Undersökning av hur agenters och kunders beteende påverkar försäljning i B2C callcenters med hjälp av stora språkmodeller och logistisk regression
Year: 2025
Type: dissertation
Abstract: This thesis investigates which agent and customer behaviours determine sales success in business-to-consumer call-centre conversations. A total of 8,479 inbound telecom calls were automatically transcribed using a Swedish-tuned Whisper model and annotated by GPT-4o against 34 agent and customer behaviour questions, derived from literature, thematic transcript coding, and semi-structured practitioner interviews with both agents and managers. Of these, 2,182 calls containing an explicit sales offer were analysed using logistic regression with sequential feature selection. The final 13-variable model (accuracy = 0.70) shows that sales probability rises sharply when agents (i) link their proposal to the customer’s stated needs (odds ratio x4.2), (ii) reassure hesitant customers by rephrasing benefits (x2.5), and (iii) solve the customer’s initial concern (x2.1). In contrast, offering multiple alternatives (x0.7), fact-heavy objection handling (x0.7), and from the customer side, stalls such as repeating previous answers (x0.5) or declaring the offer irrelevant (x0.2) strongly reduce sales conversion. This study demonstrates how large language models can scale sales-behaviour research from dozens to thousands of calls, while also highlighting the need for rigorous prompt design and validation to curb subjectivity. The results provide call-centre managers with concrete, data-backed coaching levers and offer a methodological template for future speech-analytics research.
Authors: BILL, DESIRÉE, LINDH, KOLUMBUS
Language: English
Subfield: Marketing
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7082420422
