Chatbots and Conversational AI — From FAQ Bot to Intelligent Assistant

80% of customer inquiries are repetitive. A chatbot can resolve them instantly, 24/7, with no wait time. But a bad chatbot is worse than none.

Chatbot Architecture¶

• NLU (Natural Language Understanding) — intent and entity recognition
• Dialog Management — conversation flow control, state
• Response Generation — templates or generated responses
• Backend Integration — CRM, FAQ database, ticketing

Rasa vs Azure Bot Service¶

Rasa: open-source, on-premise, full control over models. Azure Bot Service: managed, integration with Teams/webchat, LUIS for NLU. Rasa for sensitive data (on-premise), Azure for rapid MVP.

Key Metrics¶

Containment rate (% of conversations resolved without human escalation): target 70%, reality after 3 months: 62%. CSAT (satisfaction): 4.1/5. The key: quality fallback to a human operator, not an endless “I don’t understand” loop.

Chatbot = First Line of Support¶

Handles routine inquiries, gathers context for operators, available 24/7. But it must know when to say “I’m transferring you to a colleague.”