Amber
Skills Used: Conversation Design | Interaction Design | UX Research | Product Analysis | Product Pitch
Amber is Ambetter’s (CENTENE Corporation) virtual assistant. As her dedicated conversation designer, my role is to create dialogues that help Amber become a Jr. Member Services Associate with the task of guiding members through the following tasks:
Help members understand the basics of their plan (e.g., deductible, premium costs etc.)
Help members make a premium payment
Explain certain medical terminology, such as “pre-authorization”
Troubleshoot issues such as finding appropriate tax documents
Provide plan-related addresses and contact information
Help members find a medical provider
Search coverage information
At its core, Amber uses Rasa NLU, CoffeeScript and Chatette (superficially trains datasets). I work closely with my product manager, data scientist, developers, and data analyst to create, assess and finesse dialogues and flows such as the following:
Example bot-flow - Renewals (Botmock)
We wanted to explore how we might teach Amber to respond to member questions on how and if they need to renew their health insurance coverage at the end of the year. This is the second explorative iteration. While the final product cannot be shared openly, this process highlights a crucial part of my role - to explore tree-structures.
Example INTENT Analysis
Amber is made up of hundreds of intents. Above is an example of a typical intent-to-utterance analysis and mapping exercise.
A “final” product - Balance inquiries - Sketch Mock
Dialogues are always under scrutiny and evaluation. So while there is no “final” product, above is an example of a “shipped” ready-to-build mock. These are delivered to the product manager for final review and then assigned to a developer for deployment and intent-mapping.
UX Research: Prototyping & Usability tests - Proto.io
In some situations, I have to create a prototype to rapidly (3 - 7 participants) test with live users. In this scenario, I introduced a menu on the lower left of the chat view. Amber is comprised of two flows: guided and free-text. Guided flows are quick-wins for users, but they do not teach us much in terms of what people are looking for. I wanted to incentivize users to use the free-text form as much as possible. As part of that process, I decided to hide the guided-flow inside of the menu on the lower left, introduced a free-text with autocomplete functionality, and shortened Amber’s greeting to pull members to converse with her.
View original product pitch that identifies the need for this test
View UX Research plan
Listen to sample participant feedback
The test results were favorable. Members enjoyed the new, streamlined look and were excited about the autocomplete functionality. This project was successfully implemented and deployed, resulting in a 2% increase in usage and an overall increase in NLU inputs.
analysis
As part of a team, I also participate in post-conversational analysis. We have created an internal tool that downloads over 8,000 conversations with Amber. From this analysis we classify utterances, analyze the intent matched with that utterance for accuracy, assign a success rate, identify topics, and from it build a roadmap of stories Amber must answer in order to become a better virtual assistant. Here’s a sample of that analysis:
