Automated Call Compliance Monitoring Qorus Banking Innovation Awards 2023

The bank’s Group Retail adopts strict protocols and rules to ensure calls from bank sales reps to clients adhere to legal, relevant governance and performance standards. The sales reps are provided with guidelines and checklists to ensure the call covers essential elements such as verification of client identity, finding out the client investment needs and personal risk preference, explanation of the product features and product risk. Currently compliance monitoring involves supervisors listening to voice recording samples (typically less than 1%). This is a very time-consuming exercise. The project aims to transform such voice data into analyzable transcripts using Voice-To-Text (VTT) technology and which then enables further analysis using deep Natural Language Understanding (NLU) capabilities to gain insights into sales rep compliance. The Compliance Monitoring VTT-LLM program is developed in-house by the Enterprise AI Team within the Data Management Office. The team works closely with the relevant stakeholders to first understand the compliance framework, relevant governance, and marketing documents. Open source VTT technology is optimized to transcribe calls with characteristics typically encountered by the bank. Next, the call transcript is passed into an LLM, which is finetuned and prompt-engineered for the Call Compliance Monitoring application.