Kilvora
Fabric Migration

Move to Microsoft Fabric without breaking what already works.

For organisations running SQL Server and SSIS, Azure Synapse or Power BI Premium. Establish what moves, what needs redesigning and what can be retired before migration begins.

Kilvora leads the migration from your current estate to handover. It starts with a Migration Assessment and follows the Transformation Blueprint method.

SQL Server & SSIS Azure Synapse Analytics Azure Data Factory Power BI Premium (P-SKU) Alongside Databricks & Snowflake
Why now

A renewal, an estate review, or a batch that no longer keeps up.

Most migrations don't start because someone wanted a new platform. They start because a renewal, a roadmap or a failing batch forces the question.

Power BI Premium

The P-SKU renewal

A capacity or licensing review is a useful point to reassess workloads, viewer requirements and cost. Establish what the target capacity needs to support before treating migration as a licence swap.

Azure Synapse

The Synapse estate review

Dedicated SQL, serverless SQL and Spark serve different purposes. Review warehouse workloads, queries over lake files and notebook processing together, then decide which Fabric patterns fit and which workloads should stay where they are.

SQL Server & SSIS

The overnight batch

Long nightly loads, yesterday's data in this morning's reports, and failures that mean rerunning everything. These estates still work, which is exactly why they are hard to move safely.

Architecture decisions

Migrate the capability. Choose the right implementation.

A package, procedure or report is evidence of a business capability, rather than a requirement to reproduce the same design. These are candidate patterns; the Migration Assessment and the Blueprint determine the fit, including what should remain outside Fabric.

FromCandidate Fabric approach
SQL Server data warehouse→Fabric Warehouse or Lakehouse, migrated with schema conversion and reconciled table by table.
SSIS packages→Rebuilt as Fabric pipelines, Copy jobs or notebooks, with incremental loading and safe reruns. Some packages can run in Fabric for an interim period; packages that reach on-premises systems usually keep their current runtime until rebuilt.
Synapse dedicated SQL pools→Fabric Warehouse, using Microsoft's Migration Assistant for schema conversion and data movement, including incremental sync during the parallel run. Stored procedures are converted and tested before the pool is switched off.
Synapse serverless SQL pools→Lakehouse SQL analytics endpoint or Fabric Warehouse, with file access, views, SQL compatibility and security checked. Candidate patterns include querying supported files, OneLake shortcuts and curated Delta tables.
Synapse Spark pools→Fabric notebooks, Spark job definitions and environments, moved with Microsoft's Spark migration tooling where it fits. Lake databases can be mapped through OneLake shortcuts. Libraries, runtime versions and data paths are tested in the target workspace.
Synapse pipelines→Fabric pipelines, with activities, linked services, triggers and parameters converted and tested. Pipelines that call dedicated SQL or Spark move in the same wave as the work they orchestrate.
Azure Data Factory→Fabric Data Factory pipelines where suitable, using Microsoft's upgrade path and its readiness check, then reviewing what it cannot convert. Connections and schedules are configured for each environment; ADF is retained where dependencies require it.
Power BI Premium P-SKU→A Fabric F-SKU sized from measured use, including the viewer-licensing break-even, headroom for peaks and the billing model: reserved, pay-as-you-go or overage for rare peaks.
Import models on fixed schedules→Governed semantic models that refresh when their data is ready, with storage mode chosen on measured evidence.
The analytical estate is the migration scope. Operational applications and their databases may remain in place, continuing to supply data to Fabric. Retiring an old warehouse or processing service does not mean retiring its source systems.
Azure Synapse Analytics

One estate. Several migration paths.

A Synapse workspace can combine warehouse storage, SQL queries over lake files, Spark processing and the pipelines that orchestrate them. The migration plan follows their dependencies and business purpose.

Dedicated SQL

Preserve the warehouse contract

Assess tables, procedures, SQL compatibility and reporting dependencies. Keep the new warehouse in step with the old one during the parallel run, and prove business totals and performance before switching consumers.

Serverless SQL

Reassess how the lake is queried

Inventory views, external tables, file queries and storage permissions. Queries over supported files may use OPENROWSET, subject to compatibility and access checks. External table patterns are reviewed and may be replaced with file-query views or curated Delta tables.

Spark

Move processing with its dependencies

Review notebooks, job definitions, libraries, runtime versions and scheduling. Test transformations, identity, network access, runtime and capacity demand in the target environment.

Assess the connected service. A Synapse pipeline may trigger a Spark job that produces the data serverless views expose to reports. Data readiness, access controls, reconciliation and consumer cutover must work across that chain.
Migration Assessment

Turn an uncertain migration into a scope you can review.

Once you have decided to move an estate, the Migration Assessment establishes what moves, how hard each part is and in what order. It is the first phase of the engagement, and its scope is agreed around your estate. Existing assessments and designs are reviewed and reused where credible. Where it is not yet clear whether Fabric fits, start with the Readiness Assessment. The assessment produces four outputs.

