Customer Service Training using Generative AI

This semester an xLab team worked on a project for Progressive Insurance. It was the first semester of the project, and the team was tasked to create a tool to train customer service agents using Generative AI. Progressive has tens of thousands of customer service agents and has high costs in training them.

Using a finetuned large language model, the students created a chatbot that can supplement the learning of customer service agents. Utilizing both Anthropic and OpenAi’s most recent models, agents interact with the chatbot as if it were a real customer allowing them to train in a close-to-reality setting. This ensures that the agents are following the curriculum properly and learning the essential knowledge to perform the job. The students also added variation to the model and have scenarios where the chatbot customer’s mood can change so that agents can be prepared to deal with any type of customer.

The training tool is tuned with past Progressive transcripts and lessons and intelligently roleplays as if it were a customer. The team also added functionality that allows an agent to speak into a device and the device can hear and transcribe what they are saying to make it even more closely simulate a customer service phone call.