Conversational Design of a Medical Insurance Chatbot

Genesys/BCBS Tennessee

The Project

Blue Cross Blue Shield of Tennessee was building a chatbot to support member service inquiries. The purpose of this chatbot was to collect the customer’s intent and then route the customer to the correct customer service representative.

The Work

The project involved:

  • Working with the business team to review sample and tag sample data.

  • Developing a tagging guide to summarize intents and entities

  • Creating the detailed callflows for the chatbot

  • Writing all messages and creating a detailed design document for the chatbot

  • Tools included Genesys Whiteboard

The Process

Analyze training data

To create the chatbot, the first step was to understand the types of inquiries that customers might have. BCBS TN had collected sample customer requests from a variety of sources. The first step was to review these requests with the business team, develop a set of possible intents, and tag each request with the correct intent. This was an iterative process done over several weeks of review sessions. The end result was a set of training data that could be used to train the chatbot.

Create tagging guide

Once all of the customer requests had been tagged, the set of intents was summarized in a tagging guide.

Identify and define entities

Along with the intents, the key entities were extracted from the data and defined. The entities for this chatbot included items like medication names, procedure names, doctor types and pharmacy names.

Design Chat Flows

With the intents and entities defined, we now had the information needed to design and specify the chatbot. The first step of the design process was to create a detailed flow specifying the chatbot design.

Write Design Specification

The final part of the design process was to write a detailed design document specifying all chatbot messages, interactions and navigation. Once complete, the tagging guide, chat flows and design specification were provided to the development team to build the chatbot.