Reducing Payment Dispute Resolution Time via AI-Augmented Smart Routing for ShopeePay 

Project Overview

This project involved designing a machine learning-powered solution to automate and optimize customer dispute handling for ShopeePay, a leading digital wallet platform in Southeast Asia. The system uses NLP models to classify support tickets and predict dispute complexity, prioritizing high-impact cases for faster handling. Integrating Google NLP and custom-trained models with internal dashboards, this solution improved service quality while reducing operational costs. 

Context / Problem Statement:

Digital wallets often suffer from high-volume customer inquiries related to failed transactions, chargebacks, and settlement delays. ShopeePay faced increased latency in dispute resolution, leading to reduced user trust and higher support costs. 

AI-Augmented Solution:

To streamline the resolution pipeline, we designed an AI-enhanced Smart Dispute Routing System that leveraged:

• Natural Language Processing (NLP) to classify inbound support tickets by intent (e.g., “refund not received”, “duplicate payment”, “dispute rejection”).

• Named Entity Recognition (NER) to extract key metadata (transaction ID, amount, timestamp) for fast retrieval.

• Predictive Prioritization Models (XGBoost + logistic regression ensemble) to flag high-risk, high-priority cases (e.g., involving VIP accounts or past escalations).

• AI-based auto-response generation using fine-tuned LLMs (for simple cases) to offer instant resolution.

AI Stack / Tools Used:

• Python (spaCy, HuggingFace Transformers)

• FastAPI for integration with CS dashboard

• GCP NLP API for multilingual parsing

• BigQuery + Looker for real-time tracking

• SQL-based anomaly flagging

• Google Vertex AI for model deployment

Outcome & Business Impact:

• ⏱️ Reduced average resolution time by 38%

• 📉 Decreased CS agent escalations by 26%

• 📈 Improved CSAT scores by 11% in Q4

• 💸 Resulted in estimated cost savings of ~RM 240K annually in manual review overhead

My Role:

As a Data Analytics Manager, I worked cross-functionally with the customer service and payment product teams. I scoped the NLP architecture, mapped ticket taxonomies, supervised data labeling, and collaborated with engineers to embed ML scoring logic into the operational dashboard. I also led A/B testing of the auto-reply module and post-launch performance tuning.