Skip to content

README.md

Latest commit

 

History

History
177 lines (148 loc) · 7.79 KB

README.md

File metadata and controls

177 lines (148 loc) · 7.79 KB

Customer Engagement

AI-powered customer engagement system that predicts churn risk, generates personalized outreach strategies, and recommends policy optimizations for insurance retention teams.

Overview

The Customer Engagement use case coordinates three specialist agents to improve policyholder retention. It analyzes behavioral signals and policy history to predict churn probability, designs targeted outreach campaigns with optimal channel and timing selection, and identifies policy adjustments (bundling, coverage changes, discounts) that increase customer lifetime value.

Business Value

  • Proactive retention -- Early churn risk identification enables intervention before customers lapse
  • Personalized outreach -- Channel, timing, and messaging recommendations tailored to each customer's profile and preferences
  • Revenue protection -- Policy bundling and coverage optimization reduce churn while increasing premium volume
  • Resource allocation -- Risk-scored prioritization ensures retention teams focus on highest-impact customers
  • Measurable ROI -- Estimated savings calculations and retention probability forecasts support business cases

Architecture

graph TB
    Request["Client Request"] --> Runtime["AgentCore Runtime"]
    Runtime --> Orchestrator["Customer Engagement<br/>Orchestrator"]
    Orchestrator --> Agent1["Churn Predictor<br/><small>Behavioral patterns,<br/>policy history analysis</small>"]
    Orchestrator --> Agent2["Outreach Agent<br/><small>Personalized messaging,<br/>channel selection</small>"]
    Orchestrator --> Agent3["Policy Optimizer<br/><small>Coverage adequacy,<br/>bundling opportunities</small>"]
    Agent1 --> Bedrock["Amazon Bedrock<br/>(Claude)"]
    Agent2 --> Bedrock
    Agent3 --> Bedrock
    Agent1 --> S3["S3 Sample Data<br/>(Customer Profiles)"]
    Agent2 --> S3
    Agent3 --> S3
    Agent1 --> Synthesis["Result Synthesis"]
    Agent2 --> Synthesis
    Agent3 --> Synthesis
    Synthesis --> Response["Response"]
Loading

Directory Structure

use_cases/customer_engagement/
├── README.md
└── src/
    ├── __init__.py                              # Framework router + registry
    ├── strands/
    │   ├── __init__.py
    │   ├── config.py                            # CustomerEngagementSettings
    │   ├── models.py                            # EngagementRequest / EngagementResponse
    │   ├── orchestrator.py                      # CustomerEngagementOrchestrator
    │   └── agents/
    │       ├── __init__.py
    │       ├── churn_predictor.py
    │       ├── outreach_agent.py
    │       └── policy_optimizer.py
    └── langchain_langgraph/
        ├── __init__.py
        ├── config.py
        ├── models.py
        ├── orchestrator.py
        └── agents/
            ├── __init__.py
            ├── churn_predictor.py
            ├── outreach_agent.py
            └── policy_optimizer.py

Agentic Design

The CustomerEngagementOrchestrator extends StrandsOrchestrator and uses a parallel fan-out / synthesize pattern:

  1. Fan-out -- For full assessments, all three agents run in parallel via asyncio.gather (async) or run_parallel (sync), each independently retrieving customer data from S3.
  2. Targeted modes -- churn_prediction_only, outreach_only, and policy_optimization_only run individual agents for focused analysis.
  3. Synthesis -- Agent results are assembled into section-labeled markdown (Churn Prediction, Outreach Strategy, Policy Optimization). The orchestrator LLM produces a final summary with overall churn risk, recommended retention strategy, priority actions, expected retention probability, and cost-benefit analysis.

Agents

Churn Predictor

  • Role: Predicts customer churn risk by analyzing behavioral patterns, policy history, claims frequency, payment regularity, and engagement metrics
  • Data: Customer profile and history from S3 (data_type='profile')
  • Produces: Risk level (low/moderate/high/critical), churn probability (0-1), risk factors, behavioral signals, retention window (days)
  • Tool: s3_retriever_tool

Outreach Agent

  • Role: Generates personalized retention outreach strategies with optimal channel, timing, and messaging
  • Data: Customer profile from S3
  • Produces: Recommended channel (email/phone/SMS/in_app/mail), secondary channels, messaging theme, talking points, optimal timing, personalization elements
  • Tool: s3_retriever_tool

Policy Optimizer

  • Role: Recommends policy adjustments to improve customer value and reduce churn likelihood
  • Data: Customer profile and policy data from S3
  • Produces: Recommended actions (renew/upgrade/bundle/discount/adjust_coverage), coverage adjustments, bundling opportunities, estimated annual savings, value improvements
  • Tool: s3_retriever_tool

Data & Tools

Resource Description
s3_retriever_tool Retrieves customer profiles, policy data, and engagement history from S3
S3 path data/samples/customer_engagement/{customer_id}/profile.json

Request / Response

EngagementRequest

Field Type Description
customer_id str Customer identifier (e.g., POLICY001)
assessment_type AssessmentType full, churn_prediction_only, outreach_only, policy_optimization_only
additional_context str | None Optional context

EngagementResponse

Field Type Description
customer_id str Customer identifier
engagement_id str Unique engagement assessment UUID
timestamp datetime Assessment timestamp
churn_prediction ChurnPrediction | None Risk level, probability, risk factors, retention window
outreach_plan OutreachPlan | None Channel, messaging theme, talking points, timing
policy_recommendations PolicyRecommendations | None Actions, bundling, estimated savings
summary str Executive summary
raw_analysis dict Raw output from each agent

Example Request:

{
  "customer_id": "POLICY001",
  "assessment_type": "full"
}

Example Response:

{
  "customer_id": "POLICY001",
  "engagement_id": "uuid",
  "timestamp": "2026-03-25T00:00:00Z",
  "churn_prediction": {
    "risk_level": "moderate",
    "churn_probability": 0.45,
    "risk_factors": ["declining engagement score", "recent claims"],
    "behavioral_signals": ["reduced app logins", "skipped last email"],
    "retention_window_days": 90
  },
  "outreach_plan": {
    "recommended_channel": "email",
    "messaging_theme": "loyalty appreciation",
    "talking_points": ["tenure recognition", "bundling savings opportunity"],
    "optimal_timing": "mid-week morning"
  },
  "policy_recommendations": {
    "recommended_actions": ["bundle", "discount"],
    "bundling_opportunities": ["auto + home multi-policy discount"],
    "estimated_savings": 350.00
  },
  "summary": "Moderate churn risk. Recommend proactive outreach with bundling discount offer."
}

Quick Start

USE_CASE_ID=customer_engagement FRAMEWORK=strands AWS_REGION=us-east-1 \
  ./applications/fsi_foundry/scripts/deploy/full/deploy_agentcore.sh

Sample Data

Customer ID Risk Profile Description
POLICY001 Active Customer Comprehensive auto + home, 5yr tenure

Related Documentation