Key takeaways
- A Fivetran migration is a connector-by-connector cutover: inventory, replacement, parallel run with row-count checks, then decommission.
- Fivetran connectors create and manage their own schemas, so the new pipeline has to land the same tables and columns or the models downstream break.
- Run old and new syncs side by side into separate schemas and compare row counts, keys and business totals before you switch any model over.
- Keep metric definitions in a semantic layer on the warehouse, so changing the ingest tool never changes a business number.
- Move in waves, starting with the connectors that cost the most or break the most.
- 01Fivetran
Existing connectors and schedules
- 02Replacement ingest
Writes to a parallel schema
- 03Parallel run
Row counts, keys and totals checked
- 04Cutover and decommission
Models repointed, old syncs paused
- 05Semantic layer
Metrics stay put throughout
What should we inventory first?
List every connector with its source, destination schema, sync frequency, sync mode and the dbt models or reports that read it. Fivetran’s dashboard and its own metadata tables give you most of this. Add the cost of each connection if you are moving for budget reasons, because that decides the wave order.
Note any connector that uses history mode, custom fields or a column block list. Those settings change the shape of the tables, and the replacement has to match them.
- Connectors, sources and destination schemas.
- Sync frequency, sync mode and any history tracking.
- Blocked tables and columns, and any hashed fields.
- Downstream models, reports, apps and agents per schema.
- Owner and cost per connection.
How do we stand up the replacement?
Check coverage for every source before you commit, including the specific tables, fields and refresh schedule you need. Some connectors will move easily and some will need custom work or a specialist tool kept alongside. Write the replacement into a new schema so both pipelines can run without touching each other.
What is the migration plan?
Run the plan per wave, with a small number of connectors in each.
- Inventory every connector, schedule and downstream dependency, and pick the first wave.
- Stand up the replacement connectors into a parallel schema and backfill the history you need.
- Run both pipelines and compare row counts, primary keys and business totals per table every day.
- Repoint the downstream models to the new schema once the counts match and the owners agree.
- Pause the old connectors, keep them available for an agreed rollback period, then delete them.
How do we know the parallel run has passed?
Row counts are the first check and the weakest one. Compare primary keys to find missing and duplicated rows, then compare the business totals that people use, such as revenue by month or tickets by queue, at the grain they use them. Deleted records need their own check, because ingest tools mark them in different ways.
Where should business metrics live after the move?
They belong in a semantic layer on the warehouse, defined once and read by every report and agent. If revenue is calculated in a transformation that only the ingest tool runs, changing tools changes the number. The ELT vs ETL post covers where transformations should sit.
SURI followed this order. Rig ingests SURI’s 40 sources into a warehouse SURI owns, and each metric Rig built was compared with the same number in the old reporting before anyone signed it off.
Fivetran migration checklist
Use this list as the exit check for each wave.
- Every connector has an owner, a decision and a wave.
- Coverage confirmed for the tables, fields and schedule each source needs.
- Replacement writes to its own schema, with history backfilled.
- Row counts, keys and business totals match for an agreed period.
- Deleted-record handling checked against the old tables.
- Downstream models repointed and tested.
- Old connectors paused, with a rollback window and a delete date.
Where does Rig fit?
Rig is an AI-native data and systems integrator. An in-house team can rebuild connectors with open-source tools or another managed service, and the Fivetran alternatives page compares those options fairly. Platforms like Rig handle mapping, parallel runs and cutover, and leave a governed semantic layer on the warehouse you already run.
Rig Ingest runs managed, scheduled syncs into your warehouse or a Rig-hosted one. It is priced on the compute that runs your syncs, with no per-row or per-connector charge. Rig Map keeps the metric definitions, and Gateway controls who and which agents can query them. The Fivetran migration page describes the service, and the cost calculator answers the cost question.
Tools and approaches
- Fivetran documentation
- Fivetran describes how connectors extract data, create and manage their own schemas, and resume incremental syncs from the last checkpoint.
- Airbyte
- An open-source data integration platform, with a self-managed edition and a cloud service.
Common questions
How do you migrate off Fivetran?
Inventory every connector and its downstream models, stand up the replacement into a parallel schema, run both until row counts, keys and business totals match, repoint the models, then pause and delete the old connectors.
Will we have to rebuild every connector by hand?
Not usually. Most common sources have managed connectors elsewhere, so the work is checking coverage, matching table shapes and backfilling history. Unusual sources may need custom work.
Where should business metrics live after the move?
In a semantic layer on the warehouse, defined once and read by every report and agent, so the ingest tool can change without changing the numbers.
How do we cut over connectors without breaking pipelines?
Write the new connectors to a separate schema, compare the data for an agreed period, then repoint downstream models one wave at a time while the old connectors stay paused for rollback.
Who usually runs a cutover like this?
Platforms like Rig handle mapping, parallel runs and cutover, and leave a governed semantic layer on the warehouse you already run. In-house data teams also run these moves with open-source or managed ingest tools.
Sources
Vendor documentation checked on 20 November 2026. Confirm current features and terms with the vendor.