A chatbot platform provides the core capabilities for building conversational applications, including natural language understanding and dialogue management. It turns user messages into interpreted requests, routes them through conversational behavior, and supports responses or connected services. section 1.4 section 1.1.3
A useful mental model is a front door, interpreter, and coordinator: the platform starts conversations, identifies what users mean, manages dialogue, and handles cases it cannot match. Poor fallback behavior leaves users repeating themselves; a short retry followed by human escalation keeps them from getting stuck. section 2.1.1 section 1.1.3
Core platform responsibilities include:
- Recognize intent from varied natural-language utterances, rather than relying only on exact words. section 1.1.3
- Provide default conversation behaviors such as greeting users, detecting trigger words, and responding when no action matches. section 2.1.1
- Connect conversational flows to secure APIs that collect required inputs, return structured results, and expose error conditions. section 1.1.3
- Support low-code or no-code development while still requiring representative training data and testing. section 1.1.3
Cake Bot’s first release
A cake shop begins with frequently asked questions and creates Cake Bot in a conversational AI platform. The team changes the generic greeting to “Welcome to Cake Bot. How can I help you?” They also reduce the unmatched-input retry threshold from three attempts to one, after which the assistant offers a human agent. This small configuration change improves the path for users whose requests are not yet understood. section 2.0 section 2.1.1
Table 2.1 Default capabilities in a new assistant
|
Capability
|
Executed when
|
|---|---|
|
Greet customer
|
The assistant is first opened or engaged with. Opening the assistant starts a conversation.
|
|
No action matches
|
No action can be matched to the user’s message (the message is not understood). Other platforms may call this the “fallback intent.”
|
|
Trigger word detected
|
Keywords like profanity are detected.
|
|
Fallback
|
The user needs to leave the chatbot.
|
Use the default-capabilities table to inspect the standard events that start, route, or end a chatbot interaction. section 2.1.1
