Home / Coach / Coach Guide — Challenge 6: Make It Operational
Coach

Coach Guide — Challenge 6: Make It Operational

Attendee challenge: challenges/challenge-06-operationalize.md

Snapshot

Est. time 1.5–2 h
Difficulty ⭐⭐⭐⭐ (400) — stretch / finale
They build Operational alerts, chargeback/showback, RLS, and scheduled refresh
Key services Data Activator, Power BI RLS, Fabric pipelines/scheduling, FinOps

Coaching objectives

This is where reliability, cost, and performance become actionable. The team already has a Control Tower; now they prove the customer can operate it on Monday morning.

Treat this as a menu. Teams only need two success criteria, but push them toward the options that best fit their skills and showcase story.

What good looks like: the team demonstrates two live operational controls — an alert fires, a chargeback matrix attributes cost, RLS filters two roles differently, or the daily pipeline is scheduled with failure notification.

The reference path

A. Real-time alerts with Activator

Goal: create a reflex on a Gold/semantic condition and notify an operator.

  1. Pick a KPI already modeled in Gold or the semantic model: ErrorRate, P95Latency, retry rate, or daily cost variance.
  2. Confirm the data is available through a source Activator supports, such as Power BI, Eventstream, or KQL.
  3. Define the condition and threshold:
    • error rate > 2%
    • P95 latency > 15,000 ms
    • retry rate > 3%
    • daily cost variance > 20%
  4. Configure notification to Teams or email.
  5. Simulate a breach by lowering the threshold or filtering to a noisy agent; preserve the production threshold in the narrative.

Coach prompt: "If this alert fired at 06:15, who would receive it and what would they do next?"

B. FinOps chargeback/showback

Goal: attribute blended agent cost to accountable owners.

Reference modeling approach:

  1. Start with gold_cost_summary for billed/resource costs.
  2. Join to dim_resource or resource metadata for tags: team, costCenter, useCase, client, and environment.
  3. Join agent metrics or gold_agent_analytics by agent namespace where model-token cost is agent-level.
  4. Parse namespace levels from <org>.<domain>.<agent>.<version> and use them as rollup columns.
  5. Build measures for total cost, run rate, and MoM change.
  6. Put missing tags/namespaces into an explicit Unattributed bucket.

Watch for double-counting. If a model-token estimate is already included in a blended cost measure, do not add the same value again.

C. Power BI Row-Level Security

Goal: demonstrate governed views by owner/team.

Reference path:

  1. Create or identify an ownership dimension with role/team to subscription, resource group, or agent namespace mappings.
  2. In the semantic model, define two roles such as TeamA and TeamB.
  3. Add a DAX filter on the ownership/resource/agent dimension. Examples:
    • dim_resource[Team] = "FinanceOps"
    • LEFT(dim_agent[AgentNamespace], 16) = "contoso.finance"
  4. Test with View as for both roles.
  5. Assign users or groups if the tenant/workspace permits; for RVAS validation, View as is enough.

Coach prompt: "Show me the exact same report page as Role A and Role B, and tell me why the totals changed."

D. Daily pipeline automation

Goal: schedule the Control Tower operating rhythm.

Reference path:

  1. Open the Daily Refresh Pipeline from the Fabric workspace.
  2. Schedule it daily at about 06:00 UTC.
  3. Confirm it processes the current month incrementally.
  4. Verify the dependency order: Bronze → Silver → Gold → semantic-model refresh.
  5. Configure failure notification through Fabric alerts, email, or Teams webhook.
  6. Bonus: add a capacity-utilization forecast from gold_capacity_usage or a threshold visual.

The reference resources/fabric-control-tower/README.md describes the Daily Refresh schedule, retry policy, and failure notifications.

Checkpoint verification

Ask the team to demo at least two:

  1. Alert: trigger the simulated breach and show the Teams/email notification or alert history.
  2. Chargeback: open a matrix by team/use case with total cost, run rate, and MoM trend; point to the unattributed bucket if tags are missing.
  3. RLS: use View as for two roles and verify totals/details differ appropriately.
  4. Automation: open the Daily Refresh schedule and failure-notification configuration; show the last run or a test run if available.

✅ Pass when two operational features are working live and the team can explain how each supports the three business questions.

Common pitfalls & fixes

Pitfall Fix
Activator cannot see the intended data Use a supported source such as Power BI, Eventstream, or KQL; if needed, alert on an existing semantic measure instead of a raw table
Alert never fires during the demo Lower the threshold temporarily or filter to a known noisy agent; state the real production threshold
RLS roles defined but not assigned or tested Use View as at minimum; if time allows, assign Entra users/groups to roles
RLS filter is on a disconnected table Ensure the filtered dimension has relationships to the fact tables driving visuals
Chargeback double-counts model cost Decide whether token estimates are additive or already blended into cost; document the measure logic
Tags missing on resources Route unknown owners into Unattributed; make tag coverage a governance KPI
Pipeline schedule uses local time by mistake Confirm UTC vs local timezone; the reference rhythm is ~06:00 UTC
Trial capacity feature limits or tenant settings block a path Validate what the tenant allows; accept a documented demo using supported features in that environment

Talking points (mini-briefing)

  • Close the loop: detect → alert → act. Dashboards explain; operations require a trigger and an owner.
  • Showback drives behavior. When teams see their run rate and MoM change, they optimize prompts, model choice, cache hit rate, and resource sizing.
  • Governance is part of observability. RLS turns one enterprise Control Tower into safe views for many teams.
  • Production-ready means scheduled, owned, and monitored. A report that only updates manually is a demo; a scheduled pipeline with failure alerting is an operating system.
  • Unattributed cost is a signal. Missing tags are not a reason to hide cost — they are a backlog for platform governance.

If they go further

  • Add anomaly detection over error rate, token volume, or daily cost variance.
  • Wire the alert to a Teams workflow or ticketing action with owner routing.
  • Add write-back for incident notes, suppression windows, or chargeback approvals.
  • Split multiple business units into separate workspaces with shared semantic-model patterns.
  • Add a capacity-utilization forecast and alert before the Fabric capacity saturates.
  • Build a governance scorecard: tag coverage, RLS coverage, pipeline freshness, and alert ownership.

Reference assets

← Back to all content View source on GitHub ↗