Chatbot
A chatbot is a software application designed to simulate human conversation through text or voice, ranging from simple rule-based systems to sophisticated AI assistants powered by large language models.
A chatbot (also: conversational agent, virtual assistant) is software capable of engaging in text or speech dialogue with humans. The chatbot category spans a vast spectrum — from simple FAQ bots that match keywords to rules, to sophisticated AI assistants powered by large language models that can reason, write code, and engage in extended multi-turn dialogue.
Types of Chatbots
Rule-based Chatbots
Use decision trees or pattern-matching to respond. Deterministic, predictable, and easy to audit. Suitable for well-defined workflows (FAQ, appointment booking, form filling) but brittle when users deviate from expected inputs.
Retrieval-based Chatbots
Select responses from a curated database based on input similarity. More flexible than rule-based systems, but limited to pre-authored responses. Used in early customer service deployments.
Generative Chatbots (LLM-powered)
Generate responses dynamically using large language models. Can handle open-ended questions, nuanced context, and multi-turn reasoning. Modern examples: ChatGPT, Claude, Gemini, Microsoft Copilot. Risks include hallucination and inconsistency.
Hybrid Architectures
Production systems often combine rule-based intent detection (for routing and safety guardrails) with LLM generation (for flexible response creation). Retrieval-Augmented Generation (RAG) is used to ground chatbot responses in company-specific knowledge bases.
Deployment Channels
- Web widget — embedded chat window on websites
- WhatsApp Business API — dominant in Malaysia for customer communication
- Telegram bots — popular for developer communities and fintech
- Microsoft Teams / Slack — enterprise internal use
- Mobile app — native SDK integration
- Voice — phone IVR, smart speakers
Key Metrics
| Metric | Description | |--------|-------------| | Containment rate | % of conversations resolved without human escalation | | CSAT | Customer satisfaction score post-conversation | | First response time | Time from user message to first bot reply | | Intent recognition accuracy | % of user intents correctly classified | | Hallucination rate | % of factually incorrect responses (LLM bots) |
References
- Weizenbaum, J. (1966). "ELIZA — A Computer Program for the Study of Natural Language Communication Between Man and Machine." CACM 9(1), 36–45.
- MDEC (2023). Conversational AI Adoption in Malaysian Enterprises. Malaysia Digital Economy Corporation.