chatbot-flow-design

What it does

Describes how to design chatbot conversational flows with a defined set of intents, grounding in a knowledge base, and structured fallback and escalation patterns. It positions three archetypes (scripted-bot, hallucinating-bot, structured-guided-conversation) and provides guidance on when to use a design approach, how to create intent architectures (named intents, hierarchies, boundaries), grounding practices, branching logic, fallbacks, and escalation to human agents. It also covers when to deploy a chatbot and how to measure performance through conversation analytics and common failure modes. The skill emphasizes avoiding ungrounded, over-generalized, or misrouted conversations and aims for honest scope and maintainable design.

How it works

Outlines concrete design components and patterns:

• Intent architecture: define a bounded set of explicit intents, potentially with hierarchies and clear boundaries; ensure 70-90 percent coverage with intents and reserve rest for fallback.
• Knowledge-base grounding: require responses to be grounded to a knowledge base, using retrieval-augmented generation and a source-of-truth design; cite sources when answering.
• Branching and conditional logic: enable intent-driven, context-driven, and user-attribute branching with multi-turn depth, keeping branching depth manageable (3-5 turns typical).
• Fallback patterns: implement clarifying questions, suggested intents, resource handoffs, and human escalation with multiple fallback layers and honesty about limitations.
• Escalation to human: escalate based on user requests, out-of-scope intents, repeated unclear exchanges, sentiment, or high-stakes topics, passing conversation history and recognized intent.
• Conversation analytics: track intent recognition rate, resolution rate per intent, fallback rate, escalation rate, and user satisfaction to guide improvements.
• Common failure modes: recognize hallmarks such as ungrounded hallucinations, rigid scripted behavior, or misrouted flows, and address grounding and intent sets accordingly.

When to use it

Use this design approach when scoping a chatbot for the first time, auditing a chatbot that hallucinates or fails edge cases, designing the intent architecture and fallback patterns, or deciding when to escalate to humans.

What it can touch

The skill references and relies on knowledge-base grounding and escalation patterns, and specifies using retrieval-augmented grounding, source-of-truth design, and citation discipline as part of the bot's behavior. It involves designing intent architecture, branching logic, fallbacks, and escalation pathways, and measuring performance with analytics.

Caveats

Outlines limitations and risks, including the need for maintenance of intents and knowledge bases, the potential for drift and stale knowledge, and the trade-offs between scripted, hallucinating, and structured-guided approaches. It notes that the structured-guided-conversation approach requires upfront design effort and ongoing maintenance. All statements describe design principles, patterns, and measurement practices without promising outcomes beyond what is stated.