Call Transcript Analysis - Using LLMs to classify customer queries

Bright School Template Mockup

10,000+

Calls analysed

76%

Auto-tag accuracy (vs manual QA)

40%

Time to insight

Client
A leading media and telecom company, offering television, broadband, mobile, and streaming services to millions of customers across the UK and Europe.
Industry
Telecom
Company Size
10,000+
Location
London, United Kingdom
Project Duration
4 months (Feb 2024 - Jun 2024)
Framework
  • VertexAI
  • BigQuery
  • LLMs
Visit Website

Developed a pipeline to analyze customer service call transcripts using LLMs for automated summarization, sentiment analysis, and issue classification. Enabled faster insight generation and quality monitoring by extracting key themes, and customer intent from thousands of conversations.

Customer service calls are a rich source of insight, but they’re hard to scale manually. In this project, I built a system that used large language models (LLMs) to process and analyse thousands of call transcripts across Sky’s service teams.

The pipeline applied automatic summarisation, sentiment analysis, and intent classification to extract key themes from each conversation. It identified recurring issues like billing confusion, service faults, or package change requests, helping teams understand what mattered most to customers at different touchpoints.

The results were integrated into QA dashboards and customer journey analysis workflows. Analysts could now monitor trends in near real-time, without relying on slow, manual tagging or spot-checking. This supported faster escalation of emerging problems and gave stakeholders visibility into areas needing operational or communication improvements.

By turning unstructured conversations into structured data, the project enabled Sky to strengthen its customer insight capability while reducing time to insight by over 40%.

Previous
Next