| name | analyze-account-health |
| description | Summarizes B2B account health by analyzing usage patterns, engagement trends, risk signals, and expansion opportunities. Use for customer success reviews, renewal preparation, QBRs, or account prioritization. |
Analyze Account Health
Deep-dive into a B2B account's product usage to prepare for QBRs, assess renewal risk, identify expansion opportunities, or prioritize CS outreach.
Instructions
Step 0: Identify Account & Discover Context
Get the account identifier:
- Company name, org ID, account ID, or group property value
- Ask user if not provided
Search for existing work:
Use Amplitude:search_amp_entities to find existing dashboards, charts, or notebooks for this account. If found, ask user if they want fresh analysis or to review existing.
Resolve how accounts are modelled — do this before any query.
Every step below breaks down "by account", and there are three different ways a
project represents one. Check in this order and reuse the answer throughout:
get_group_types — if the project has a group type (commonly org id or
company), that is the account. Account attributes are group
properties: get_properties({propertyType: 'group', groupType: '<type>'}),
referenced with scope: 'group' plus group_type. To count accounts rather
than users, set count_unique_by to the group type.- If there is no group type, the account is usually a user property
(
company, org_name) — scope: 'user'.
- Failing both, an event property carrying the org (
org id, org url).
If a group property comes back as Invalid group property … for group type …,
the query engine's registry is missing it — that is a platform gap, not a
naming mistake. Do not retry spelling variants. Fall back to the user- or
event-level equivalent from (2)/(3), and tell the user which representation you
used, since the numbers are not interchangeable.
Step 1: Quick Health Triage
Use Amplitude:query_amplitude_data to run these queries in parallel:
Usage Trend:
- Event:
_active, Metric: uniques, Group by: the account property resolved in Step 0 (with its scope)
- Time: Last 60 days, daily interval
- Shows: Activity increasing or decreasing?
Engagement Quality:
- Calculate DAU and MAU for account
- Get DAU/MAU ratio (stickiness)
- Shows: How engaged are active users?
User Momentum:
- Active user count week-over-week
- Shows: Team growing or shrinking?
Classify Health:
- Healthy: Growing MAU, DAU/MAU >40%, positive WoW
- At-Risk: Flat/declining MAU, DAU/MAU 20-40%, negative WoW
- Critical: Steep decline, DAU/MAU <20%, sustained negative WoW
Step 2: User-Level Analysis
Use Amplitude:query_amplitude_data with user-level groupBy:
Power Users:
- Top 3-5 users by event volume (champions to leverage)
Churned Users:
- Users active in previous period but not current (retention risks)
License Utilization:
- Active users in last 30 days vs total seats
Step 3: Feature Usage Analysis
Use Amplitude:query_amplitude_data grouped by events/features:
Feature Breadth:
- Which core features are being used (ask user for 5-10 key features)
- Adoption rate per feature
Feature Trends:
- Usage over last 90 days per feature
- Identify growing vs declining features
Focus based on health:
- If At-Risk/Critical: Find abandoned features (used 60-90 days ago, not in last 30)
- If Healthy: Find expansion opportunities (premium features not yet tried)
Step 4: Account Feedback Analysis
Get feedback sources:
Use Amplitude:use_amplitude_ai_feedback with facet: "sources" to see what's available.
Get feedback insights:
Use Amplitude:use_amplitude_ai_feedback with facet: "insights" filtered by:
- ampId for each user in the account
- dateStart/dateEnd: Last 90 days
- types:
bug, painPoint, complaint, request, lovedFeature
Get specific mentions:
For top 3-5 insights, use Amplitude:use_amplitude_ai_feedback with facet: "mentions" to get quotes.
Correlate with behavior:
- Complaint about Feature X? Query their usage of Feature X
- Request for Feature Y? Check if they hit limits Y would solve
- Praise for Feature Z? Validate they're heavy users of Z
Step 5: Present Account Health Report
Structure output as follows:
Account Health Report: [Account Name]
Executive Summary
[2-3 sentences: Health score, key trend, primary recommendation]
Health Score: [🟢 Healthy | 🟡 At-Risk | 🔴 Critical]
[One sentence rationale with key metric]
Key Metrics
| Metric | Current | Trend | Status |
|---|
| MAU | X | ↑↓→ Y% | 🟢🟡🔴 |
| DAU/MAU | X% | ↑↓→ Y% | 🟢🟡🔴 |
| License Utilization | X% | ↑↓→ | 🟢🟡🔴 |
| Features Adopted | X/Y | ↑↓→ | 🟢🟡🔴 |
🚨 Risk Factors (if any)
- [Issue] - [Impact]
- Usage data: [metric/trend]
- Customer feedback: [theme with X mentions] - [representative quote]
✅ Positive Signals
- [What's working] - [Evidence from usage + feedback]
👥 User Intelligence
Champions (Leverage)
- [User ID/Name]: [Activity summary] - Action: [Specific CS recommendation]
At Risk (Engage)
- [User ID/Name]: [Last active date / declining pattern] - Action: [Check-in recommendation]
Inactive (>30 days)
- [Count] users ([X]% of licenses)
💡 Top Pain Points & Requests
Pain Points
- [Theme] (X mentions)
- [Concise description]
- Evidence: [Behavioral data] + "[Quote]" - [Source, Date]
- Action: [What to do]
Feature Requests
- [Theme] (X mentions)
- [What they want]
- Evidence: "[Quote]" - [Source, Date]
- Roadmap status: [On roadmap/Not planned/Considering]
What They Love ❤️
- [Feature]: "[Quote]"
📊 Feature Adoption
High Usage: [Feature] - [X users] (↑Y%)
Declining: [Feature] - [X users] (↓Y%) - Investigate
Untapped (Upsell): [Premium feature] - Could solve [pain point]
🎯 Recommendations
🔥 This Week
- [Specific action with user/contact name]
📅 This Month
- [Strategic action with context]
💰 Expansion Opportunities
- [Upsell signal with evidence]
📎 Details
- Analysis Date: [Date]
- Timeframe: [Last X days]
- Confidence: [High/Medium/Low based on data volume]
Best Practices
- Always name users - CS needs who to contact, not aggregates
- Connect feedback to behavior - Validate complaints with usage data
- Be specific in recommendations - "Call Sarah about Feature X" not "improve engagement"
- Show trends, not snapshots - Direction matters more than point-in-time
- Flag data gaps - Note low volume, missing properties, or incomplete data
- Prioritize by impact - Focus on issues affecting multiple users or champions
Common Patterns
Churn Risks:
- Champion churned + declining overall usage
- Multiple complaints about same issue + behavioral evidence of friction
- License utilization declining + negative feedback
Expansion Signals:
- Hitting plan limits (users, API, storage)
- Requests for premium features + high engagement
- New users being added + positive feedback
