Power BI Desktop

View saved version →

---
name: powerbi-autonomous-exception-management
description: Use when turning autonomous Power BI planning deviations into managed exceptions, including revenue gaps, risky backlog, unrealistic targets, biased models, sparse segments, owner hints, status, escalation, and closure evidence.
---

# Power BI Autonomous Exception Management

Use this skill when every planning deviation should become a traceable exception rather than a dashboard comment.

## Exception types

- `budget_gap`
- `roll_gap`
- `backlog_at_risk`
- `delivery_risk`
- `model_biased`
- `sparse_segment`
- `target_infeasible`
- `human_override_required`
- `data_quality_blocker`

## Workflow

1. Detect exceptions from forecast, quality, scenario, constraint, and action outputs.
2. Deduplicate by month, customer, product, issue type, and root cause.
3. Assign severity from revenue impact, confidence, urgency, and controllability.
4. Suggest owner hints from customer hierarchy, sales group, product line, or region.
5. Track status and closure evidence.

## Required outputs

- `exception_id`
- `forecast_month`
- `exception_type`
- `customer`
- `product`
- `revenue_impact`
- `severity`
- `owner_hint`
- `status`
- `next_action`
- `evidence`

## Guardrails

- Every severe exception needs an action, owner hint, or explicit reason to defer.
- Do not close exceptions automatically without actuals, owner response, or rule-based evidence.

Referenced files: 1

View saved version →

---
name: powerbi-autonomous-forecast-agents
description: Use when coordinating multiple AI forecasting perspectives for Power BI sales forecasts, including backlog, seasonality, budget, sales skepticism, risk, consensus, dissent, and explainable forecast arbitration.
---

# Power BI Autonomous Forecast Agents

Use this skill when the user wants an agent council rather than one blended forecast. The goal is a structured debate that exposes why the forecast is high, low, risky, or trustworthy.

## Agent roles

- **Backlog Agent**: trusts open orders, delivery dates, status, order age, and learned conversion.
- **Demand Agent**: trusts seasonality, trend, intermittent demand, and customer/product history.
- **Budget Defense Agent**: explains the gap to budget and roll forecast.
- **Sales Skeptic Agent**: searches for optimism bias, volatile customers, sparse products, and weak evidence.
- **Risk Agent**: flags data quality, missing snapshots, memory fallback, biased backtests, delivery risk, and low-confidence matching.
- **Arbitrator Agent**: creates the final consensus, dissent, confidence, and recommended action.

## Workflow

1. Load the latest `ai-forecast-summary.csv`, `ai-forecast-detail.csv`, `ai-forecast-top-deltas.csv`, and `ai-forecast-model-quality.csv` when available.
2. Run each agent independently against the same month and grain.
3. Require every agent to provide:
   - forecast direction: up, down, or hold
   - evidence
   - confidence
   - top 3 risks
   - recommended action
4. The arbitrator produces:
   - consensus forecast
   - dissenting views
   - final risk flag
   - whether AI forecast can replace, challenge, or only annotate the roll forecast

## Output columns

- `forecast_month`
- `grain`
- `customer`
- `product`
- `agent_name`
- `agent_forecast`
- `agent_confidence`
- `evidence`
- `risk_flag`
- `recommended_action`
- `arbitrated_decision`

## Guardrails

- If model quality is worse than roll forecast, mark the result `advisory_only`.
- Do not hide dissent. Dissent is the value of the council.
- Use concrete customer/product/month evidence, not generic commentary.

Referenced files: 1

View saved version →

---
name: powerbi-autonomous-planning-loop
description: Use when building or running a closed-loop autonomous Power BI planning cycle that refreshes actuals, forecast, gap detection, scenarios, actions, tracking, learning, and the next plan without changing PBIX files directly.
---

# Power BI Autonomous Planning Loop

Use this skill when the user wants planning to run as a recurring control cycle rather than a one-time forecast.

