Snowflake to Fabric Case Study: Migrating Data and Reporting to Microsoft Fabric
A representative scenario: prove parity, then cut over reports.
This Snowflake to Fabric case study walks through a representative migration of data and Power BI reporting onto Microsoft Fabric: why teams move, how the architecture changes, how to plan the work, how to prove the numbers match and what to measure afterwards.
It is a scenario, not a client story. It names no client and quotes no measured client figures, and the facts about Snowflake mirroring, OneLake and Direct Lake come from Microsoft documentation. For the options analysis, see our guide to Snowflake to Microsoft Fabric reporting options.
About This Snowflake to Fabric Case Study Scenario
The scenario is a mid-sized organisation with finance, sales and operations reporting built on a Snowflake warehouse. Power BI reports read from it through import models and some DirectQuery. Data lands from several source systems, transformations run in Snowflake, and each reporting team keeps its own refresh schedule and access rules.
Treat the numbers you see in your own estate as the real case. Use this walkthrough as a template, and replace each placeholder assumption with your own baseline before you plan.
Why Teams Move Reporting from Snowflake to Fabric
The reasons are usually practical rather than technical. Power BI is already the reporting standard, so teams want fewer hops between the data and the report. They also want data in an open format that Fabric workloads, including Spark, SQL and Power BI, can all read from one copy in OneLake.
Cost visibility matters too. Reporting queries on a warehouse and reporting capacity can be paid for twice. Our Microsoft Fabric business case guide shows how to build a cost model that finance can trace, and our SKU guide for executives explains how to size the capacity.
Snowflake to Fabric Case Study Architecture: Before and After
Before the move, data passes through the Snowflake warehouse and then into Power BI import models. After the move, selected data lands in OneLake as Delta tables, is organised into bronze, silver and gold layers, and is read by Power BI through Direct Lake semantic models.
Representative before and after architecture, not a specific client.
Three routes get the data into OneLake, and a real migration often mixes them.
Mirroring
Database mirroring creates a synchronised Delta copy of selected Snowflake tables in OneLake. Fabric compute for replication is free and mirroring storage is free up to a capacity-based limit, but Snowflake compute is still charged when changes are read, and mirroring runs continuously. Mirror only the tables you need, and keep your Fabric capacity in the same cloud region as Snowflake to avoid egress charges and extra latency.
Iceberg shortcuts
If tables are stored as Iceberg, a OneLake shortcut can present them as Delta tables without copying data. In the other direction, Delta tables in OneLake can be read by Snowflake as Iceberg tables, which supports a coexistence period.
Pipelines and copy jobs
Use Fabric Data Factory pipelines for scheduled batch loads or when you need to transform data on the way in. If the team already uses dbt, the dbt adapter for Fabric Data Warehouse lets a project that targets Snowflake be pointed at a Fabric warehouse with a configuration change, though SQL differences still need testing.
Approach and Timeline
The plan below shows the order of work, not durations. Set durations from your own inventory, because the number of tables, reports and business owners drives the schedule far more than data volume.
| Phase | What happens | Exit signal |
|---|---|---|
| 1. Assess | Inventory objects, reports, users and costs; pick a first domain | Scope and baseline recorded |
| 2. Land | Mirror or shortcut tables into OneLake; set security and workspaces | Data readable in Fabric |
| 3. Model | Build gold tables and Direct Lake semantic models | Reports run on Fabric |
| 4. Validate | Parallel run through a full reporting cycle | Written parity sign-off |
| 5. Cut over | Move users; keep Snowflake read-only for a rollback window | Users on Fabric |
| 6. Retire | Switch off migrated workloads once contract dates allow | Cost and scope closed |
Move by business domain, not by technology layer. One domain, such as finance, goes through all six phases before the next starts. Our Microsoft Fabric migration checklist keeps each wave consistent, and the guide to migrating to Microsoft Fabric from any platform covers other sources.
Parity Validation: How to Prove the Numbers Match
Reports are trusted because the numbers agree with what people already know. Validate in three layers, and keep the evidence.
Layer 1: Tables
Compare row counts and checksums of key columns for each migrated table, and check null counts and date ranges.
Layer 2: Measures
Compare the business measures that drive decisions, such as revenue or order counts, by period and by key dimension. Agree the tolerance with the business owner before you test, so a rounding difference is not argued afterwards.
