kienbui1995/analyze-dashboard

Analyze Amplitude Dashboard

When to Use

• A stakeholder asks for a summary of what a dashboard is showing
• You need to prepare talking points for a weekly or monthly business review
• An executive wants a written briefing instead of raw charts
• You need to synthesize multiple metrics into a single coherent story
• Identifying the most important signal across a dashboard with many charts

Core Jobs

1. Load All Charts

Call mcp__Amplitude__get_dashboard to retrieve the dashboard configuration and chart list. Then use mcp__Amplitude__query_charts to fetch current data for all charts. Do not analyze charts individually in isolation — the goal is synthesis across the entire dashboard.

Note: query all charts before drawing conclusions. A metric that looks alarming in isolation may be explained by another chart on the same dashboard.

2. Identify the Story Across Metrics

Look for the narrative thread connecting the charts:

• What is the overall health signal? (Positive momentum, stable, declining, or mixed)
• Are the leading indicators consistent with the lagging indicators?
• If the North Star metric is up, are the supporting metrics aligned?
• If the North Star is flat or down, which upstream metrics explain it?
• Are there contradictions across charts that need investigation?

The story is not a list of what each chart shows. It is the meaning that emerges when the charts are read together.

3. Surface Top 3-5 Findings Ranked by Business Impact

Select the most important findings across the entire dashboard. Rank by business impact:

• Impact tier 1: Changes that affect revenue, retention, or growth rate directly
• Impact tier 2: Changes that affect leading indicators of revenue or retention (activation rate, feature adoption, engagement depth)
• Impact tier 3: Changes in operational metrics that may explain tier 1 or 2 findings

For each finding, provide: the specific metric, the direction and magnitude of change, the time window, and why it matters to the business.

4. Highlight Anomalies and Explain Context

Flag any metric that deviates unexpectedly from its trend. For each anomaly:

• Describe what is unusual (not just "it dropped" — by how much, compared to what baseline)
• Offer the most likely explanation based on what other charts show
• Flag whether the anomaly is confirmed or requires further investigation
• Note if it could be a tracking artifact (incomplete day, instrumentation break)

5. Correlate with User Feedback if Available

If the dashboard includes qualitative signals (NPS, CSAT, support ticket volume), correlate them with the quantitative metrics:

• Are users reporting friction in areas where quantitative metrics also show drop-off?
• Are there positive qualitative signals that explain quantitative growth?
• Contradictions between qualitative and quantitative signals often reveal the most interesting insights

6. Provide Specific Recommendations

Conclude with 2-4 concrete recommended actions. Each recommendation should:

• Be directly tied to a specific finding from the dashboard
• Name the team or person who should act on it
• Specify a timeline (this sprint, this week, this month)
• Have a clear success metric — how will you know if the recommendation worked?

MCP Tools

• mcp__Amplitude__get_dashboard — load dashboard structure and chart list
• mcp__Amplitude__query_charts — fetch current data for all charts in the dashboard
• mcp__Amplitude__render_chart — visual rendering of individual charts for pattern recognition
• mcp__Amplitude__get_feedback_insights — correlate quantitative signals with user feedback (if available)

Key Concepts

• Narrative synthesis: Translating a collection of charts into a single coherent story with a beginning (context), middle (findings), and end (recommendations).
• Leading indicator: A metric that changes before the outcome metric (e.g., activation rate predicts retention).
• Lagging indicator: A metric that reflects past performance (e.g., monthly revenue reflects decisions made weeks ago).
• Business impact ranking: Prioritizing findings by their consequence for revenue, retention, or growth — not by the size of the percentage change.
• Anomaly context: An anomaly is only meaningful when compared to a baseline. Always state both the deviation and the baseline.
• Contradictions: When two metrics tell opposite stories, that contradiction is often the most important finding.

Output Format

Output is written in narrative paragraphs readable by executives — not labeled database fields or bullet dumps.

Structure:

1. Opening (1-2 sentences): The single most important thing the dashboard shows right now.
2. Overall health (1 paragraph): The big-picture read on the product area — momentum direction, key drivers.
3. Key findings (3-5 findings, each 2-3 sentences): Specific metrics with numbers, direction, magnitude, and business significance.
4. Risks and anomalies (1 paragraph): What to watch — deviations that need attention or investigation.
5. What's working (1 paragraph): Positive signals worth amplifying or doubling down on.
6. Recommendations (2-4 bullet points): Concrete next actions, each tied to a specific finding.

Every factual claim must include a specific number. No vague language ("metrics look good," "some improvement").