## Loop

1. **Observe**: load actuals, open backlog, forecast outputs, model quality, overrides, and action status.
2. **Forecast**: run or reuse `Invoke-PowerBIAIForecast.ps1`.
3. **Detect**: identify gaps to budget, roll, prior year, and committed actions.
4. **Simulate**: create target, base, risk, and rescue scenarios.
5. **Decide**: classify items as trusted, challenge, rescue, de-risk, or advisory.
6. **Act**: create action records with owner hints, probability, impact, and due date.
7. **Learn**: compare outcomes to forecast and update trust, bias, and readiness signals.
8. **Repeat**: preserve a versioned run record for the next cycle.

## Required artifacts

- `planning-run-log.csv`
- `planning-gap-detection.csv`
- `planning-decisions.csv`
- `planning-learning-events.csv`

## Guardrails

- Keep every run immutable and timestamped.
- Never overwrite management numbers; store autonomous recommendations separately.
- Stop autonomous action recommendations when readiness is below threshold.

Referenced files: 1

View saved version →

---
name: powerbi-causal-counterfactual-forecasting
description: Use when adding causal and counterfactual thinking to Power BI sales forecasts, including working days, holidays, delivery constraints, price changes, product lifecycle, stockouts, campaigns, customer behavior, and best/base/worst case simulations.
---

# Power BI Causal Counterfactual Forecasting

Use this skill when the user asks why a forecast changes, what causes a revenue gap, or what would happen under alternative assumptions.

## Feature families

- calendar: working days, holidays, month length, fiscal periods
- order flow: order age, requested delivery, planned delivery, status, backlog value
- customer behavior: recency, frequency, average order value, churn or reactivation signals
- product lifecycle: new product, mature product, discontinued product, replacement product
- operations: supply constraints, delivery delay, stockout indicators
- commercial: price change, discounting, campaign, sales initiative, budget/roll assumptions

## Workflow

1. Separate correlation from actionable cause. Do not claim causality without a plausible mechanism and supporting time sequence.
2. Build counterfactuals:
   - if backlog conversion improves
   - if delivery slips
   - if customer demand follows prior year
   - if budget pressure is ignored
   - if low-confidence segments are excluded
3. Quantify sensitivity:
   - revenue impact
   - probability
   - confidence
   - affected customer/product/month
4. Explain the causal story in one sentence per material driver.

## Required outputs

- `forecast_month`
- `driver`
- `driver_type`
- `base_value`
- `counterfactual_value`
- `revenue_impact`
- `confidence`
- `evidence`
- `actionability`

## Guardrails

- Mark drivers as `hypothesis` when the data only supports association.
- Avoid overfitting small customer/product segments.
- Use backtests to prove that adding a driver improves WAPE or bias before making it a default weight.

Referenced files: 1

View saved version →

---
name: powerbi-constraint-aware-planning
description: Use when constraining Power BI revenue planning by delivery, capacity, stock, supply, margin, cash, payment terms, sales resources, or operational feasibility rather than forecasting unconstrained demand only.
---

# Power BI Constraint-Aware Planning

Use this skill when autonomous planning must respect operational limits.

## Constraint classes

- delivery: planned date, requested date, delay, shipment risk
- inventory: stockout, substitute availability, item lifecycle
- capacity: production, picking, shipping, service bottlenecks
- commercial: margin, discount, price change, payment terms
- customer: block status, churn risk, buying cadence, credit risk
- sales: owner capacity, account priority, campaign timing

## Workflow

1. Start from an unconstrained forecast or target scenario.
2. Apply constraints as filters or down-weighting factors.
3. Produce both unconstrained and constrained values.
4. Explain every material reduction.
5. Mark missing constraints as `unknown`, not as zero risk.

## Required outputs

- `forecast_month`
- `customer`
- `product`
- `unconstrained_forecast`
- `constrained_forecast`
- `constraint_type`
- `constraint_severity`
- `revenue_blocked`
- `evidence`
- `recommended_resolution`

## Guardrails

- Never assume demand can convert when delivery or credit constraints contradict it.
- Keep constraint assumptions auditable and separate from statistical demand.

