FAQ answers from your own knowledge base:
no guessing, no made-up policy
The same twenty questions get asked a hundred times a week, and the answer depends on who is typing. An agent answers from your actual documents every time, and says so honestly when a question is not covered.
The process today
The same handful of questions repeat constantly, pricing, hours, policy on a specific edge case, whether a product does a specific thing, and every one gets typed out fresh by whoever is on shift, pulled from memory rather than a document, which means the answer quality depends entirely on who happens to be answering and how well they remember the current policy.
The second cost is drift. Policies change, prices update, a product gets a new feature, and the knowledge in a support team’s collective memory updates unevenly: someone who has been doing the job for a year answers with last year’s policy because nobody told them it changed, and a customer gets a confidently wrong answer that then needs correcting, which damages trust more than a slow answer would have.
The third cost is coverage. Questions arrive outside working hours, and a question with a documented, unambiguous answer waits anyway, simply because the document lives in a shared drive that nobody checks outside working hours either.
What the agent does
The agent answers from a knowledge base built out of your actual documents, pricing sheets, policies and FAQ pages, using retrieval so every answer is grounded in something real rather than generated freely. When a question matches something in the knowledge base, it answers immediately, any hour, in the channel the customer already used: Telegram, WhatsApp, your site’s chat, wherever.
When a question does not match anything in the knowledge base, the agent says so honestly and hands off to a person with the question attached, instead of producing a plausible-sounding guess. This single rule, grounded answers or an honest hand-off, is the difference between a knowledge base agent and a chatbot that occasionally invents a policy that does not exist.
Your team edits the knowledge base through an admin panel, no developer needed, so a policy change or a new product feature is reflected in the next answer within minutes of being added. The agent also logs every question asked, which becomes the most useful input for finding gaps: topics customers ask about that your documentation does not actually cover yet.
The same knowledge base can serve more than one surface at once: a chat widget on the site, a Telegram bot, and a simple search box on a help center page, all answering from the same source so an update made once applies everywhere instead of needing to be copied into three separate places by hand.
What stays with humans
Anything the knowledge base does not cover, and anything that reads as a complaint, a dispute, or a request rather than a question, hands off to a person. The knowledge base content itself, what counts as the correct answer, is written and maintained by your team; the agent retrieves and phrases it, it does not decide policy.
Guards
Answers are grounded in retrieval against your actual documents, with a hand-off default when nothing matches closely enough, rather than a model free to answer from general knowledge. Every conversation is logged, useful both for quality review and for spotting documentation gaps. A kill switch and manual review period precede full launch.
Price and timeline
| Option | Price | What it covers | Timeline |
|---|---|---|---|
| Single automation | from $500 | One knowledge base, one channel, admin panel | 3 to 7 days |
| Department package | from $2,500 | FAQ answering plus multilingual support and ticket triage | 2 to 4 weeks |
Running cost is typically $20 to $80 a month in model usage depending on question volume.
Related
Pair this with multilingual support replies if your customers write in more than one language, and with support ticket triage so questions the knowledge base cannot answer land with the right person automatically. See warranty and complaint intake for the structured version of a question that is really a complaint. Full package details are on the AI agents service page; for a real knowledge base editable without a developer, see the visa centre AI support bots case study, and for a wider multi-channel agent built on the same grounding principle, the seven-channel AI sales agent case study.
Want your actual documents turned into instant, honest answers? Get in touch and send us what you have.
Tired of doing this by hand? We can take the whole routine off your team, not just this step: Routine takeover, from $400 →
FAQ
How much does FAQ automation cost?
From $500 for one knowledge base and one channel, live in 3 to 7 days. A larger document set, multiple languages, or several channels usually run $900 to $2,000.
How long before it is live?
3 to 7 days once we have your documents, FAQ pages, policies and pricing sheets to build the knowledge base from.
Which tools does it connect to?
Telegram, WhatsApp, your site's chat widget, and a RAG layer over whatever documents you give us: PDFs, Notion, Google Docs, a help center export. An admin panel lets your team edit answers directly.
What if the AI does not know the answer?
It says so and hands off to a person, with the question attached, rather than inventing a plausible-sounding answer. This is enforced by grounding every answer in your actual documents, not by hoping the model stays honest.
How do we keep the knowledge base up to date?
Through the admin panel, no developer needed: your team edits or adds an answer directly, and it is live for the next question. The log of asked questions also shows which topics need a documented answer that does not exist yet.