YEAR
2023-2025
DELIVERABLES
conversation design
content strategy
flow design
node architecture
intent mapping
Nora Chatbot
As a Conversational AI Designer at Nationwide, I work with the Digital Assistants team to refine, rebuild and transform our insurance and retirement solutions chatbots.
To get started, let's take a quick look at how our conversational style guide shaped my content choices for Nora. Next, I'll dive into the flow-building process.
Refining and establishing Nora's persona
I teamed up with a fellow conversational designer to develop a comprehensive style guide that defined our chatbot's persona and voice. By outlining tone, style, grammar, and mechanics, we ensured consistency across interactions. This guide played a crucial role in rebuilding Nora's NLU model, laying the foundation for more effective and engaging conversations.
Who is Nora?
Nora is a digital assistant that does not "consider" herself human. She is designed to be a non-agentic helper driven by NLU models that respond to common member inquiries while maintaining Nationwide customer experience principles and ultimately lower call volume.
Nora's Qualities
Reassuring
Warm
Personal
Confident
Straightforward
Helpful
Grammar & mechanic guidelines
Here are some rules we live by to make sure Nora's voice is consistent across all intents:
-
Contractions are a great way to create a casual, approachable tone. Use them often.
-
Numbers should be spelled out unless it is in reference to a year or in reference to money.
-
Avoid ellipses...there is no situation where Nora should be trailing off in thought.
-
Use exclamation points extremely sparingly: these can be interpreted differently person to person which can cloud the intended message.
-
Avoid emojis and iconography: emojis are not always operating system universal and create a more casual tone than the intent of Nora.
Designing conversation flow charts
In an effort to improve Nora's NLU model, I worked with our dev team to create dozens of flowcharts in Figma to help identify user pain points, create new intents, and restructure conversational flows across account management, claims, billing and payments.
.png)
Building conversations in Voiceflow
I worked closely with our dev team to rebuild those conversations in Voiceflow, incorporating feedback from our business and customer support teams to craft more effective responses. At this stage, I also wrote hundreds of sample utterances that our team used to test and refine Nora. Below is a preliminary view of the conversation design work I did in Voiceflow.
The result:
After releasing MVP3 of Nora in Q1 of 2024, we were able to achieve our goal of at least 80% successful intent matches across all domains. This improved our overall NLU accuracy by 38%.
