Automated Analysis of Call Transcripts Using Large Language Models

Analysis of customer service call transcripts can be challenging due to high volumes of unstructured textual data. To overcome this challenge, Co-operators Insurance Group seeks to explores the application of large language models (LLMs) such as GPT3 in analyzing call transcripts to extract structured data to be used in further modelling. Specifically, the project aims to design an end-to-end pipeline to filter call transcripts based on user keywords and call outcomes, extract transcript data using LLMs and format extracted information in a structured dataset. Extracted data from call transcripts could include classifications of the sentiment of the client and/or call representative, classification of client-expressed reason(s) for cancellation, and identification of call representative tactics attempting to retain the client. Co-operators insurance aims to apply this pipeline to the analysis of calls relating to mid-policy cancellations to understand the causes of these cancellations and potential methods for mitigating cancellations using insights related to customer service approach.

Faculty Supervisor:

Ayesha Ali;Fred Liu

Student:

Partner:

Co-operators (General Insurance)

Discipline:

Computer science

Sector:

Finance and Insurance

University:

University of Guelph

Program: