business-intelligence

これはレビュー時に保存された抜粋です。完全で最新の内容は外部ソースを確認してください。

---
name: business-intelligence
description: >
  Business intelligence across dashboard design, visualization, and reporting
  automation. Use when designing dashboards, building KPI frameworks, automating
  reports, creating data stories, or optimizing BI tool performance.
license: MIT + Commons Clause
metadata:
  version: 1.0.0
  author: borghei
  category: data-analytics
  updated: 2026-03-31
  tags: [bi, dashboards, visualization, reporting, insights]
---
# Business Intelligence

The agent operates as a senior BI specialist, designing dashboards, defining KPI frameworks, automating reporting pipelines, and translating data into executive-ready narratives.

## Clarify First

Before designing the dashboard, confirm these inputs. If any is unknown or vague, ASK — do not assume:

- [ ] **Audience** — executive, operational, or self-service (sets the layout, altitude, and metric count per page)
- [ ] **Key questions + refresh cadence** — what decisions the dashboard drives and how fresh the data must be (scopes the metrics and the live-vs-extract choice)
- [ ] **KPI definitions** — formula, data source, owner, and RAG thresholds per metric (these are the exact fields the KPI template and `metric_validator.py` require)

Stop rule: ask only the 2-3 that most change the output. If the user says "just draft it," proceed and list your assumptions at the top of the artifact.

## Workflow

1. **Clarify the reporting need** -- Identify the audience (executive, operational, self-service), the key questions the dashboard must answer, and the refresh cadence. Validate that required data sources exist and are accessible.
2. **Define KPIs and metrics** -- For each metric, specify the formula, data source, granularity, owner, and RAG thresholds using the KPI definition template below.
3. **Design the dashboard layout** -- Apply the visual hierarchy (most important metric top-left, summary-to-detail flow top-to-bottom). Select chart types using the chart selection matrix. Limit to 5-8 visualizations per page.
4. **Build the semantic layer** -- Define metric calculations, hierarchies, and row-level security in the BI tool's semantic model so consumers get consistent numbers.
5. **Automate reporting** -- Configure scheduled delivery (PDF/email, Slack alerts) and threshold-based alerts with the patterns below.
6. **Validate and iterate** -- Confirm KPI values match source-of-truth queries. Check dashboard load time (<5 s target). Gather stakeholder feedback and refine.

## KPI Definition Template

```yaml
# Copy and fill for each metric
kpi:
  name: "Monthly Recurring Revenue"
  owner: "Finance"
  purpose: "Track subscription revenue health"
  formula: "SUM(subscription_amount) WHERE status = 'active'"
  data_source: "billing.subscriptions"
  granularity: "monthly"
  target: 1200000
  warning_threshold: 1080000   # 90% of target
  critical_threshold: 960000   # 80% of target
  dimensions: ["region", "plan_tier", "cohort_month"]
  caveats:
    - "Excludes one-time setup fees"
    - "Currency normalized to USD at month-end rate"
```

## Dashboard Design Principles

**Visual hierarchy:**
1. Most important metrics at top-left
2. Summary cards flow into trend charts flow into detail tables (top to bottom)
3. Related metrics grouped; white space separates logical sections
4. RAG status colors: Green `#28A745` | Yellow `#FFC107` | Red `#DC3545` | Gray `#6C757D`

**Chart selection matrix:**

| Data question | Chart type | Alternative |
|---------------|-----------|-------------|
| Trend over time | Line | Area |
| Part of whole | Donut / Treemap | Stacked bar |
| Comparison across categories | Bar / Column | Bullet |
| Distribution | Histogram | Box plot |
| Relationship | Scatter | Bubble |
| Geographic | Choropleth | Filled map |

## Executive Dashboard Example

```
+------------------------------------------------------------+
|                   EXECUTIVE SUMMARY                         |
| Revenue: $12.4M (+15% YoY)   Pipeline: $45.2M (+22% QoQ)  |