A chatbot developer's guide to handling erroneous situations
Eboreime, Jeffrey (2023)
2023
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2023071324570
https://urn.fi/URN:NBN:fi:amk-2023071324570
Tiivistelmä
The aim of this research was to evaluate chatbots' conversation flow designs, especially their errorhandling mechanisms and strategies. The purpose was to suggest practices to improve conversation quality and increase user retention and chatbot interactions by attempting to directly resolve common causes of failing chatbot conversations. The concern was that many conversations fail and result in escalation to second-line support channels. These types of conversations should be actively managed rather than using generic responses that may not work to restore conversation flow.
The project outcomes presented chatbot design solutions and communication strategies to improve ways of creating conversation flow. The focus was not on how chatbots ought to be built, but on how to improve responsivity when it did not know how to communicate with its users.
The studied chatbots were deployed in the finance sector. All data and findings, though anonymized, were based on real conversation history. The research showed how changing different elements of chatbots' ways of communicating with their users affected conversation flow and various indicators of conversation quality.
How a chatbot should behave in an erroneous situation may not be easily defined, as not all chatbots are deployed to serve the same function, such as a self-service chat or a contact deflection method. Methods that work for one instance of a chatbot, may result in adverse effects on another. Therefore, conversation design, system responses, and overall chatbot behavior should be tailored for each deployment
The project outcomes presented chatbot design solutions and communication strategies to improve ways of creating conversation flow. The focus was not on how chatbots ought to be built, but on how to improve responsivity when it did not know how to communicate with its users.
The studied chatbots were deployed in the finance sector. All data and findings, though anonymized, were based on real conversation history. The research showed how changing different elements of chatbots' ways of communicating with their users affected conversation flow and various indicators of conversation quality.
How a chatbot should behave in an erroneous situation may not be easily defined, as not all chatbots are deployed to serve the same function, such as a self-service chat or a contact deflection method. Methods that work for one instance of a chatbot, may result in adverse effects on another. Therefore, conversation design, system responses, and overall chatbot behavior should be tailored for each deployment