Moving SSIS and SSAS to Microsoft Fabric without breaking the reports.
Microsoft's investment has moved to Fabric. SSIS is in maintenance mode, and the automated converters handle about half of a real package library. This guide covers what converts, what does not, and how to prove the new pipelines match the old ones before anyone depends on them.
- Fabric Data Factory is Microsoft's recommended landing zone
- Legacy Integration Services components are deprecated
- Expect 40 to 60 percent automated conversion on real portfolios
- Multidimensional cubes need rebuilding as tabular models
- Old and new run side by side until every output reconciles
Where the Microsoft BI stack stands in 2026
SQL Server Integration Services still ships, but it is no longer where Microsoft invests. With the 2025 release, the Legacy Integration Services Service and the SSIS Package Store were deprecated, and new integration capability lands in Fabric Data Factory first. For Analysis Services, the direction is Power BI semantic models in Fabric, with Direct Lake reading straight from OneLake. Reporting Services reports map to Power BI paginated reports.
Fabric Data Factory, not standalone Azure Data Factory, is now the recommended target for new pipelines. Azure-SSIS Integration Runtime still exists as a lift-and-shift bridge, and it has a place for packages that will be retired soon anyway, but it keeps you on a legacy runtime at cloud prices.
What converts automatically, and what does not
Vendor demos show high conversion rates. On production portfolios, the realistic figure is 40 to 60 percent. The remainder is where the work, and the risk, lives.
Straightforward data flows
Source-to-staging copies, file ingestion, simple lookups and derived columns, and packages that follow a consistent pattern across many tables. These map cleanly to Fabric pipelines and dataflows.
Script tasks and custom components
C# and VB script tasks, third-party components, complex expressions, package configurations and transaction handling have no automatic equivalent. Each one is a small piece of logic that has to be understood and rebuilt, usually in a notebook or a dataflow.
Row-by-row and procedural logic
Packages that loop over rows, call stored procedures per record, or depend on execution order across packages. These are better redesigned as set-based transformations than translated line by line.
Multidimensional cubes and MDX
SSAS multidimensional has no direct equivalent in Fabric. Cubes are rebuilt as tabular semantic models, and reports that depend on MDX calculations are re-expressed in DAX. This is where report outputs most often drift, so it is where reconciliation matters most.
The reconciliation discipline
A converted pipeline that runs without errors is not a migrated pipeline. It is a candidate. The legacy packages keep running, the new pipelines run beside them, and every output is compared: row counts per table, control totals on every measure, hashes of key columns, and the actual numbers on the reports executives read.
We run that comparison through at least one full business cycle, usually a month-end, so that the slow-changing dimensions, late-arriving facts and year-to-date calculations are all exercised. Only when the new output matches, or the differences are explained and accepted, does a workload cut over. Rollback is simply switching the report back to the old source, because the old source is still there.
This is stage three of our method, and it is the reason our migrations do not become the ones people describe as having been burned by. See the four stages.
Fabric, Databricks or both
Fabric suits Microsoft-aligned organisations whose consumers live in Power BI and whose data engineering is SQL-shaped. Licensing is capacity-based, which simplifies cost for mid-sized estates.
Databricks suits heavy Spark, data science and machine learning workloads, multi-cloud estates, and teams that already work in notebooks.
Both is common at larger organisations: the lakehouse and engineering in Databricks, the semantic layer and reporting in Fabric, with OneLake shortcuts between them. The assessment makes this call based on your workloads and your team, not on a vendor preference.
What it unlocks
A modern lakehouse is the foundation every AI use case needs. Copilot in Fabric and machine learning on OneLake work only when the data is there and trusted. Governance improves because lineage is captured by the platform rather than reverse-engineered from package XML. And the 2016-era servers that ran SSIS and SSAS can be decommissioned, which removes an end-of-support problem at the same time. Read the SQL Server 2016 guide.
The questions you're already asking
Can we lift and shift SSIS into Azure-SSIS Integration Runtime instead?
Will our Power BI reports break?
Do we have to rebuild our cubes?
How do you prove the numbers match?
Find out how much of your SSIS and SSAS estate converts, and what needs rework.
The Legacy-to-AI Readiness Assessment inventories every package, cube and report, rates conversion effort, and gives you a sequenced plan. Fixed scope, fixed price, no obligation to continue.
Legacy-to-AI Readiness Assessment
Fixed scope. Everything delivered is yours to keep.
- Duration
- 2 to 3 weeks
- Delivers
-
- System and data risk map
- Deadline exposure: end-of-support and policy dates that touch your systems
- Sequenced modernisation plan
- 2 to 3 AI use cases for the next 90 days
- Executive readout
- Price
- Fixed, quoted up front
- Commitment
- None beyond the assessment