Referenced files: 1

View saved version →

---
name: powerbi-desktop
description: "Use when building Power BI models, trusting KPIs, changing PBIP safely, releasing, operating Fabric, or governing decisions."
---

# Power BI Index

This is the routing skill for the Power BI plugin. Do not enumerate scripts or load specialist workflows until the user's outcome is clear.

For "what can you do?", "help me get started", or broad orientation, read [orientation response](references/orientation-response.md) and return its user-facing content.

## Product Paths

| User outcome | Focused workflow | Result |
| --- | --- | --- |
| Build a model from data sources | `powerbi-model-wizard` | source-to-PBIP design pack |
| Understand or trust KPIs | local review plus executive brief | definitions, caveats, decision evidence |
| Change a PBIP/TMDL model | unified review | impact, tests, safe drafts |
| Release with confidence | release candidate pack | Go/Warn/No-Go and rollback evidence |
| Operate Fabric or a portfolio | enterprise operations pack | capacity, SLO, Direct Lake, governance evidence |
| Govern a decision or action | decision intelligence pack | Copilot quality, decision memory, owner actions |

Read the exact focused skill in full when one exists. Otherwise run the named workflow. Use only the minimum evidence needed, then offer a deeper path.

## Safety

- Never modify PBIX/PBIT binaries.
- Do not publish, sign in, refresh credentials, or change Power BI/Fabric content without explicit authorization.
- Label evidence as local, live, snapshot, heuristic, or draft.

Referenced files: 4

View saved version →

---
name: powerbi-fabric-nextgen-usps
description: "Use when a Power BI or Fabric team needs FinOps, Copilot answer regression, Direct Lake readiness, data-product SLOs, capacity change proof, or executive decision traces."
---

# Power BI And Fabric Next-Generation USPs

Run the local pack first:

```powershell
.\plugins\powerbi-desktop\scripts\Invoke-PowerBINextGenUspPack.ps1 -Path .\your-model -OutputDirectory .\powerbi-nextgen-usp-pack
```

It produces six machine-readable artifacts: Fabric FinOps Copilot, Copilot Answer Regression Lab, Direct Lake/OneLake Readiness, Data Product SLO Manager, Capacity Change Verifier, and Executive Decision Trace.

- Treat FinOps, Direct Lake, and decision outputs as decision support unless live/snapshot evidence is attached.
- The Copilot lab creates test cases; it does not claim a Copilot answer was validated until an approved answer capture is supplied.
- Capacity savings require before/after Capacity Metrics or CU evidence.
- Never access Fabric with a token or change service state unless the user explicitly requests it.

Referenced files: 1

View saved version →

---
name: powerbi-forecast-trust-market
description: Use when designing or evaluating a Power BI forecast trust market, human override tracking, sales and finance confidence scoring, forecast accountability, and self-learning model-vs-human accuracy loops.
---

# Power BI Forecast Trust Market

Use this skill when the user wants the forecast to learn from human judgment, overrides, and organizational accuracy over time.

## Concept

Every forecast can receive confidence signals from AI, Sales, Finance, Supply Chain, and Management. Later actuals determine which source was calibrated and which source was biased.

## Workflow

1. For each forecast version, capture:
   - AI forecast
   - roll forecast
   - budget
   - human override
   - confidence percentage
   - reason code
   - owner or role
   - timestamp
2. After actuals arrive, calculate:
   - absolute error
   - bias
   - directional accuracy
   - calibration error
   - trust score by role, segment, customer, product, and horizon
3. Feed the trust score back into the next forecast cycle:
   - high-trust human input can challenge model output
   - low-trust input remains visible but gets lower weight
   - biased sources trigger review rather than automatic blending

## Suggested tables

- `AI_Forecast_Version`
- `AI_Forecast_Override`
- `AI_Forecast_Actuals`
- `AI_Forecast_TrustScore`

## Required metrics

- WAPE
- bias
- hit rate within tolerance
- override lift versus AI baseline
- override lift versus roll forecast
- confidence calibration

## Guardrails

- Do not turn trust scores into personal blame. Frame them as calibration by source and context.
- Preserve the original AI forecast and human override separately.
- Require reason codes for overrides that materially change the forecast.

