with typical durations for a mid-sized estate:
| Phase | Typical duration | What happens |
|---|---|---|
| Assess | 2 to 3 weeks | Inventory, usage analysis, wave plan, target architecture and cost model |
| Pilot wave | 4 to 6 weeks | One business domain migrated end to end, reconciled and in production |
| Scale waves | 8 to 20 weeks | Remaining domains migrated in parallel streams |
| Decommission | 2 to 4 weeks | Final consumers switched, legacy system archived and shut down |
That adds up to roughly four to eight months. A small estate with little procedural code can be faster. An estate with thousands of procedures, many business owners and strict sign-off takes longer.
Why estimates slip
- Unknown consumers. The Excel workbook, the partner extract, the application nobody listed. Every one found late costs days of testing.
- Procedural code underestimated. Cursor-heavy procedures and dynamic SQL need rewriting, not translating.
- Reconciliation left to the end. Proving the numbers match takes time per table. Done at the end, it all lands at once.
- Moving targets. Changes made to the legacy system during the migration have to be made twice.
- Big-bang cutover. Switching everything at once concentrates risk and delays. Waves let consumers move one at a time.
How to go faster
- Retire before you migrate. Usage logs typically show a large share of objects unused. Archive them instead of migrating them.
- Federate first. Query the old warehouse in place from the new platform, for example with Lakehouse Federation on Databricks, so new work starts on day one while tables move domain by domain.
- Use tooling for conversion and reconciliation, such as Databricks Labs’ Lakebridge, and test its output closely.
- Reconcile per wave, not at the end.
- Freeze the legacy system for changes where the business can accept it.
Estimate your own
Count your tables in use, procedures and jobs, and downstream consumers, then try the migration estimator. It turns those counts into engineer-days, calendar weeks for your team size and a suggested number of waves. For platform-specific plans, see the migration guides, for example Teradata to Databricks or Redshift to Databricks.
The short version
Migrations take months because of logic, consumers, access and proof, not data volume. Count tables in use, procedures and consumers; plan assess, pilot, waves and decommission; retire unused objects first; and reconcile as you go.
Sources
- Lakebridge overview, Databricks Labs
- Lakebridge reconcile, Databricks Labs
- What is Lakehouse Federation?, Databricks documentation
Not sure where your own platform stands? Take the free 3-minute data platform scorecard. Need people to do this alongside your team? See senior data engineers for hire. Unfamiliar terms are explained in the data engineering glossary.