Design Recommendations for Building an Efficient Chatbot
Recently, I built a chatbot for an e-commerce website.The main goal of this chatbot was to facilitate customer purchase in minimum conversational steps. This required building a fast product recommenders, intuitive cart/checkout experience and much more.
Based on my learnings while building a e-commerce chatbot , here are few design recommendations to consider if you decide to build a chatbot for yourself.
-
Enable user to reach target state in minimum number of steps
-
Support for platforms and devices used by the majority of user base
-
Use menus over too broad and generic questions to simplify your NLP logic
-
Design the bot to allow user to change the course of conversation any time during the conversation
-
Implement global message handlers to support common user actions like help , start over, cancel, skip etc
-
Acknowledge the user input immediately, even if your bot is going to take some time for processing
-
Priorities ‘value over volume’ in chatbot conversation. Ask specific questions to reduce the complexity of the response handling logic
-
If you providing multiple search results or probable answers for the user query, use a ‘confidence score’ to indicate higher probability of correctness of the answer or recommendation
-
Use ‘Back channel mechanism’ to facilitate interaction between bot and the client web page.
-
Use a root dialog to initiate other dialogs during the conversation. The root dialog should always be running.