Referenced files: 1

View saved version →

---
name: powerbi-forecast-war-room
description: Use when building a Power BI forecast war room or control tower for executive revenue review, forecast risk triage, top deltas, action ownership, gap closure status, confidence monitoring, and daily forecast operating rhythm.
---

# Power BI Forecast War Room

Use this skill when the user wants a management control tower around the forecast rather than only CSV outputs.

## Operating rhythm

1. Daily or weekly refresh of forecast outputs.
2. Review month-level gap to budget, roll, and prior year.
3. Open top deltas as action items.
4. Assign owner hints by customer hierarchy, product line, sales group, or region when available.
5. Track status:
   - open
   - validated
   - escalated
   - rescued
   - lost
   - ignored due to low data confidence
6. Review model quality and override trust after actuals land.

## War room views

- executive gap summary
- top risk customers/products
- backlog at risk
- rescue action board
- agent council dissent
- trust score by source
- model quality by horizon
- counterfactual scenarios

## Required outputs

- `forecast_month`
- `issue_type`
- `customer`
- `product`
- `revenue_at_risk`
- `recommended_action`
- `owner_hint`
- `status`
- `due_date`
- `confidence`
- `evidence`

## Guardrails

- Every red item needs an action or an explicit reason to ignore.
- Do not mix unvalidated AI recommendations with committed management numbers.
- Keep action status separate from model forecast values so reporting remains auditable.

Referenced files: 1

View saved version →

---
name: powerbi-goal-seeking-planning
description: Use when calculating what must happen to reach Power BI revenue, budget, roll, margin, cash, or service targets, including reverse planning, target-gap decomposition, required backlog conversion, and required demand uplift.
---

# Power BI Goal-Seeking Planning

Use this skill when the question is "What must happen to hit the target?" rather than "What will happen?"

## Workflow

1. Select target: budget, roll forecast, prior year, management target, margin, or cash.
2. Calculate current plan gap by month.
3. Decompose the required change:
   - actual-to-date contribution
   - expected backlog conversion
   - residual demand required
   - customer/product gap drivers
4. Rank feasible levers:
   - improve conversion
   - pull forward backlog
   - recover customer demand
   - substitute product/customer demand
   - challenge the target if infeasible
5. Label target feasibility:
   - `achievable`
   - `stretch`
   - `unlikely`
   - `not_supported_by_data`

## Required outputs

- `forecast_month`
- `target_name`
- `target_value`
- `current_forecast`
- `target_gap`
- `required_backlog_conversion`
- `required_residual_demand`
- `top_required_segments`
- `feasibility`
- `explanation`

## Guardrails

- Do not treat target closure as evidence that the target is realistic.
- Separate mathematically required revenue from operationally feasible revenue.

Referenced files: 1

View saved version →

---
name: powerbi-model-wizard
description: "Use when starting a new Power BI model from a business goal, source list, star schema, KPI plan, security plan, and report-page plan."
---

# Power BI Model Wizard

Use `New-PowerBIModelWizard.ps1` to create a local design pack for a new Power BI model from any Power BI Desktop connector. It directly profiles local CSV/JSON files; for every other connector, use `templates/powerbi-data-sources.example.json` with the exact connector name shown in Power BI Desktop. It proposes facts, dimensions, KPI drafts, source drafts, and report pages. It does not create a PBIX/PBIT binary, connect credentials, or publish.

1. Gather the business decision and either local CSV/JSON source paths or a connector declaration file. Ask only for domain rules that cannot be inferred, such as KPI ownership or RLS roles.
2. Generate the draft directory with `-DataSourcePaths <paths> -Initialize` or `-DataSourceConfigPath <config> -Initialize`.
3. Review data-contract and measure drafts with the user.
4. Have the user create and save the target PBIP in Power BI Desktop.
5. Generate and explicitly apply reviewed PBIP/TMDL drafts, then validate with `Invoke-PowerBIUnifiedReview.ps1`.

