> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pylar.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Customer Churn Predictor

> Build an AI agent that identifies at-risk customers and recommends retention actions using usage and engagement data

## Overview

A Customer Churn Predictor agent powered by Pylar analyzes customer usage patterns, engagement metrics, support history, and billing data to identify customers at risk of churning and recommend proactive retention actions.

## What the Agent Needs to Accomplish

The agent must:

* Analyze customer usage patterns and engagement metrics
* Track support ticket frequency and sentiment
* Monitor billing and payment history
* Identify churn risk indicators
* Generate churn risk scores
* Recommend specific retention actions
* Track retention campaign effectiveness

## How Pylar Helps

Pylar enables the agent by:

* **Unified Customer Data**: Combining usage, engagement, support, and billing data
* **Real-time Analysis**: Querying current customer behavior patterns
* **Multi-Source Integration**: Joining data from product analytics, CRM, support, and billing systems
* **Pattern Recognition**: SQL views identify churn indicators
* **Actionable Insights**: Generating retention recommendations based on data

```mermaid theme={null}
graph LR
    subgraph "Data Sources"
        ANALYTICS[Product Analytics]
        CRM[CRM]
        SUPPORT[Support]
        BILLING[Billing]
    end
    
    PYLAR[Pylar<br/>Unified Views]
    
    TOOLS[MCP Tools<br/>calculate_churn_risk<br/>get_usage_patterns<br/>get_retention_actions]
    
    AGENT[Churn Predictor]
    TEAM[CS Team]
    
    ANALYTICS --> PYLAR
    CRM --> PYLAR
    SUPPORT --> PYLAR
    BILLING --> PYLAR
    PYLAR --> TOOLS
    TOOLS --> AGENT
    AGENT --> TEAM
    
    style PYLAR fill:#FF4017,stroke:#CC3300,color:#fff
```

## Without Pylar vs With Pylar

### Without Pylar

**Challenges**:

* ❌ Data scattered across multiple systems (analytics, CRM, support, billing)
* ❌ Complex ETL pipelines to combine data
* ❌ Difficult to get real-time churn risk scores
* ❌ Manual analysis of churn indicators
* ❌ No unified customer view
* ❌ Time-consuming to update models
* ❌ Limited ability to track retention actions

**Implementation Complexity**:

* 4-5 different data source integrations
* Complex data pipeline setup
* Custom ML model development
* Churn scoring infrastructure
* \~6-8 weeks development time

### With Pylar

**Benefits**:

* ✅ Single endpoint for all customer data
* ✅ Real-time churn risk calculation
* ✅ Easy to update churn indicators
* ✅ Unified customer 360 view
* ✅ Built-in analytics on retention actions
* ✅ Simple query-based risk scoring

**Implementation Complexity**:

* Connect 4-5 data sources (1.5 hours)
* Create churn analysis views (3 hours)
* Build MCP tools with AI (1.5 hours)
* Connect to agent builder (15 minutes)
* **Total: \~6-7 hours**

## Step-by-Step Implementation

### Step 1: Connect Data Sources

Connect all customer data sources:

1. **Connect Product Analytics** (Usage metrics, feature adoption, engagement)
2. **Connect CRM** (Account data, contract info, renewal dates)
3. **Connect Support System** (Ticket history, satisfaction scores)
4. **Connect Billing System** (Payment history, invoices, subscription status)

