01
Semantic-layer architecture
Composite models, calculation groups, and shared datasets in Premium / Fabric capacity. Reports inherit from the model rather than each rebuilding the same logic. Documented model ownership and contribution governance.
Enterprise Platforms · ATLAS
Power BI Consulting Services.
Semantic-model-first Power BI architecture: composite models, RLS / OLS, calculation groups, and DAX optimization. Reports inherit from a governed semantic layer rather than competing for the same data through ad-hoc Power Query.
- Practice
- Enterprise Platforms
- Department
- ATLAS
The problem
Most Power BI estates are two thousand reports, all slightly different.
The familiar shape: a Power BI tenant with hundreds of workspaces; reports that each query the data warehouse independently with their own Power Query and DAX; conflicting truth across executive dashboards; performance issues that show up only at month-end; row-level security applied inconsistently across reports that share a data source; and DAX measures duplicated dozens of times with subtle variations. Per-report development is how Power BI estates accumulate technical debt.
We design Power BI estates the way enterprise data platforms are designed: with a governed semantic layer at the center, deployment pipelines that enforce model ownership, RLS / OLS that lives in the model rather than in individual reports, and a tenant-level governance frame that scales. The investment is real up front; the dividend is reports that don't break each other when the source schema evolves and a tenant that compounds analytical value rather than accruing report sprawl.
Where it ships
5 use cases, in production.
Specific applications we’ve built and operated. Not speculative: every example below is grounded in a real shipped engagement.
- 02
Power BI estate cleanup
Audit and remediation of existing tenants. Workspace structure rationalization, dataset consolidation, RLS / OLS implementation, performance tuning, and the deprecation plan for legacy Excel-driven reporting.
- 03
Embedded analytics
Power BI Embedded into customer-facing applications with row-level security per tenant, custom theming aligned to product design, and a deployment pipeline that survives report iterations.
- 04
10x
report performance
DAX optimization
Performance triage on slow reports. Storage engine vs formula engine analysis, calculation-group migration from sprawling measures, and aggregation tables for high-volume facts. Most engagements achieve 10x query-time reduction on the worst offenders.
- 05
Fabric migration
Migration of existing Power BI estates onto Microsoft Fabric. Lakehouse / warehouse extraction, OneLake-based shared datasets, and the integration story across Synapse, Data Factory, and Real-Time Analytics.
How we engage
4 phases, named in the SOW.
Each phase has a deliverable, an owner, and an acceptance criterion. Not slogans, operating rules.
- 01
Tenant assessment
Audit of existing Power BI tenant: workspace topology, dataset inventory, report duplication, RLS / OLS posture, performance hotspots, capacity utilization. Output is the prioritized remediation roadmap.
- 02
Semantic-model design
Composite model design, calculation-group strategy, RLS / OLS pattern, and the dataset ownership / contribution model. Documented model contracts that downstream reports consume.
- 03
Deployment pipelines and governance
Deployment pipelines (Dev → Test → Prod) for shared datasets and report packages. Tenant-level governance: workspace creation policy, certification process, sensitivity labels, and audit. Capacity utilization monitored.
- 04
Cadence after launch
Quarterly model evolution review, performance triage on incoming reports, and capacity-utilization optimization. The semantic layer compounds value when it has a continuous owner.
Capabilities
What’s in scope.
- Semantic model architecture: composite models, calculation groups, hybrid
- DAX engineering: optimization, storage engine analysis, aggregation tables
- Row-level / object-level security with identity integration
- Power BI deployment pipelines and source-controlled dataset definitions
- Power BI Embedded for customer-facing applications
- Fabric architecture: OneLake, lakehouse, warehouse, Real-Time Analytics
- Tenant governance: workspace policy, certification, sensitivity labels, audit
Stack
Tools we use in production.
- Power BI
- Power BI ServicePremiumFabricEmbeddedReport Server
- Modeling
- Power QueryDAXTabular EditorDAX StudioTabular Object Model
- Source platforms
- SynapseSnowflakeDatabricksBigQueryAzure SQLPostgres
- Fabric
- OneLakeLakehouseWarehouseData FactoryReal-Time Analytics
- DevOps
- Power BI Deployment PipelinesAzure DevOpsPower BI REST APIPBIP source files
Selected work
Quantified outcomes, not adjectives.
Common questions
Asked before the first call.
- 01
Semantic model or per-report, what's the difference?
Per-report development is how Power BI estates accumulate technical debt. Each report owns its own Power Query and DAX, often duplicating logic that lives in other reports. Semantic-model-first development centralizes the data model, business logic, and security in a governed dataset; reports become thin views over that layer. The investment is real up front; the dividend is an estate that scales without breaking.
- 02
Do we need Premium or Fabric?
Depends on workload. Pro licenses work for small teams with simple data volumes. Premium Per User (PPU) is a cost-effective middle ground. Premium / Fabric capacity is warranted when you need shared datasets, deployment pipelines, larger model sizes, or paginated reports. We model the licensing and recommend the right tier with explicit cost projections.
- 03
Can you fix our slow reports?
Yes: DAX optimization is one of our most common engagements. Storage engine vs formula engine analysis, calculation-group migration, aggregation tables for high-volume facts, and Power Query optimization. We typically achieve 10x query-time reduction on the worst offenders within a 6–10 week engagement.
- 04
How do you handle RLS / OLS?
RLS at the model layer, not in individual reports. Identity integrated with Microsoft Entra ID (or USERPRINCIPALNAME-based dynamic RLS where the org's identity model warrants it). OLS for column-level sensitivity. Tested through dedicated test users; security validated as part of CI before deployment.
- 05
Do you do Fabric migrations?
Yes: Fabric migrations are increasingly common as customers adopt OneLake and the unified analytics platform. We design the migration as a phased program: lakehouse extraction first, then dataset re-platforming, then report cutover. Most migrations land in 4–8 months with no analytical reporting downtime.
- 06
What does a Power BI engagement cost?
Tenant assessment: 2–4 weeks, $25K–$60K. Semantic-layer rebuild: 3–6 months, $200K–$700K. Power BI Embedded build: 2–4 months, $150K–$500K. DAX / performance optimization: 6–10 weeks, $80K–$200K. Fabric migration program: 6–12 months, $400K–$1.5M. Brackets published honestly so visitors self-qualify before the first call.
Within Enterprise Platforms
Other capabilities in this practice.
Talk to us
Bring a Power BI problem. We’ll bring a senior engineer.
A senior engineer plus the ATLAS department lead joins the first call. No discovery gauntlet, no junior reps.