Score Churn Risk from Support Ticket Patterns

Analyze an account's support ticket history and produce a churn risk score (0-100) with specific signals and recommended intervention.

Prerequisites

• Normalized ticket data from intercom-conversations-export fundamental
• Tags applied via intercom-ticket-tagging fundamental
• Account-level aggregation (all tickets for a given company/contact)

Input: Account Ticket Summary

Before calling the scoring API, aggregate the following per account:

{
  "account_id": "company_123",
  "account_name": "Acme Corp",
  "plan": "pro",
  "mrr": 299,
  "tenure_months": 8,
  "tickets_last_30_days": 7,
  "tickets_last_90_days": 12,
  "avg_monthly_tickets_historical": 2,
  "categories": {"bug": 4, "how-to": 2, "billing": 1},
  "severities": {"critical": 1, "high": 2, "medium": 3, "low": 1},
  "avg_csat_rating": 2.8,
  "repeat_issues": ["csv-export-fails", "slow-dashboard-load"],
  "sentiment_trend": "declining",
  "last_ticket_date": "2025-03-28",
  "escalation_count": 2,
  "competitor_mentions": ["competitor-x"],
  "cancellation_intent_signals": 1
}

API Call

POST https://api.anthropic.com/v1/messages
{
  "model": "claude-sonnet-4-20250514",
  "max_tokens": 1000,
  "messages": [{
    "role": "user",
    "content": "You are a churn risk scoring agent. Given an account's support ticket history, produce a churn risk score and recommended action.\n\nAccount data:\n{account_summary_json}\n\nScoring rules:\n- Base score starts at 0\n- Ticket volume: +5 per ticket above their monthly average (last 30 days)\n- Severity: +15 per critical, +8 per high\n- CSAT below 3.0: +15\n- Repeat issues (same problem twice+): +10 per repeat\n- Declining sentiment trend: +10\n- Escalations: +10 per escalation\n- Competitor mentions: +15 per mention\n- Cancellation intent: +25\n- Billing-category tickets: +10\n- Tenure < 3 months with high tickets: +10 (new customer struggling)\n- Cap at 100\n\nRespond in JSON:\n{\n  \"score\": <0-100>,\n  \"risk_level\": \"low|medium|high|critical\",\n  \"top_signals\": [\"signal 1\", \"signal 2\", \"signal 3\"],\n  \"recommended_action\": \"specific action\",\n  \"urgency\": \"immediate|this-week|this-month\",\n  \"talking_points\": [\"point for CS rep to address\"]\n}"
  }]
}

Risk Level Thresholds

• Low (0-25): Normal ticket patterns. No intervention needed. Monitor.
• Medium (26-50): Elevated activity. Add to watch list. Proactive check-in within 2 weeks.
• High (51-75): Clear distress signals. CS outreach within 48 hours. Prepare retention offer.
• Critical (76-100): Imminent churn risk. Immediate escalation to CS lead or founder. Personal call within 24 hours.

Output

Store the score and signals in your CRM (Attio) as custom attributes on the company record:

PUT https://api.attio.com/v2/objects/companies/records/{record_id}
{
  "data": {
    "values": {
      "support_churn_score": [{"value": 72}],
      "support_risk_level": [{"value": "high"}],
      "support_top_signals": [{"value": "3 critical bugs in 30 days, competitor mention, declining CSAT"}],
      "support_recommended_action": [{"value": "CS call within 48h to address recurring export failures"}],
      "support_score_updated_at": [{"value": "2025-03-28T10:00:00Z"}]
    }
  }
}

Also fire a PostHog event for trend tracking:

POST https://app.posthog.com/capture/
{
  "api_key": "{posthog_project_key}",
  "event": "support_churn_score_calculated",
  "distinct_id": "{account_id}",
  "properties": {
    "score": 72,
    "risk_level": "high",
    "tickets_30d": 7,
    "source": "support-issue-tracking"
  }
}

Cost

• Claude Sonnet per scoring call: ~$0.01-0.02
• At 100 accounts scored weekly: ~$4-8/month
• At 1000 accounts scored weekly: ~$40-80/month