| name | metric-dashboard |
| description | Design metric dashboards and KPI tracking plans for products and features. Defines what to measure, how to measure it, alert thresholds, and dashboard layout. Covers product, business, and technical metrics. |
| argument-hint | [product, feature, or area to build metrics for] |
Metric Dashboard Skill
Design a comprehensive metric dashboard and KPI tracking plan for any product or feature.
When to Use
- User needs to define metrics for a new product or feature
- User is setting up monitoring and alerting
- User needs to design a dashboard layout
- User says
/metric-dashboard followed by the product/feature
- Any time measurement strategy needs to be defined
Framework: Metric Dashboard Design (5 Steps)
Step 1: Define the Metric Hierarchy
North Star Metric (NSM):
The single metric that best captures the value your product delivers.
- Must reflect user value, not just business value
- Must be measurable with current instrumentation
- Formula: NSM = [engagement unit] per [user segment] per [time period]
Decompose into a metric tree:
North Star Metric
├── Input Metric A (e.g., new users)
│ ├── Sub-metric A1
│ └── Sub-metric A2
├── Input Metric B (e.g., activation rate)
│ ├── Sub-metric B1
│ └── Sub-metric B2
└── Input Metric C (e.g., retention)
├── Sub-metric C1
└── Sub-metric C2
Step 2: Categorize Metrics
Product Metrics:
- Acquisition: How users find you (sign-ups, installs, registrations)
- Activation: First value moment (onboarding completion, first action)
- Engagement: Core usage (DAU/MAU, session length, feature adoption)
- Retention: Coming back (D1/D7/D30, cohort retention curves)
- Revenue: Monetization (ARPU, conversion, LTV, churn)
Technical Metrics:
- Performance: Latency (p50, p95, p99), throughput, error rate
- Reliability: Uptime, incident count, MTTR
- Infrastructure: CPU/memory utilization, cost per request
AI/ML Metrics (if applicable):
- Quality: Accuracy, hallucination rate, eval scores
- Safety: Content policy violation rate, false refusal rate
- Cost: Cost per inference, token usage
- Latency: Time to first token, tokens per second
Business Metrics:
- Revenue: MRR, ARR, revenue growth rate
- Unit economics: CAC, LTV, LTV/CAC ratio
- Market: Market share, competitive win rate
Step 3: Set Targets & Alerts
For each metric, define:
| Metric | Current | Target | Alert Threshold | Owner |
|---|
| NSM | X | Y | Z | PM |
| Metric A | | | | |
| Metric B | | | | |
Alert levels:
- Warning (yellow): Metric trending below target — investigate
- Critical (red): Metric below threshold — immediate action required
- Anomaly: Unexpected spike or drop — auto-detect and notify
Step 4: Design Dashboard Layout
Executive Dashboard (1 screen):
- NSM trend (last 30/90 days) — large, prominent
- 4-6 key metrics with sparklines and trend arrows
- Traffic light status (green/yellow/red) for each area
- Notable events annotated on the timeline
Operational Dashboard (detailed):
- Real-time metrics for the current day/hour
- Breakdowns by segment (platform, geography, user type)
- Funnel visualization (acquisition → activation → retention)
- Experiment results (A/B test outcomes)
Technical Dashboard (if applicable):
- System health (latency, error rate, uptime)
- Model performance (eval scores, cost, throughput)
- Infrastructure utilization and cost
Step 5: Measurement Plan
For each metric, document:
- Definition: Exact formula, including/excluding criteria
- Data source: Which event, table, or API
- Instrumentation: What needs to be logged/tracked
- Granularity: How often updated (real-time, hourly, daily)
- Segments: Key breakdowns (platform, country, user tier)
- Owner: Who monitors this metric
Output Format
Generate a complete metric plan in markdown with:
- Metric hierarchy (tree diagram)
- Metric definitions table
- Targets and alert thresholds
- Dashboard layout description
- Measurement plan
Common Pitfalls to Avoid
- Vanity metrics: Big numbers that don't reflect value (total sign-ups vs. active users)
- Too many metrics: 5-8 key metrics max on the exec dashboard
- No baselines: Always show current state before setting targets
- Missing guardrails: Every optimization metric needs a counter-metric
- No segmentation: Averages hide problems — always break down by segment