01 / Estate and dependencies

Know what the business relies on

Inventory the agreed estate, usage, data flows, downstream consumers and manual interventions. Record business owners, criticality and required availability times.

02 / Decisions and compatibility

Decide what belongs where

Record a move, redesign, retire or retain decision for every item, with complexity and compatibility: SQL objects, SSIS components, Spark runtimes and libraries, file formats and storage dependencies. Identify candidate target patterns and the security, sensitive data and governance requirements the design must meet.

03 / Economics and service

Test whether the move makes sense

Baseline cost, runtime, freshness and reliability. Model capacity, licensing, storage, support and transition costs, with assumptions and uncertainty made explicit.

04 / Contractable delivery

Leave with a contractable wave plan

Sequence waves by dependency, business criticality and risk. Set acceptance criteria, cutover and rollback outlines, responsibilities, and a delivery scope with estimates and dependencies for the agreed engagement model.

From assessment to design. The Transformation Blueprint then turns these decisions into the target architecture and migration design. A review point confirms which workloads can proceed, which need attention and which should remain in place, before any fixed price for delivery is agreed.
Databricks, Snowflake and other platforms

Coexist first. Migrate only what earns it.

Not every estate needs to move to use Fabric. Where Databricks or Snowflake already does its job well, Fabric can read that data without copying pipelines across, so Power BI and Fabric workloads can use it while engineering stays where it is.

Azure Databricks

Mirror the Unity Catalog

Selected schemas and tables appear in Fabric as a mirrored catalog, without duplicating the data. Access, refresh behaviour and semantic model storage mode are designed around it.

Snowflake

Mirror into OneLake

Snowflake tables are mirrored into OneLake for reporting and analytics in Fabric. Cost on both platforms, latency and security are checked before reports depend on them.

Other sources

Assessed case by case

Teradata, Oracle and legacy ETL estates go through the Migration Assessment like any other. Whether to integrate, migrate selectively or leave them in place is decided on the evidence.

Where code has to move off another platform, conversion is planned with your platform team or a specialist in that platform. Kilvora leads the assessment, the Fabric design, the build on the Fabric side, reconciliation and cutover.
Readiness and remediation

Resolve the right gaps at the right stage.

Moving data does not resolve unclear definitions, weak ownership or uncontrolled mappings. Findings are classified by their effect on the workloads being migrated.

Before the affected workload moves

Critical prerequisites

Resolve blockers such as unapproved access, missing reconciliation rules or an untested recovery dependency. Record an owner and the evidence needed to close each item.

During design and migration

Required delivery actions

Implement agreed changes to ingestion, mappings, quality controls, security and operations alongside the relevant migration wave.

After the initial migration

Planned improvements

Keep lower-priority improvements visible in an owned backlog, with a rationale for deferral and conditions for revisiting them.

Remediation can run before or alongside migration. Data Estate Remediation addresses agreed quality, definition, ownership and mapping gaps. Workloads can proceed when their prerequisites are met; unrelated gaps need not hold up the whole estate.
How it works

Confirm the starting point, design, migrate and prove the service.

01
Migration Assessment, then the Blueprint

Assess and design

The Migration Assessment reviews existing evidence, assigns prerequisite actions and gives every item in scope a decision: move, redesign, retire or retain. The Blueprint then sets the target architecture. You get a business case and a wave plan that delivery is measured against. What the assessment covers →

02
Fixed scope or embedded

Migration in waves

Waves follow dependencies, business criticality and risk. Each wave lands in development, is tested, and is promoted through test to production with nothing hard-coded to one environment.

03
Before anything is switched off

Parallel run and cutover

Old and new run side by side, with reconciliation of row counts, totals and business measures at agreed intervals, automated where practical. Freshness, access controls and failure recovery are checked too. Cutover happens only after the agreed number of clean days, and switching back is rehearsed and timed first.

04
After cutover

Right-sizing and handover

Capacity is measured under real load and sized to it. Named owners take responsibility for monitoring, incident response, escalation and platform decisions. Your team practises the runbooks and recovery steps before handover is accepted. The old estate is retired only after retention and recovery requirements are met.

What keeps it safe

Six rules for a platform your team can operate.

Measure before you change

Nothing moves until its current run time, freshness and cost are recorded, so the improvement can be shown.

Retire before you migrate

Unused items are identified and retirement is agreed with business owners, subject to retention and regulatory obligations.

The old system stays in charge

It remains the source of truth until the new platform reconciles to it, day after day.

Rollback is rehearsed

Switching back is tested and timed before cutover, not worked out during an incident.

Recovery is proven

Restore or rebuild procedures are tested for the chosen workloads against agreed recovery time and data loss objectives.

Ownership transfers before Kilvora leaves

Named owners practise recovery, understand escalation and accept responsibility for running the platform.

How it is delivered

Day-rate engagement or fixed-scope delivery.

Readiness, remediation, architecture and implementation are included as the agreed work requires. The starting point reflects the evidence and design already in place, with review points before workloads move.