Layer 3: Reports
Open the old and new report side by side for the same filters, including a month-end close, and record any difference with its cause. Test security by signing in as users from different roles.
Security needs its own check
Mirroring does not replicate Snowflake row-level or column-level security policies, so recreate them in Fabric and test them. Row-level security also affects Direct Lake: on a SQL analytics endpoint it causes queries to fall back to DirectQuery, while a Direct Lake on OneLake model does not apply SQL-based row-level security. Plan where each rule will live.
- Use mirroring, Iceberg shortcuts or pipelines, and mix them where it helps.
- Mirror only the tables you need, and keep Fabric in the same region as Snowflake.
- Migrate one business domain at a time through all phases.
- Prove parity at table, measure and report level, with agreed tolerances.
- Recreate security in Fabric; mirroring does not copy Snowflake policies.
Snowflake to Fabric Case Study Results: What to Measure
A real case study reports measured results. This scenario deliberately does not, because any figure here would be invented. Instead, set these baselines before the migration and measure the same things afterwards.
| Metric | How to measure it | Evidence |
|---|---|---|
| Reporting cost | Total cost of Snowflake plus BI compute before, against Fabric capacity plus OneLake storage after | Invoices for the same months and scope |
| Report load time | Median and slowest-visual time for the same reports | Performance Analyzer on both platforms |
| Data freshness | Delay from source change to report | Mirroring status and refresh logs |
| Parity | Share of checked measures inside tolerance | Parity test log |
| Data copies | Number of places the same table is stored | Platform inventory |
| Incidents | Failed loads and report errors per month | Monitoring and ticket history |
Compare like with like: the same months, the same reports and the same scope. Include Snowflake costs you keep, because a partial move can leave both bills in place for a while.
Lessons Learned from Snowflake to Fabric Migrations
Watch for reseeds
A reseed is a full reload of a table. DDL changes, tools such as dbt that drop and recreate tables on a schedule, and stopping and restarting mirroring can all trigger one, and each costs Snowflake compute in proportion to table size. Schedule schema changes outside mirroring windows and monitor the mirroring status page.
Check unsupported objects early
External, transient, temporary and dynamic tables are not mirrored, and views sync every 12 hours. Find these in the inventory, because they often sit behind important reports.
Mirrored data is read-only
You cannot add calculated columns on a mirrored database. Reference the mirrored tables with shortcuts from a lakehouse and add logic there.
Do not retire too soon
Keep Snowflake read-only for a rollback window after cutover, and check contract end dates early so the timeline and the renewal date are aligned.
Snowflake to Fabric Case Study FAQ
Is this a real client case study?
No. It is a representative scenario built from the way Snowflake to Fabric migrations are usually structured, using Microsoft documentation for the facts. It contains no client names or measured client figures. Real results depend on your data, workloads and contracts, so measure your own baseline.
Do I have to migrate everything from Snowflake?
No. Fabric can mirror Snowflake tables into OneLake, and it can read Snowflake Iceberg tables through shortcuts, so you can move reporting first and keep other workloads where they are. A full move is one option, not the default.
How does Fabric mirroring from Snowflake work?
Mirroring continuously replicates selected Snowflake tables into OneLake as Delta tables. Fabric compute used for replication is free, and mirroring storage is free up to a limit based on your capacity size. Snowflake warehouse compute and cloud services costs still apply while changes are read.
What does mirroring not copy?
Row-level and column-level security policies are not replicated, so you must recreate equivalent security in Fabric. External, transient, temporary and dynamic tables are not supported, and views and materialized views sync every 12 hours.
Which Power BI mode should reports use after the move?
Direct Lake reads Delta tables from OneLake without importing a copy. Direct Lake on OneLake does not fall back to DirectQuery, while Direct Lake on SQL analytics endpoints can fall back when SQL row-level security or non-materialised views are involved. Test report performance before you cut over.
How do I prove the numbers match?
Run both platforms in parallel for at least one full reporting cycle, such as a month-end close. Compare row counts, key aggregates and the finished reports, agree a tolerance with the business, and record sign-off before you retire anything.
To plan your own move, see our guide to choosing a Microsoft Fabric migration partner, our Snowflake services and our Microsoft Fabric migration service, or request a free migration assessment.