Microsoft Fabric ADS Session

Phase-Specific Fabric Questions

Phase 1: Context Discovery

Ask about:

• Business problem or opportunity driving this initiative
• Industry and regulatory context
• Greenfield project vs. migration from existing system (Synapse, Power BI Premium, on-prem SQL, Databricks)
• Key stakeholders and decision-makers
• Timeline and budget constraints
• Success criteria (what does "done" look like?)
• KPIs, latency targets, and cost envelope
• Microsoft 365 licensing (E3/E5) — determines Fabric capacity entitlements

Adapt: If user mentions migration, read references/migration-patterns.md. If user names a specific industry, read references/industry-templates.md for starter context.

Phase 2: Current Landscape

Ask about:

• Data sources (databases, APIs, files, streams, SaaS platforms)
• Current data platform (if migrating: Synapse, Power BI Premium, ADF, Databricks, on-prem SQL)
• Data volumes and growth rate
• Real-time vs. batch requirements
• Existing Power BI usage (workspace count, semantic models, user count)
• Data governance and cataloging needs (Purview integration)
• Sensitive data classification (PII, PHI, financial)

See references/probing-questions.md for deep-dive question banks when the user's answers are vague or incomplete.

Phase 3: Security & Networking

Ask about:

• Network topology (Managed Private Endpoints, Managed VNet, on-premises gateway)
• Identity provider (Entra ID, conditional access policies, guest access)
• Data access control model (workspace roles, item-level sharing, row-level security, OLS)
• Regulatory compliance (HIPAA, SOC2, GDPR, FedRAMP, industry-specific)
• Encryption requirements (at-rest Microsoft-managed, customer-managed keys via Fabric CMK)
• Sensitivity labels and information protection policies

Phase 4: Operational Requirements

Ask about:

• HA/DR requirements and RPO/RTO targets
• Multi-region or single-region deployment
• Capacity sizing priorities (F SKU selection, autoscale, burst capacity, smoothing)
• Monitoring and alerting requirements (Fabric Capacity Metrics app, Azure Monitor)
• Environment strategy (dev/test/prod via Deployment Pipelines)
• Tagging and cost allocation strategy (capacity-level, workspace-level)
• Operating model: who owns data products, who approves access, who triages incidents

Use references/trade-offs-and-failure-modes.md for domain-specific failure scenarios to raise proactively during this phase.

Readiness Gate

After each phase, internally track information completeness using references/readiness-checklist.md.

Decision Logic

IF all must-have items are gathered:
    → Proceed to diagram generation
IF some should-have items are missing:
    → State assumptions explicitly, ask user to confirm or correct
IF must-have items are missing:
    → Ask targeted follow-up questions (max 2 additional turns)
IF user says "just generate something" or expresses impatience:
    → Generate with sensible defaults, document all assumptions

Always tell the user what you know and what you're assuming before generating.

Fabric Diagram Components

When generating diagrams, use these Fabric-specific node shapes (extends the generic style guide from the architecture-diagramming skill):

Pattern Selection

Match gathered requirements to a Fabric architecture pattern. Read references/fabric-patterns.md for the pattern catalog.

Common patterns:

Generating the Diagram

Generate a Mermaid flowchart diagram based on the gathered requirements. Use the pattern templates in references/fabric-patterns.md as a starting point, then customize based on the specific requirements gathered.

Follow the architecture-diagramming skill's style guide for general Mermaid conventions (arrow styles, subgraph naming, layout direction). Apply the Fabric-specific node shapes listed above.

For rendering, Architecture Recap format, and iteration workflow, defer to the architecture-diagramming skill.

Optional: Workload Profiling

These questions are not a mandatory phase — the customer may or may not raise workload-specific topics during the session. Have this content ready to deploy when the conversation naturally moves toward workloads, but do not force it as a separate phase.

If the customer discusses workloads, ask about:

• ETL/ELT pipelines (Spark Notebooks vs Dataflows Gen2 vs Data Factory Pipelines — complexity, frequency, SLAs)
• SQL analytics workloads (ad-hoc queries, warehouse sizing, concurrency)
• BI/reporting and self-service analytics (Power BI user count, semantic model complexity, Direct Lake eligibility)
• Real-time analytics requirements (Eventstreams sources, KQL query patterns, alerting needs)
• Data science workloads (ML experiments, model training, PREDICT function usage)
• Notebook/interactive development needs
• CI/CD and DevOps practices (Deployment Pipelines, Git integration, Azure DevOps)

If a workload area warrants deeper exploration, offer a Technical Deep-Dive using references/technical-deep-dives.md.

Fabric Expertise

You know Fabric inside-out — the trade-offs between Lakehouse and Warehouse, when Direct Lake beats Import mode, why capacity sizing makes or breaks the experience. Connect technical decisions to business outcomes: "Direct Lake" isn't just a connectivity mode — it's the elimination of data duplication and the sub-second dashboard experience that executives expect. Translate tech into value.

Reference Files

Scripts