Relieving Call Centers with AI

Dr. Dietrich Wettschereck
Tags
Machine Learning
Data & Analytics
Automation
AI
Innovation
Published 13. February 2024

The white paper describes the implementation of a ChatGPT-powered VoiceBot for efficient handling of customer orders in call centers, using Azure AI Speech for natural language interaction. The system aims to reduce call center workload, improve customer experience and is scalable at a very reasonable cost per call. Qvest provides support for the implementation of innovative AI projects.

A VoiceBot in Real Life

On behalf of our customer, we at Qvest developed a prototype of a VoiceBot that, based on the Azure services OpenAI and AI-Speech, queries information from callers in a free dialog in order to generate a structured service case. In the specific project, the VoiceBot had to query the company name of the calling customer, the zip code and the quantity and size of the items that a logistics company commissioned by our customer was to collect from the caller. Imagine that the items in question are old electronic devices or raw materials that need to be recycled and that accumulate at irregular intervals at the caller's premises in terms of type and quantity.

The aim of the project was to supplement current services with innovative solutions and future-oriented technologies. Currently, the majority of callers use an existing call center solution to place service orders. This leads to the call center being overloaded, especially at peak times, and thus to unpopular waiting times when ordering a simple service. The VoiceBot described here is therefore intended to relieve customer service and delight customers. In addition, such a VoiceBot in the cloud can be scaled on-demand with practically no limits and additional functionality can be added with relatively little effort.

For our first prototype, the VoiceBot should receive a phone call to find out the name and zip code of the store or market for which the end customer is calling, as well as the type and number of bins to be collected. The conversation with the bot should be as natural as possible and not follow a fixed structure. In later expansion stages, the VoiceBot will also answer questions about the recycling process, process other service requests and generally serve as the first point of contact for all of the customer's services. Callers will only be forwarded to the call center for complicated issues or if they explicitly request it.

System Architecture

A VoiceBot can be implemented in Microsoft Azure with relatively few components. The intelligent components of the system are Azure AI Speech  for converting spoken language into text and vice versa and Azure OpenAI Service for ChatGPT. Integration into a VoiceOverIP system that calls the VoiceBot via a REST API, for example, is still necessary to achieve full functionality. In the current use case, the connection was established via Azure Communication Services.

The following figure illustrates how the individual components are used. The entire process is repeated until the caller hangs up or, optionally, the VoiceBot ends the call because all the necessary information for the respective use case has been requested.

Get this white paper
Would you like to learn more?
Enter your data here and click on "Download now“. You will receive a download link to your email address afterwards. Please note that the delivery may take a few minutes.
Request instant free download access to the full white paper
Relieving Call Centers with AI - How ChatGPT is used to accept customer orders without waiting times
Agree to data processing*