A voice bot that misunderstands a user can quickly make a simple task frustrating. Error handling helps the bot recover from recognition problems and guide users back on track. section 9.2.4
Keep recovery manageable for the user:
- Break tasks into clear steps, checking that users are ready before moving on. section 9.2.4
- When speech is misunderstood, ask a clarifying question or offer repetition; simplify responses to help restore the conversation. section 9.2.4
When required information is missing
In a claim-status chat, an LLM asks for a member ID, claim date, and claim amount. When the user does not know the requested details, the bot keeps asking for the remaining information, risking frustration rather than handling the gap. section 10.2.1
Listing 10.10 LLM generated responses for when user doesn’t have all the information
Instruction: You are a conversational designer. You are building a chatbot to help users find information about their insurance claims.
Insurance claims include the following information: a claim date, a member ID, and a claim amount.
Generate conversational responses until you have collected all three pieces of information.
When you have all the information, respond with a #1
payload in this format: {"memberID": #1
"(the member id collected)", "claimDate": #1
"(the claim date collected)", "claimAmount": #1
"(the claim amount collected)"} #1
System: How can I help you? #2
User: I need information on my claim. #2
System: I can help you with that. What is your member #2
ID? #2
User: I don't know it. #3
System: What is your claim date?
User: I don't know that either. #3
System: What is your claim amount?
User: $1000
System: I can help you with that. What is your member #4
ID? #4
See the claim-status exchange where the user lacks the requested information. section 10.2.1
