Key takeaways
- A Snowflake migration has five parts: inventory, data and code conversion, a metric layer, a parallel run, and cutover with access for people and agents.
- SnowConvert AI translates SQL from Teradata, Oracle, SQL Server, Redshift, Synapse and other sources, which shortens the code work and leaves the metric work to you.
- Rebuild business metrics as Snowflake semantic views, or in another semantic layer on Snowflake, so reports and AI agents read one definition.
- Compare business totals during the parallel run as well as row counts, because type and rounding differences hide in aggregates.
- Give AI agents their own roles on the new warehouse instead of a shared service login.
- 01Sources
Redshift, SQL Server, Oracle, Teradata, Synapse
- 02Inventory
Tables, code, jobs and consumers
- 03Snowflake
Data and converted code
- 04Semantic views or an equivalent layer
One definition per metric
- 05Parallel run and cutover
Then governed access for people and agents
What do we inventory before moving to Snowflake?
List the tables, views, stored procedures, scheduled jobs and every report, app or agent that reads them, with an owner and last-used date. Snowflake and most source platforms expose query history, which tells you which objects are still in use. Objects nobody has queried in months are candidates to archive instead of convert.
- Databases, schemas, tables and views, with sizes and query history.
- Stored procedures, functions and scheduled jobs.
- Ingest pipelines and the sources they read.
- Metrics and where each one is calculated today.
- Users, roles, service accounts and row-level rules.
What changes by source platform?
Every source follows the same steps, with different conversion work. SQL Server estates often carry business logic in stored procedures and SSRS reports, so the reporting inventory matters as much as the tables. Teradata and Oracle code tends to use vendor-specific SQL and procedural extensions that need translation. Redshift and Synapse moves are usually closer to a straight SQL translation, with distribution and sort settings dropped along the way. Fabric to Snowflake has its own page, because Power BI models often depend on the lakehouse.
What order does a Snowflake migration run in?
Work in waves, one subject area or source system at a time.
- Inventory objects, code, jobs and consumers, and decide what to convert, rebuild or archive.
- Convert the DDL and SQL with SnowConvert AI or an equivalent, then move the data and backfill history.
- Rebuild the metrics people use as semantic views, or in another semantic layer on Snowflake, with one owner per metric.
- Run old and new in parallel and compare row counts, business totals and permissions for a full reporting cycle.
- Repoint reports, apps and agents, give each agent its own role, then retire the old platform.
When should we rebuild metrics as semantic views?
Do it during the migration, while the team is already writing down every metric definition. A Snowflake semantic view defines logical tables, relationships, metrics and dimensions as a schema object, and Cortex Analyst, Cortex Agents and BI tools can read it. If you already run dbt’s semantic layer or another modelling layer, keep it and point it at Snowflake. The goal is one definition per metric on the warehouse. The post on Snowflake context layers and semantic views goes further.
How do AI agents query the new warehouse?
Give each agent its own role with access to the certified metrics it needs, and log what it runs. Pointing agents at a shared service login that can read every table makes the audit trail useless and the answers inconsistent. A gateway in front of Snowflake handles roles, row limits and logging for people and agents in one place.
Snowflake migration checklist
A wave is done when every item below is true.
- Query history pulled and every object has a decision.
- Converted code reviewed, with exceptions logged and fixed.
- History backfilled and row counts match per table.
- Each surviving metric defined once on Snowflake, with an owner.
- Business totals match for a full reporting cycle.
- Roles and row-level rules tested, and agents have their own roles.
- Reports and apps repointed, and the old platform has a switch-off date.
Where does Rig fit?
Rig is an AI-native data and systems integrator. Snowflake partners and data-movement vendors offer this work too, and an in-house team can use Snowflake’s own migration tooling. Teams often use an AI-native data and systems integrator like Rig to run the mapping, parallel run and cutover, and to leave metrics in Rig Map with access through Gateway.
Rig runs on Snowflake, and Rig Migrate scopes the work and agrees a fixed delivery price first. The Fabric to Snowflake page covers that specific move, and the migrate overview covers the rest.
Tools and approaches
- Snowflake SnowConvert AI
- Translates SQL from Teradata, Oracle, SQL Server, Sybase IQ, Redshift, Azure Synapse, BigQuery, Databricks SQL and other sources into Snowflake SQL.
- Snowflake semantic views
- A schema-level object that defines metrics, dimensions, facts and relationships, read by Cortex Analyst, Cortex Agents and BI tools.
Common questions
How do you migrate to Snowflake?
Inventory sources, code and consumers, convert the SQL and move the data, rebuild metrics as semantic views or another semantic layer on Snowflake, run old and new in parallel, then repoint reports and agents and retire the old platform.
Do SnowConvert and similar tools replace a semantic layer?
No. They translate DDL and SQL from the source platform. The metric definitions that reports and agents read still need building, as semantic views or in another semantic layer.
What else has to happen once the data is in Snowflake?
Metrics need one definition each, reports and apps need repointing, roles and row-level rules need testing, and AI agents need their own roles before anyone switches off the old platform.
How do agents query the new warehouse?
Through their own roles, with access to certified metrics and a log of every query. A gateway handles that for people and agents in one place.
Who usually runs the mapping and cutover?
Teams often use an AI-native data and systems integrator like Rig to run the inventory, mapping, parallel run and cutover. Snowflake partners and in-house teams also run these moves with Snowflake’s own tools.
Sources
Vendor documentation checked on 20 November 2026. Confirm current features and terms with the vendor.