The short answer
An Azure Data Factory to Fabric migration now starts with Microsoft's own tool: you mount the factory in a Fabric workspace, run the built-in assessment, and move the pipelines marked Ready into Fabric Data Factory with their triggers switched off. Self-hosted integration runtimes, metadata-driven designs with parameterised linked services, some triggers and a few activities still need hand work, and the bill changes from pay-per-run to a fixed Fabric capacity. Most mid-sized factories can be moved in phases over a few months, but there is no deadline forcing you to do it this year.
That last point matters. Microsoft has not announced a retirement date for Azure Data Factory, and its own guidance says ADF and Fabric can run side by side. The reason to move is where new investment is going: Copilot, OneLake, Direct Lake reporting and built-in deployment pipelines arrive in Fabric, not in ADF. The reason to wait is that some of what your factory does today is still in preview or not supported.
This guide covers what the tool really moves, how to triage a factory pipeline by pipeline, a worked cost and effort example for a 60-pipeline estate, the cutover method we use, and the cases where staying on ADF for another year is the right call. It is written for data and analytics leads at mid-sized companies, including pharma and life sciences teams running commercial and patient data through ADF.
Numbers to have in front of you
Azure Data Factory vs Fabric Data Factory: what actually changes
Fabric Data Factory looks familiar, but it is not ADF with a new logo. Microsoft's upgrade planning guide lists the architectural differences that drive most of the effort:
| Area | Azure Data Factory | Fabric Data Factory | What it means for you |
|---|---|---|---|
| Datasets | Separate, reusable objects | Defined inline inside each activity | Shared dataset logic is repeated per activity after migration |
| Linked services | Can be fully parameterised | Connections do not take dynamic properties; activities can pass dynamic content to the connection object | Metadata-driven frameworks need redesign, not a straight copy |
| Global parameters | Global parameters | Fabric variable libraries | Different patterns and data types; Microsoft publishes a conversion guide |
| Identity | Managed identity | Workspace identity | Source permissions must be re-granted to the new identity |
| Key Vault | Mature, all auth types | Narrower Key Vault reference support | Check every secret your factory reads before you plan the move |
| Scheduling | Shared triggers, managed centrally | Schedules set per pipeline, no shared trigger hub | One trigger running ten pipelines becomes ten schedules |
| On-premises access | Self-hosted integration runtime | On-premises data gateway | A new install, new networking checks and a new owner |
| Mapping data flows | Spark-based data flows | Mapping data flow transforms in Dataflow Gen2, in preview | Eligible flows migrate; others are rebuilt in Power Query, SQL or notebooks |
| Child pipelines | Execute Pipeline activity | Invoke Pipeline activity | Renamed, with a different connection requirement |
| Custom code | Custom activity | Azure Batch activity | Same function under a new name |
Some activities have no direct equivalent. Microsoft lists U-SQL (Data Lake Analytics, itself a deprecated service) and the Validation activity, which you rebuild with Get Metadata, loops and If conditions. Change data capture moves to Fabric Copy jobs, Airflow DAGs move to Fabric's Apache Airflow jobs, and SSIS packages keep running in ADF and are called from Fabric. If your factory depends heavily on any of these, the migration is a redesign project, and it should be budgeted as one.
The three ways to migrate
Mount first
Adding your existing factory to a Fabric workspace as an Azure Data Factory item gives a live, read-only view alongside your Fabric work. Execution stays in Azure. It costs nothing in risk and is the right first step for every estate: you can catalogue what exists, assign owners and plan cutover by domain.
Built-in upgrade
From ADF or from the mounted item, run the assessment, choose the pipelines you want, map linked services to Fabric connections and migrate. Connections are created automatically for common sources such as Blob Storage, ADLS Gen2, Azure SQL and Cosmos DB. Migrated schedule triggers arrive disabled so nothing double-loads.
Manual rebuild
Pipelines with unsupported activities, parameterised connection frameworks or event triggers are rebuilt by hand. This is slower but often the moment to remove duplicate pipelines, adopt OneLake shortcuts instead of copies, and move transformation into a lakehouse or warehouse.
How to triage a factory, pipeline by pipeline
Run the assessment, export the result and give every pipeline one of four actions. The status tells you what the tool thinks; the business owner tells you whether it matters.
| Assessment status | Default action | Typical effort | Watch for |
|---|---|---|---|
| Ready | Migrate with the built-in tool, then validate | Low: mostly testing | Hidden dependencies on shared triggers, global parameters or Key Vault secrets |
| Needs review | Migrate, then fix flagged settings by hand | Medium | Connection mapping, identity changes, dynamic expressions that reference datasets |
| Coming soon | Leave in ADF, mounted in Fabric, and re-check each quarter | None now | Do not rebuild something Microsoft is about to support |
| Not compatible | Rebuild, replace, or keep in ADF if the pipeline is stable | High | U-SQL, Validation activities, unsupported connectors, event triggers |
Before any of that, delete what nobody uses. Most factories that have run for a few years carry pipelines that load tables no report reads. Cross-check the last run date and the downstream consumers, retire the dead ones, and the migration gets smaller before it starts.
Worked example: cost and effort for a 60-pipeline factory
The figures below are planning assumptions for illustration, not benchmarks. They show how we size an ADF to Fabric migration for a mid-sized company with one factory, a self-hosted integration runtime for an on-premises ERP, and about 40 Power BI users.
