The agent answers from what you've taught it. This article covers where that knowledge comes from, how it's structured, and how it improves.
The sources
Your knowledge base. The primary source. Help articles, product documentation, policies, and FAQs, structured so the agent can retrieve the right passage for the right question. It serves your customers who search it directly, and your agent, which reads it on every enquiry.
Your website content. Published pages, product descriptions, and policy pages. This is what a first version of the agent trains on, which is why it can be answering customers within days.
Your past conversations. The most useful source you have. Ticket history shows what customers actually ask, in the words they use, and what your team actually answers. It reveals the questions nobody thought to document.
Your live data. Order records, account details, billing status. Not knowledge exactly — the agent queries these at the moment of the conversation rather than learning them. It's what lets a policy answer become a specific one.
If You Don't Have A Knowledge Base
Common. We write it.
Discovery surfaces most of it: the kickoff call, your team's canned responses, whatever lives in someone's head or a shared doc. Where a source exists we structure it. Where none does, we draft and you approve.
You end up with a customer-facing help center you didn't have before, which is a second asset out of the same work.
How It Improves
When | What happens |
Setup | Trained on your content, tested against your enquiry history, tuned before launch |
Launch | Monitored closely, corrections applied quickly |
Ongoing | Conversations reviewed, missed answers corrected, knowledge base expanded where gaps appear |
The pattern is the same throughout. Real conversations reveal what the agent doesn't know. Missing knowledge gets written and added. Coverage widens.
💬 Still Have Questions?
Need more help? Our team responds to every message within 24 hours. Start a conversation and we'll point you in the right direction.