Microsoft Fabric for mid-market · from assessment to adoption.
A guide built from our active project portfolio in Belgium and Estonia · framework, architectures, pitfalls and pricing for organisations between 50 and 500 employees.
What is Microsoft Fabric?
Microsoft Fabric consolidates Power BI, Synapse, Data Factory and parts of Azure ML into a single SaaS platform. Implementing it well in a mid-market context is harder than the marketing suggests.
Most Microsoft Fabric content is written for Fortune 500 enterprise teams, organisations with dedicated data engineering departments, separate BI and ML platforms, mature governance frameworks, and capacity to absorb a 12-month implementation programme. That is not who we work with.
One of the niches that Sparkle's approach fits really well is organisations between 50 and 500 employees, the "upper SME" segment in Belgium and Estonia. Generic Fabric playbooks designed for enterprises lead growing teams toward over-engineered architecture and over-provisioned capacity. This guide draws on our active project portfolio and is tuned for the constraints that 50 to 500 employee organisations actually face, framework, architectures, pitfalls and pricing included.
When Fabric is not the right call yet. If you have a single Power BI workspace, a few hundred megabytes of data and no roadmap to grow, you probably do not need Fabric capacity at all: Power BI Pro is enough. Fabric earns its place once you have multiple sources to unify, a governance or compliance need, real-time or AI ambitions, or a legacy platform (Qlik, SAP BO, on-prem SQL) to retire.
Sparkle's Microsoft Fabric implementation framework.
Five phases · each with a clear deliverable, a defined timebox, and the option for the customer to stop or change direction.
Assessment · Fabric Ideation Session
Single working session: map business goals, audit data sources, identify the three to five highest-impact use cases, sketch the roadmap concept. The output becomes the working document for the rest of the project.
Architecture and design
Choose between two reference architecture packages, size capacity, design security and access. An architecture document your tech team can verify and your CFO can budget against.
Build and migration
OneLake setup, Lakehouses with medallion (Bronze, Silver, Gold), ingestion pipelines, semantic model for Direct Lake, legacy migration. Iterative · every two weeks we ship a slice the business can use.
Adoption and enablement
Where many platform investments quietly fail. We train your team, document runbooks, set up a citizen developer programme if appropriate, and stay engaged while first real workloads go live.
Managed run
Optional · ongoing managed service for capacity, monitoring, optimisation, evolution. Senior team, from kickoff through delivery to support · same partner, no body-shop swap-out.
One integrated architecture, two Lakehouse packages.
Microsoft Fabric is built around a single unified architecture. Real-time analytics, AI workloads and traditional reporting all live in the same OneLake · they are capabilities you switch on as needed, not separate platforms to choose between.
Fabric-native Lakehouse
For teams without existing dbt or analytics-engineering practice · ship fast on Microsoft's stack with everything integrated out of the box.
- Storage, OneLake medallion (Bronze, Silver, Gold) on Delta Parquet
- Transformation, Notebooks + Pipelines, no separate orchestrator
- Analytics layer, Semantic model in Power BI Direct Lake · sub-second performance
- Governance, Microsoft Purview · classifications, lineage, sensitivity labels
- Capacity, Right-sized at F2 to F64 for the first three months
Lakehouse + dbt
For teams already on dbt or with strong analytics-engineering culture · keep dbt for transformation logic, use Fabric for storage, BI and AI.
- Storage, OneLake medallion on Delta Parquet (same backbone)
- Transformation, dbt Cloud or dbt Core, Fabric Warehouse as compute
- Analytics layer, Semantic model in Power BI Direct Lake · sub-second performance
- Governance, Microsoft Purview + dbt-native testing, documentation, lineage
- Capacity, Right-sized at F2 to F64 for the first three months
One integrated architecture. Real-time analytics (Eventstream + KQL), AI workloads (Azure AI Foundry), and traditional reporting (Power BI) all live in the same OneLake. Switch them on as you need them · no separate platform decisions, no double migrations.
