AI chatbot
What is AI chatbot?
A conversational interface that answers in natural language, and in business usually grounds those answers in your own content through retrieval-augmented generation rather than the model's general training. That grounding is what keeps it accurate and lets it cite sources. It is the most widely deployed enterprise AI pattern, and the easiest to get wrong when the underlying data is poor.
Why it matters
The demo is deceptively easy; the deployment is where they fail. A chatbot that answers well in a scripted walkthrough meets three problems in production the demo never showed. It gets asked things outside its remit and needs to decline rather than improvise. It hits questions it cannot answer confidently and needs to hand to a person instead of guessing. And it gets judged, sooner or later, on a single confident wrong answer to a real customer. Grounding the answers in trusted content handles accuracy, but scope, escalation and honest uncertainty are design decisions, not defaults. The measure that matters is not how many queries it deflects from the queue but how many it actually resolves, because a deflected question that leaves the customer angry has not been handled, only hidden.
In practice
A support assistant answers from the approved knowledge base, refuses politely when asked something off-topic, and routes anything it is unsure about to a human with the conversation attached. The team tracks resolved issues, not deflected tickets, and watches the escalation rate as a health signal. A bot that looked efficient by sending frustrated customers in circles would have scored well on exactly the wrong metric.