Day rate

Embedded architecture and engineering

I work alongside your team as the architect and, where appropriate, as the engineer. The Migration Assessment, design, remediation and migration progress through agreed priorities and waves, with effort and milestones reviewed as evidence develops.

  • Suits redesign work and estates with gaps in documentation
  • Your team learns the platform as it is built
Fixed price

Fixed scope platform delivery

Each wave is a bounded piece of platform delivered to an agreed definition of done, with defined outputs, dependencies and acceptance criteria. The Migration Assessment establishes the scope before the price is fixed, so the commitment is evidence-based rather than optimistic.

  • Defined deliverables, acceptance tests and reconciliation criteria for each wave
  • Anything new goes through change control
  • A small, well-understood estate can be a single fixed-scope engagement
The models can be combined. A defined migration wave can use fixed-scope delivery, while an evolving workstream uses a day-rate engagement. Responsibilities, boundaries and change control are agreed explicitly.
The business case

Establish whether the migration earns its investment.

The Migration Assessment compares a measured baseline with a target model. Savings are a conclusion to test, alongside service quality and operational risk.

01 / Current state

What does today cost?

Platform and licensing costs, operational effort, batch duration, data freshness and incident history. Identify costs that will remain after migration.

02 / Target and transition

What will the move require?

Model recurring capacity, storage, licensing and support, plus one-off delivery, parallel running, training and retirement costs.

03 / Decision

What improves, and when?

Compare net recurring cost, service outcomes and risk. Calculate payback where there is a net saving, and test how it changes with demand and delivery assumptions.

A credible case includes the uncomfortable numbers. Dual-running costs, retained systems and ongoing support belong in the model. If the economics do not support migration, the assessment says so.
Questions

Migration questions.

Can you migrate a Synapse estate that uses serverless SQL and Spark?

Yes. The Migration Assessment covers serverless SQL, Spark and Synapse pipelines alongside dedicated SQL where they fall within the agreed scope. Serverless SQL queries are checked against the target data and query model; Spark processing is checked for runtime, library, security and orchestration compatibility; pipelines are mapped to the work they orchestrate. Shared dependencies determine the wave plan, and any workload that should remain in Synapse is recorded explicitly.

Do we have to migrate everything at once?

No. The Migration Assessment produces a wave plan, and each wave runs in parallel with the old system before anything is switched off. Many organisations move the warehouse and its loads first and leave reporting until the data has reconciled.

Can our SSIS packages run in Fabric as they are?

Do not assume that every package can move unchanged. Fabric has a preview activity that runs SSIS packages stored in OneLake, but it cannot currently reach on-premises systems or private endpoints, and it does not support custom components. The Migration Assessment checks each package's connectors, components and network dependencies, then recommends a rebuild, an interim run in Fabric, or keeping the current runtime until a rebuild is due.

Will our users notice the change?

The agreed acceptance criteria cover business measures, availability and any planned changes in report behaviour. Report numbers are checked against the old reports before cutover, and anything that will look different is agreed with you in advance.

How long does a migration take?

It depends on the estate, which is why it starts with the Migration Assessment. Its scope and timing are agreed around your estate, and its wave plan gives you a timeline for each wave.

How is the Migration Assessment different from the Readiness Assessment?

The Readiness Assessment answers whether Fabric fits your workloads and what must change first. The Migration Assessment starts once the move is decided. It establishes what exactly moves, how complex each part is, what depends on what, what it will cost and the order of the waves. If Fabric suitability is already settled, you can start with the Migration Assessment.

We run Databricks or Snowflake. Do we have to migrate?

Not necessarily. Fabric can mirror an Azure Databricks Unity Catalog or a Snowflake database, so Power BI and Fabric workloads can use that data without moving your pipelines. The Migration Assessment establishes whether anything should move, workload by workload. Where code does need converting, it is planned with your platform team or a specialist in that platform.

What do you need from us?

Access to the estate in your tenant with least-privilege permissions, someone who knows the business meaning of the data, and a person who can sign off reconciliation results. Client data remains in your approved environment; evidence handling and any exports are agreed upfront.

Can you work from an assessment or architecture we already have?

Yes. Existing evidence, decisions and specifications are reviewed for completeness and relevance to the migration scope. Credible work is reused, gaps are investigated and required changes are agreed. A new Readiness Assessment, or a full Migration Assessment from scratch, is not automatically required.

We've already switched from Premium to Fabric. Is this still relevant?

Often, yes. A straight SKU swap leaves the old refresh patterns in place. If the capacity is struggling or the cost is higher than expected, start with the Cost & Performance Audit; it establishes workload behaviour and whether optimisation should precede further migration.

Next step

Discuss the estate you need to migrate.

A short conversation covers your estate, objectives, existing assessments and deadlines, and how a Migration Assessment would be scoped. Kilvora then leads the migration through a day-rate engagement or fixed-scope delivery.