How I Automated the Chargeback Dispute Resolution Process
My Role(s):
Project Lead & Architect: Designed the end-to-end workflow of the automation system, from data ingestion to generating actionable dispute responses.
Data Analyst & Engineer: Developed SQL queries and Python scripts to fetch historical transaction data, analyze patterns, and integrate multiple data sources.
AI Integration Specialist: Collaborated with a developer to fine-tune GPT-4o prompts, ensuring the model produces concise, persuasive evidence-based summaries.
Optimization & QA: Iterated on the system, refining prompts and layouts based on real-case outcomes, feedback, and win/loss analysis.
Process & Workflow:
Outcome:
This project significantly reduced time-to-resolution for disputes, improved consistency in chargeback responses, and increased the recovery rate for friendly fraud cases. The tool now empowers agents to manage a larger volume of cases with less manual effort, offering a scalable solution for future dispute handling needs.
Data Preparation and Integration:
Extracted and merged relevant transaction, authorization, and chat data using SQLAlchemy and Python.
Consolidated insights from multiple sources (e.g., historical bookings, Ekata data, AVS/CVV responses).
Evidence Gathering & Analysis:
Implemented logic to identify key messages in chat logs that either authenticated the customer or contradicted unauthorized transaction claims.
Worked with key members in various departments to gather all relevant data for a given reason label and product combination.
AI-Generated Summaries:
Developed dynamic prompts that generate customized dispute summaries based on case-specific data, such as itinerary details, AVS/CVV matches, and billing-passenger alignment.
Automated the selection and A/B testing of summaries, tracking prompt versions and outcomes to optimize recovery rates.
Dispute Template Creation:
Collaborated with developers on building a Jodit-based template editor, ensuring all sections of the dispute form were easily editable, including tables and dynamic evidence fields.
Integrated a print-friendly layout to improve usability for agents and bankers reviewing dispute cases.
Analytics and Performance Monitoring:
Set up tracking for evidence types per case and monitored agent edits to measure performance and evaluate tool effectiveness.
Analyzed win/recovery rates to compare outcomes between automated and manual dispute processes, continuously refining the system for optimal results.