Assessment result. After retiring unused pipelines, 60 remain: 35 Ready, 12 Needs review, 8 Coming soon and 5 Not compatible.
| Work item | Assumption | Person-days |
|---|---|---|
| Validate Ready pipelines | 35 x 0.5 day | 17.5 |
| Fix Needs review pipelines | 12 x 2 days | 24 |
| Rebuild Not compatible pipelines | 5 x 4 days | 20 |
| Coming soon pipelines | Stay in ADF, mounted | 0 |
| On-premises data gateway to replace the SHIR | Install, networking, tests | 3 |
| Global parameters to variable libraries | One library per environment | 2 |
| Parallel run and reconciliation | Two weeks of side-by-side loads | 10 |
| Cutover and hypercare | Switch schedules, watch first runs | 3 |
| Total | 79.5 |
That is roughly 16 weeks for one engineer, or eight to ten weeks with two working in parallel, with the 8 Coming soon pipelines left in ADF until Microsoft supports them.
Run cost. ADF bills per activity run, per data integration unit hour and per data flow vCore hour. Fabric bills a capacity whether or not pipelines are running, although pay-as-you-go capacities can be paused. If testing shows the estate fits on an F8, the list price is $1,051.20 a month on pay-as-you-go or $625.33 a month on a one-year reservation, according to AlphaBold's August 2026 pricing guide. That capacity is shared with any lakehouse, warehouse and Power BI work you put on it. Below F64, the 40 report viewers still need Power BI Pro, which Microsoft lists at $14 per user per month, so $560 a month stays in the budget either way.
Now change one assumption. If the factory is metadata-driven, with one generic pipeline reading a control table and parameterised linked services, most of it will land in Needs review or Not compatible, and the effort can double. That design choice, not the pipeline count, is the number to check first.
How to run the migration, step by step
Inventory and clean up
List every pipeline, linked service, integration runtime, trigger and global parameter. Retire pipelines with no recent runs or no downstream consumers.
Mount the factory in Fabric
Add the factory as an Azure Data Factory item in a Fabric workspace so the team can see ADF and Fabric work side by side while nothing changes in production.
Run the assessment and triage
Export the readiness status for every pipeline and assign migrate, fix, wait or rebuild, using the table above and the business owner's view of criticality.
Build the landing zone
Create workspaces per environment, Fabric connections, the on-premises data gateway, workspace identity permissions and variable libraries before the first pipeline moves.
Migrate in waves
Start with low-risk copy pipelines in non-production, then move by business domain. Keep finance and month-end loads for a later wave.
Run in parallel and reconcile
Run old and new pipelines against the same sources and compare row counts, totals and key fields in the target tables until they match for an agreed period.
Cut over and switch off
Enable Fabric schedules, disable the ADF triggers for the same pipelines, watch the first runs closely and keep a documented way back for each wave.
Common ADF to Fabric migration mistakes
Treating the assessment as the plan
Ready means the tool can move it, not that it will behave the same. Shared triggers, secrets and identity still need checking.
Forgetting the gateway
The self-hosted integration runtime does not migrate. The on-premises data gateway is a new install with its own owner, updates and network rules.
Double loading
Enabling Fabric schedules before disabling ADF triggers runs the same load twice and can duplicate rows in append-only tables.
Sizing capacity by guesswork
Pick a capacity from measured consumption on a trial or test workload, not from the SKU a vendor slide suggests.
Rebuilding what is coming
Rewriting a Coming soon pipeline by hand burns effort Microsoft may save you a quarter later.
No reconciliation sign-off
Without written proof that old and new outputs match, the first odd number in a report will be blamed on the migration.
When not to migrate yet
Regulated data, and when to bring in Azure Data Factory consultants
For pharma and life sciences teams, the pipelines that matter most are usually the ones loading syndicated prescription data, CRM activity, specialty pharmacy and hub feeds into a commercial data warehouse. Those loads have vendor delivery windows, restatements and licence terms, and some carry patient-level data. A migration has to preserve the late-file and duplicate-file checks, keep access as narrow as it is today under workspace roles and sensitivity labels, and prove that weekly brand numbers did not move. Our guide to the pharma commercial data warehouse covers what those loads feed.
Many teams do the Ready pipelines themselves and bring in Azure Data Factory consultants for the parts that need judgement: the triage, the gateway and identity setup, rebuilding metadata-driven frameworks and running reconciliation. If you are budgeting for outside help, our breakdown of Microsoft Fabric consultant cost shows how rates and project totals are built. If the move is also a chance to change the reporting layer, see Power BI consulting, and for pipeline work on other platforms see our ETL services.
Greenwolf's founder came up through pharma commercial operations consulting at ZS, and our Microsoft Fabric consulting and Azure data engineering team plans and runs these migrations in your own tenant. We are an independent consultancy and do not claim a Microsoft partnership; we will also tell you when staying on ADF is the cheaper answer.
Questions about migrating Azure Data Factory to Fabric
Is Azure Data Factory being retired?
How do I migrate Azure Data Factory pipelines to Fabric?
Does the self-hosted integration runtime migrate to Fabric?
Can mapping data flows be migrated to Fabric?
Is Fabric cheaper than Azure Data Factory?
How long does an ADF to Fabric migration take?
What happens to ADF global parameters in Fabric?
Want a straight answer on whether to move your factory?
Send us the shape of your ADF estate: pipeline count, integration runtimes and what the loads feed. We will tell you what the assessment is likely to show, what it would take to migrate, and whether waiting is the better call.
Part of our Data Management and Engineering hub. Start with our data management services.