Get data in without rebuilding ETL. Two recent Fabric capabilities matter for mid-market: Mirroring replicates an operational database (Azure SQL, Cosmos DB, Snowflake and more) into OneLake in near real time at no compute cost, and OneLake shortcuts query data where it already lives (ADLS, S3, Snowflake) without copying it. Both let you light up Fabric before migrating anything.
Design for Direct Lake, do not assume it. The sub-second performance needs a well-modelled Lakehouse and semantic model built together; under certain conditions queries fall back to DirectQuery and lose the speed. And even small teams benefit from deployment pipelines and Git (Dev, Test, Prod): changes stay reviewable and nobody edits reports straight in production.
Common pitfalls · and how to avoid them.
Patterns from our active project portfolio · the mistakes that cost the most time and money are the same in almost every engagement.
Over-provisioning capacity at the start
Teams often buy F32 or F64 to be safe. For most mid-market projects, F8 or F16 is enough for the first three months. Capacity scales up easily, much harder to defend at budget review when usage is 15%.
Trying to migrate everything at once
The temptation to lift-and-shift all reports in one go is real and almost always wrong. Pick one business domain (sales, finance, ops), migrate it end-to-end including users, learn what worked, then expand.
Treating Fabric like Azure Synapse Analytics
Synapse muscle memory leads to over-engineered architectures. Fabric's SaaS posture means many design patterns from Synapse (separate compute pools, network gateways, storage account configs) are no longer needed. Trust the platform's defaults more than you think.
Ignoring data governance until after launch
Governance added late costs three to five times more than governance baked in. Set up data ownership, classification, sensitivity labels and Microsoft Purview integration in Phase 2, not Phase 5.
Underinvesting in adoption and training
The platform is only as valuable as the number of people who use it well. Budget at least 15-20% of total project cost for adoption, training, and citizen developer enablement.
Not planning for capacity cost monitoring
Without FinOps tagging and capacity governance from week one, capacity bills creep up unnoticed. Auto-pause non-prod, tag every workload, review monthly · or watch the budget walk away.
Building Lakehouses without semantic models for Direct Lake
The 10x performance benefit of Direct Lake on Power BI requires a Lakehouse + semantic model designed together. Skip the model and you lose the magic · Lakehouse becomes just another storage tier.
What does Microsoft Fabric actually cost?
Capacity is the platform bill. The build is a separate conversation.
Microsoft Fabric is licensed by capacity (F-units) · F2, F4, F8, F16, F32, F64, F128, F256, F512, F1024, F2048. You pick a tier and pay per hour of capacity, with the option to pause non-prod environments to lower the bill.
For mid-market: capacity typically lands in the F2-F64 range, with a sweet spot around F8-F16. Most customers under 500 employees stay below F32 for the first 12 months. Scale up only after monitoring shows you need it · scaling up takes minutes, not weeks.
The Sparkle build cost is separate · framework-driven, fixed-price options, transparent. Scoped to a focused first wave: one business domain end-to-end, including adoption, then scaling with scope from there.
West Europe pay-as-you-go, per month · pause non-prod to lower it · reserved capacity (1-year) cuts roughly 40%.
The biggest mid-market cost lever: from F64 upward, report viewers no longer need individual Power BI Pro licences (the capacity covers them). Below F64, every viewer needs a Pro licence. For teams with dozens of viewers this often flips the total cost of ownership, so calculate the break-even between a smaller capacity plus many Pro licences and F64 with none.
Customer outcomes · Nabuminds.
Microsoft Fabric mid-market in practice · how a fast-growing Belgian-Estonian company moved from complexity to control.
From complexity to control · a scalable analytics transformation
Nabuminds was scaling fast and buckling under fragmented data. We built a Microsoft Fabric medallion lakehouse on OneLake, a Power BI semantic model with Direct Lake performance, and a Purview-governed foundation that the team could keep extending. Self-service for 40+ users in eight weeks.
Ready for a Fabric Ideation Session?
Three hours, your data, our seniors. We map your three highest-impact use cases, sketch the architecture, size your capacity, and give you a clear go/no-go on whether Fabric is right for your stage.