A business chatbot searches the company’s own documents for an answer before writing anything, then phrases a reply from what it found. That is why the answer reflects your pricing and terms rather than the model’s general knowledge.
Step one is retrieval: the system finds the relevant passages in your documents. Step two is phrasing: the model writes an answer from those passages. Quality depends almost entirely on step one. If a document is vague or out of date, the model cannot rescue it - it only relays what it found.
The most laborious part of a rollout is not technical. It is getting the pricing, terms and instructions into order - often for the first time in the company’s history. That has value on its own. Plenty of teams discover that staff had been giving customers different answers because no single source existed.
A chatbot handles repeated questions well. It handles negotiation, exceptions and context-dependent judgement badly. So a properly built system knows its own boundary and hands over. A bot that answers everything is dangerous - not because it is wrong, but because it is confidently wrong.
Every call to the model costs money. That makes a chatbot a supported service rather than a one-off build, with an ongoing cost and ongoing supervision. The cost is predictable and tracks conversation volume. If someone offers a chatbot for a single fixed fee, ask who pays the model in month two.
A chatbot is only as good as your documents. Before considering one, check whether a single place exists where the prices and terms are written down correctly.
8 Cristoforo Castelli St, Tbilisi 0159, Georgia · +995 596 16 66 88 · info@zenlumex.com