### Step 2: Create Churn Analysis Views

**Customer Engagement Score View**:

```sql theme={null}
CREATE VIEW customer_engagement_score AS
SELECT 
  c.customer_id,
  c.customer_name,
  c.subscription_tier,
  c.contract_start_date,
  c.contract_end_date,
  DATEDIFF(CURRENT_DATE, c.contract_end_date) as days_until_renewal,
  -- Usage metrics
  u.active_days_last_30,
  u.features_used_count,
  u.total_actions_last_30,
  u.last_activity_date,
  DATEDIFF(CURRENT_DATE, u.last_activity_date) as days_since_last_activity,
  -- Support metrics
  s.ticket_count_last_30,
  s.avg_satisfaction_score,
  s.avg_resolution_time_hours,
  -- Billing metrics
  b.total_invoices,
  b.payment_failures_count,
  b.days_overdue,
  -- Engagement score calculation
  CASE 
    WHEN u.active_days_last_30 >= 20 THEN 100
    WHEN u.active_days_last_30 >= 15 THEN 75
    WHEN u.active_days_last_30 >= 10 THEN 50
    WHEN u.active_days_last_30 >= 5 THEN 25
    ELSE 0
  END as usage_score,
  CASE 
    WHEN s.avg_satisfaction_score >= 4.5 THEN 100
    WHEN s.avg_satisfaction_score >= 4.0 THEN 75
    WHEN s.avg_satisfaction_score >= 3.5 THEN 50
    ELSE 25
  END as satisfaction_score
FROM crm.customers c
LEFT JOIN analytics.user_usage u ON c.customer_id = u.customer_id
LEFT JOIN support.ticket_summary s ON c.customer_id = s.customer_id
LEFT JOIN billing.payment_summary b ON c.customer_id = b.customer_id;
```

**Churn Risk Score View**:

```sql theme={null}
CREATE VIEW churn_risk_score AS
SELECT 
  e.*,
  -- Risk indicators
  CASE 
    WHEN e.days_since_last_activity > 30 THEN 1 ELSE 0
  END as low_activity_flag,
  CASE 
    WHEN e.ticket_count_last_30 > 5 THEN 1 ELSE 0
  END as high_support_flag,
  CASE 
    WHEN e.payment_failures_count > 0 THEN 1 ELSE 0
  END as payment_issue_flag,
  CASE 
    WHEN e.days_until_renewal < 30 THEN 1 ELSE 0
  END as renewal_approaching_flag,
  -- Churn risk score (0-100, higher = more risk)
  (
    (CASE WHEN e.days_since_last_activity > 30 THEN 30 ELSE 0 END) +
    (CASE WHEN e.active_days_last_30 < 10 THEN 25 ELSE 0 END) +
    (CASE WHEN e.ticket_count_last_30 > 5 THEN 20 ELSE 0 END) +
    (CASE WHEN e.payment_failures_count > 0 THEN 15 ELSE 0 END) +
    (CASE WHEN e.avg_satisfaction_score < 3.5 THEN 10 ELSE 0 END)
  ) as churn_risk_score,
  -- Risk level
  CASE 
    WHEN (
      (CASE WHEN e.days_since_last_activity > 30 THEN 30 ELSE 0 END) +
      (CASE WHEN e.active_days_last_30 < 10 THEN 25 ELSE 0 END) +
      (CASE WHEN e.ticket_count_last_30 > 5 THEN 20 ELSE 0 END) +
      (CASE WHEN e.payment_failures_count > 0 THEN 15 ELSE 0 END) +
      (CASE WHEN e.avg_satisfaction_score < 3.5 THEN 10 ELSE 0 END)
    ) >= 50 THEN 'High Risk'
    WHEN (
      (CASE WHEN e.days_since_last_activity > 30 THEN 30 ELSE 0 END) +
      (CASE WHEN e.active_days_last_30 < 10 THEN 25 ELSE 0 END) +
      (CASE WHEN e.ticket_count_last_30 > 5 THEN 20 ELSE 0 END) +
      (CASE WHEN e.payment_failures_count > 0 THEN 15 ELSE 0 END) +
      (CASE WHEN e.avg_satisfaction_score < 3.5 THEN 10 ELSE 0 END)
    ) >= 25 THEN 'Medium Risk'
    ELSE 'Low Risk'
  END as risk_level
FROM customer_engagement_score e;
```

**Retention Action Recommendations View**:

```sql theme={null}
CREATE VIEW retention_actions AS
SELECT 
  r.*,
  CASE 
    WHEN r.low_activity_flag = 1 THEN 'Send re-engagement email campaign'
    WHEN r.high_support_flag = 1 THEN 'Schedule success manager call'
    WHEN r.payment_issue_flag = 1 THEN 'Contact billing team for payment assistance'
    WHEN r.renewal_approaching_flag = 1 AND r.churn_risk_score > 30 THEN 'Offer renewal discount'
    WHEN r.avg_satisfaction_score < 3.5 THEN 'Request feedback and address concerns'
    ELSE 'Monitor closely'
  END as recommended_action,
  CASE 
    WHEN r.churn_risk_score >= 50 THEN 'Urgent'
    WHEN r.churn_risk_score >= 25 THEN 'High Priority'
    ELSE 'Normal'
  END as action_priority
FROM churn_risk_score r
WHERE r.churn_risk_score > 20;
```