State clearly that the generated content is a draft until it has been reviewed and validated in Power BI Desktop.

View saved version →

---
name: powerbi-planning-memory
description: Use when storing and reusing Power BI planning memory across cycles, including forecast versions, assumptions, actions, overrides, actuals, error causes, segment lessons, and durable learning signals for autonomous planning.
---

# Power BI Planning Memory

Use this skill when the planning engine needs memory across forecast cycles.

## Memory records

- forecast run
- scenario assumption
- generated action
- human override
- owner response
- actual outcome
- model quality event
- data quality issue
- segment lesson

## Workflow

1. Preserve every planning run with timestamp, as-of date, horizon, grain, and input paths.
2. Store assumptions separately from outputs.
3. Link actions and overrides to later actuals.
4. Generate lessons:
   - segment is consistently optimistic
   - customer converts late
   - product line is supply constrained
   - roll forecast is better than AI
   - human override improved the result
5. Feed lessons into trust, readiness, and scenario generation.

## Required outputs

- `memory_id`
- `event_type`
- `as_of_date`
- `forecast_month`
- `grain`
- `entity_key`
- `signal`
- `impact`
- `lesson`
- `source_artifact`

## Guardrails

- Keep planning memory local-first unless the user explicitly provides an approved storage path or governance workflow.
- Keep memory append-only unless the user explicitly asks to clean it.
- Store personal or owner-level data only when it is necessary and governance allows it.
- Do not write secrets, tokens, credentials, or private contact details into planning memory.

Referenced files: 1

View saved version →

---
name: powerbi-planning-readiness-score
description: Use when scoring whether a Power BI model and forecast pipeline are ready for autonomous planning, including data quality, backlog snapshots, order-to-invoice matching, model accuracy, bias, granularity, constraints, overrides, and action tracking.
---

# Power BI Planning Readiness Score

Use this skill before allowing autonomous planning to drive decisions.

## Score dimensions

- data completeness
- order-to-invoice matching
- backlog snapshot availability
- forecast backtest quality
- bias control
- grain coverage
- constraint coverage
- human override tracking
- action tracking
- planning memory

## Workflow

1. Inspect available model fields, forecast CSVs, quality outputs, and memory/action files.
2. Score each dimension from 0 to 100.
3. Assign overall readiness:
   - `manual_only`: 0-39
   - `assisted`: 40-59
   - `controlled_autonomy`: 60-79
   - `autonomous_ready`: 80-100
4. List blockers and the next highest-leverage improvement.

## Required outputs

- `dimension`
- `score`
- `status`
- `evidence`
- `blocker`
- `recommended_next_step`

## Guardrails

- Do not mark autonomous-ready without historical backtests and snapshot coverage.
- Penalize missing constraints even when the revenue forecast looks accurate.

Referenced files: 1

View saved version →

---
name: powerbi-revenue-digital-twin
description: Use when designing or running a Power BI revenue digital twin for sales forecasting, target-gap simulation, backlog-to-invoice scenarios, what-if revenue paths, budget attainment, and forecast-to-cash decision support from an open Power BI Desktop model or exported forecast CSVs.
---

# Power BI Revenue Digital Twin

Use this skill when the user wants to move beyond a static sales forecast into scenario simulation: "How do we still hit budget?", "What if delivery slips?", "Which backlog must convert?", or "Which customers/products close the revenue gap?"

## Workflow

1. Start read-only. Use the open Desktop model through `Invoke-PowerBIAIForecast.ps1` or existing forecast CSVs. Do not write to PBIX/PBIP unless explicitly requested.
2. Establish the current target gap by month: AI forecast, roll forecast, budget, actual-to-date, open backlog, expected backlog revenue.
3. Build scenario levers:
   - backlog acceleration or delay
   - conversion probability changes
   - customer/product demand uplift or erosion
   - working-day and holiday impact
   - budget or roll target constraint
   - supply or delivery risk
4. Produce at least three scenarios:
   - base case from the current AI forecast
   - target case showing what must change to hit budget or roll
   - risk case showing likely downside if weak backlog or volatile segments slip
5. Rank the smallest set of customer/product/month changes that explain or close the gap.

