G

GenAI Voice Analytics Pipeline

Architected and deployed a GenAI-powered voice analytics pipeline at Citi, leveraging Whisper, LLMs, and embedding-based clustering to process 50M+ customer calls. Extracted complaint themes and sentiment, delivering $14M in cost savings and $72M in incremental revenue.

Citi
2023-2024
$86M total business impact through cost savings and revenue generation

Project Overview

This large-scale voice analytics solution processed millions of customer service calls to extract actionable insights. The pipeline combined state-of-the-art speech recognition with advanced natural language understanding to identify customer pain points, sentiment trends, and operational inefficiencies.

Technical implementation: • Distributed speech-to-text processing with Whisper models • Multi-stage NLP pipeline for intent and sentiment analysis • Advanced embedding techniques for call clustering and categorization • Real-time dashboard for operational insights • Automated alert system for critical customer issues • Integration with existing CRM and customer service platforms

The solution provided unprecedented visibility into customer interactions at scale, enabling data-driven improvements across the entire customer service operation.

Key Challenges

  • Processing 50M+ calls efficiently at scale
  • Handling diverse languages and accents
  • Maintaining privacy and compliance with financial regulations
  • Achieving real-time processing for live monitoring

Outcomes & Impact

  • $14M in operational cost savings
  • $72M in incremental revenue through improved customer experience
  • Reduced customer complaint resolution time by 35%
  • Enhanced customer satisfaction scores across all channels

Technologies Used

WhisperBERTFlan T5LlamaDockerKubernetes

Company

Citi

2023-2024