### Step 3: Create MCP Tools with AI

**Tool 1: Get Churn Risk Score**

* Prompt: "Create a tool to get churn risk score for a customer by customer ID"
* AI generates: `get_churn_risk(customer_id: string)`

**Tool 2: List High-Risk Customers**

* Prompt: "Create a tool to list customers with high churn risk scores"
* AI generates: `get_high_risk_customers(risk_level: string, limit: number)`

**Tool 3: Get Retention Recommendations**

* Prompt: "Create a tool to get recommended retention actions for at-risk customers"
* AI generates: `get_retention_recommendations(customer_id: string)`

**Tool 4: Analyze Churn Patterns**

* Prompt: "Create a tool to analyze churn patterns and identify common indicators"
* AI generates: `analyze_churn_patterns(days_back: number, min_risk_score: number)`

### Step 4: Test and Publish

1. Test churn risk calculation
2. Verify high-risk customer identification
3. Validate retention recommendations
4. Publish tools
5. Connect to agent builder

## Example Agent Interactions

### Scenario 1: High-Risk Customer Identification

**User**: "Show me customers at high risk of churning"

**Agent** (using Pylar tools):

1. Calls `get_high_risk_customers("High Risk", 50)`
2. Analyzes results:
   * "I found 23 high-risk customers:
     * Acme Corp: 65 risk score, last active 42 days ago, 3 payment failures
     * TechCo: 58 risk score, 8 support tickets last month, low satisfaction
     * ..."

### Scenario 2: Retention Action Recommendation

**User**: "What should we do about customer ID 12345?"

**Agent** (using Pylar tools):

1. Calls `get_churn_risk("12345")`
2. Calls `get_retention_recommendations("12345")`
3. Responds:
   * "Customer 12345 has a 72 churn risk score (High Risk)
   * Indicators: No activity for 38 days, 6 support tickets, payment failure
   * Recommended action: Schedule success manager call + offer renewal discount
   * Priority: Urgent"

### Scenario 3: Churn Pattern Analysis

**User**: "What are the common patterns in churned customers?"

**Agent** (using Pylar tools):

1. Calls `analyze_churn_patterns(90, 50)`
2. Analyzes patterns:
   * "Common churn indicators in last 90 days:
     1. 68% had no activity for 30+ days
     2. 45% had payment failures
     3. 52% had low satisfaction scores (less than 3.5)
     4. 38% had 5+ support tickets"

## Outcomes

### Churn Prevention

* **Early Detection**: 40% earlier identification of at-risk customers
* **Retention Rate**: 25% improvement in customer retention
* **Proactive Actions**: 3x increase in proactive retention outreach
* **Churn Reduction**: 30% reduction in overall churn rate

### Data-Driven Decisions

* **Risk Scoring**: Automated, real-time churn risk scores
* **Action Recommendations**: Data-driven retention strategies
* **Pattern Recognition**: Identification of common churn indicators
* **Effectiveness Tracking**: Monitor retention campaign success

### Business Impact

* **Revenue Protection**: Save \$500K+ in annual revenue through retention
* **Customer Lifetime Value**: 20% increase in average CLV
* **Customer Satisfaction**: Improved satisfaction through proactive outreach
* **Team Efficiency**: 50% reduction in manual churn analysis time

## Best Practices

1. **Regular Monitoring**: Check churn risk scores weekly
2. **Action Follow-up**: Track retention action effectiveness
3. **Model Refinement**: Update risk indicators based on outcomes
4. **Communication**: Integrate with CRM for automatic outreach
5. **Analytics**: Use Evals to track agent churn prediction accuracy

## Next Steps

* [Customer Support Agent Example](/examples/customer-support-agent) - Build a support agent
* [Sales Assistant Example](/examples/sales-assistant) - Analyze sales pipeline
* [Revenue Operations Agent Example](/examples/revenue-operations-agent) - Track revenue metrics