Call center automation with AI

Client

Under a non-disclosure agreement (NDA)

Project description

To identify the exact reasons for high costs and call center load, our team conducted an analysis of the recorded call database. It turned out that, despite the presence of a mobile application, many customers still call to book tickets.

To solve business challenges, we've developed a Proof of Concept (PoC) solution, which includes the following functions: call listening, voice to text and text to voice translation, as well as data exchange with the core back-end system. In addition, the solution offers the ability to request additional information from the client to clarify the request or transfer to the operator if necessary. All operations are performed in real-time, creating the illusion of interaction with a real person.

We replaced ChatGPT with Dialogflow for process modeling, and as a tool for extracting data from user requests, such as departure and arrival points, departure date and time, etc.. After a series of tests for speech-to-text (STT) transcription, we chose AssemblyAI because of its streaming transcription and WebSocket API. Elevenlabs was chosen as the tool for text-to-speech (TTS) conversion due to the finer voice settings compared to OpenAI.

As a result, we created a solution as a voice bot that, based on a custom flow, allowed the customer to book tickets for trips and cancel them without operator involvement.

Key activities