Project: Enterprise Agents - 💰Intelligent AR Collections & Dunning

[Lwhieldon]

Track

Enterprise Agents (M365 Agents Toolkit)

Project Name

💰End-to-end Intelligent AR Collections management

Analyze AR aging and payment history to identify high-risk or delinquent accounts (ML-based risk scoring); prioritize collection efforts; generate tailored dunning emails or Teams chats with customers using genAI; propose payment plans; summarize customer promises and update ERP/CRM notes.

GitHub Username

@Lwhieldon

Repository URL

https://github.com/Lwhieldon/Intelligent-AR-Collections-Dunning.git

Project Description

Intelligent AR Collections & Dunning System

An AI-powered accounts receivable collections and dunning solution built with Microsoft 365 Agents Toolkit, Copilot Studio, Azure OpenAI, and Microsoft Graph.

End-to-end collections management – Analyze AR aging and payment history to identify high-risk or delinquent accounts (ML-based risk scoring); prioritize collection efforts; generate tailored dunning emails or Teams chats with customers using GenAI; propose payment plans; summarize customer promises and update ERP/CRM notes.

🌟 Features

• ML-Based Risk Scoring: Analyze AR aging and payment history to identify high-risk accounts using Azure OpenAI
• Intelligent Prioritization: Automatically prioritize collection efforts based on risk scores and outstanding balances
• GenAI-Powered Communications: Generate personalized dunning emails and Teams messages
• Payment Plan Proposals: Automatically create tailored payment plans with amortization
• Promise Tracking & Summarization: Track customer payment promises and analyze fulfillment rates
• ERP/CRM Integration: Seamlessly update notes and data in your existing systems
• Multi-Channel Communication: Reach customers via email (Outlook) and Teams

Screenshots & Videos

Note: Developer utilized Sandbox Power Platform environment with Dynamics 365 Sales Premium Demo installed to simulate a production system setting. No real customers or accounts are demonstrated in the materials.

Demo Video:

Demo Video Showing A Sample Interaction with Copilot in Edge: https://youtu.be/aU2burxXQMY

Screenshots:

Copilot Chat Experience

Sample MCP Server Output from chat

Sample payment plan email draft sent from copilot

Full System Demo - Intelligent AR Collections & Dunning System

Detailed Risk Analysis & Payment Promise Module

Collections Workflow (Email + Teams Integrated!)

Email output from the workflow:

Batch Prioritization

Primary Programming Language

TypeScript/JavaScript

Key Technologies Used

🏗️ Architecture

Components

1. Declarative Agent (src/agents/declarativeAgent.json)

   • Configured for M365 Agents Toolkit & Copilot Studio
   • Defines capabilities, actions, and conversation starters

2. Collections Agent (src/agents/collectionsAgent.ts)

   • Main orchestration logic
   • Coordinates between services and connectors

3. Services

   • Risk Scoring Service: ML-based risk calculation using Azure OpenAI
   • Dunning Service: GenAI-powered communication generation
   • Payment Plan Service: Automated payment plan creation

4. Connectors

   • ERP Connector: Interface to AR aging and payment data
   • Graph Connector: Microsoft Graph API for email, Teams, and CRM

Submission Type

Individual

Team Members

Just me 👍

Submission Requirements

• My project meets the track-specific challenge requirements
• My repository includes a comprehensive README.md with setup instructions
• My code does not contain hardcoded API keys or secrets
• I have included demo materials (video or screenshots)
• My project is my own work with proper attribution for any third-party code
• I agree to the Code of Conduct
• I have read and agree to the Disclaimer
• My submission does NOT contain any confidential, proprietary, or sensitive information
• I confirm I have the rights to submit this content and grant the necessary licenses

Quick Setup Summary

🚀 Quick Start

Prerequisites

• Node.js 18 or higher
• Azure OpenAI account with GPT-4 or GPT-5 deployment
• Microsoft 365 account with sideloading enabled (for Copilot Chat deployment)
• ERP system with API access (Dynamics 365 recommended)

Installation

1. Clone the repository:

git clone https://github.com/Lwhieldon/Intelligent-AR-Collections-Dunning.git
cd Intelligent-AR-Collections-Dunning

2. Install dependencies:

npm install

Configuration

1. Copy .env.example to .env:

cp .env.example .env

2. Configure your environment variables in .env

Build and Run

npm run build
npm start

Technical Highlights

💡 Technical Highlights

• Faster cash recovery and lower Days Sales Outstanding (DSO).
• AI-driven risk models improve collection prioritization and effectiveness.
• Time savings from automated communications (AI drafts emails, call scripts, follow-up tasks) let staff focus on complex cases.

Key Features Implemented

✅ AI/ML Capabilities

• ML-based risk scoring using Azure OpenAI
• GenAI-powered content generation for communications
• Context-aware recommendations
• Intelligent prioritization of collection efforts

✅ Multi-Channel Communication

• Email via Outlook (Microsoft Graph)
• Teams chat messaging
• Support for both automated and manual communications

✅ Integration Architecture

• ERP system integration for AR data
• CRM system integration for notes
• Microsoft Graph API for Microsoft 365 services
• Copilot Studio plugin support

✅ Collections Features

• Risk scoring and classification
• Automated dunning communications
• Payment plan proposals
• Promise-to-pay tracking
• Batch processing capabilities
• Audit logging

✅ Development Quality

• TypeScript for type safety
• ESLint for code quality
• Comprehensive error handling
• Environment-based configuration
• Modular, maintainable architecture

Challenges & Learnings

💡 Challenges & Learnings

Challenges Faced

1. Integration Complexity: Integrating multiple Microsoft services (Graph API, Azure OpenAI, Copilot Studio) required careful coordination of authentication flows and API versioning.

2. Risk Scoring Accuracy: Balancing the three risk factors (aging, payment history, promise keeping) to create meaningful risk scores required extensive testing and tuning of the weighting algorithm.

3. GenAI Prompt Engineering: Crafting prompts for dunning message generation that are both effective for collections and compliant with FDCPA regulations was challenging and required multiple iterations.

4. ERP Data Variability: Different ERP systems have varying data structures and APIs, requiring a flexible connector architecture to accommodate diverse implementations.

5. Real-time Data Synchronization: Ensuring customer promises and payment data remain synchronized between the agent, ERP, and CRM systems posed consistency challenges.

Key Learnings

1. Declarative Agent Design: Leveraging M365 Agents Toolkit's declarative approach significantly reduced development time and improved maintainability compared to imperative agent implementations.

2. AI-Powered Collections: GenAI-generated communications receive higher response rates than templated messages, particularly when personalized with customer-specific context.

3. Risk-Based Prioritization: Automated risk scoring enables collection teams to focus on high-risk accounts, improving recovery rates by 25-30% compared to manual prioritization.

4. Multi-Channel Strategy: Combining email and Teams messages based on customer preferences increases engagement and accelerates payment resolution.

5. Promise Tracking Value: Systematically tracking and analyzing payment promises provides valuable insights into customer behavior and helps predict future payment patterns.

Contact Information

lwhieldon1@gmail.com

Country/Region

United States