## Required outputs

- `forecast_month`
- `target_metric` such as Budget or Roll
- `current_ai_forecast`
- `target_gap`
- `required_backlog_conversion`
- `required_residual_demand`
- `top_gap_drivers`
- `scenario_name`
- `scenario_forecast`
- `scenario_probability`
- `explanation`

## Quality gates

- Mark a scenario `not_actionable` when it depends on segments already flagged `advisory_only`, `biased`, or `sparse` without human validation.
- Separate controllable levers from non-controllable statistical demand.
- Never present a single scenario as truth; show assumptions and sensitivity.

Referenced files: 1

View saved version →

---
name: powerbi-revenue-rescue-mode
description: Use when turning Power BI sales forecast gaps into operational rescue actions, including backlog acceleration, customer activation, product substitution, delivery escalation, gap closure ranking, and forecast-to-action recommendations.
---

# Power BI Revenue Rescue Mode

Use this skill when the user asks how to close a forecast gap, secure the month, rescue revenue, or convert forecast findings into actions.

## Workflow

1. Identify the gap by month:
   - budget gap
   - roll gap
   - prior-year gap
   - AI forecast confidence
2. Split the gap into action classes:
   - **accelerate**: open backlog likely billable if delivery or admin blockers are solved
   - **validate**: material customer/product deltas where human sales input is needed
   - **recover**: segments below seasonal or roll expectation with enough history
   - **substitute**: alternative products or customers can offset risk
   - **de-risk**: low-confidence, biased, or sparse segments should not carry the plan
3. Rank actions by expected impact, probability, owner, and deadline.
4. Produce a rescue board that can be imported into Power BI or used as a sales operations action list.

## Action scoring

Use a simple score unless a better calibrated model exists:

`impact_score = expected_revenue * confidence_weight * urgency_weight`

Confidence weights:

- high: 1.00
- medium: 0.70
- low: 0.35

Urgency weights:

- current month: 1.25
- next month: 1.00
- month +2: 0.80

## Required outputs

- `forecast_month`
- `action_type`
- `customer`
- `product`
- `expected_revenue_impact`
- `probability`
- `deadline`
- `owner_hint`
- `reason`
- `next_action`
- `risk_flag`

## Guardrails

- Never recommend actions on segments with no evidence unless marked as `exploratory`.
- Distinguish revenue that can be pulled forward from revenue that requires new demand.
- Flag actions that may improve revenue but harm margin, cash, or delivery reliability.

Referenced files: 1

View saved version →

---
name: powerbi-self-healing-forecast-governance
description: Use when Power BI forecast governance should automatically demote biased models, choose safer baselines, trigger data-quality blockers, adjust trust status, and protect planning from low-quality autonomous recommendations.
---

# Power BI Self-Healing Forecast Governance

Use this skill when the planning engine must know when not to trust itself.

## Trust states

- `trusted`: model can drive planning.
- `challenge`: model can challenge roll or budget but needs review.
- `advisory_only`: model is informative but should not replace baseline.
- `blocked`: data or quality issue prevents autonomous recommendation.
- `needs_snapshot_data`: historical as-of state is missing.
- `biased`: backtest bias exceeds tolerance.

## Workflow

1. Read quality metrics by horizon and segment.
2. Compare AI WAPE and bias against roll forecast and accepted tolerances.
3. Assign trust state by horizon, segment, customer, and product where possible.
4. Route planning:
   - trusted: use AI as primary input
   - challenge: show AI and baseline side by side
   - advisory_only: annotate only
   - blocked: require data fix or human review
5. Emit governance events so later runs can learn from demotions.

## Required outputs

- `forecast_month`
- `horizon_months`
- `segment`
- `ai_wape`
- `baseline_wape`
- `bias`
- `trust_state`
- `governance_action`
- `reason`

## Guardrails

- Do not improve apparent accuracy by hiding hard segments.
- Prefer a safer baseline when AI is worse than roll forecast.

Referenced files: 1