An internal knowledge base for repeat questions

An internal knowledge base lets your team find the answer to a repeated question without interrupting the person who always knows it. It works when handbooks and policies sit in one place, can be searched in seconds and every answer shows the document it came from, so nobody has to take it on faith.

Imagine a consultancy of twelve people. Every week someone asks in the team chat how many support hours the monthly plan includes, how an after-hours visit is billed or who approves a discount. The same person usually answers, drops what they were doing, and sometimes replies from memory with a rule that changed months ago. Nobody means any harm: asking is faster than searching.

Why the team keeps asking the same thing

The answer almost always exists. It sits in a PDF attached to an old email, in a shared folder with three versions of the same procedure or in the head of whoever wrote it. Finding it takes longer than messaging a colleague, and even when someone finds it, they cannot tell whether that version still applies.

So a knowledge base that actually cuts repeated questions has to meet two conditions. Finding the answer must take less time than asking for it. And the answer must say where it comes from, so the reader can open the source and check it.

Which documents to upload first

Start from the questions, not from the whole shared drive. For two weeks, write down every question that comes up more than once in the team chat or in the hallway. That list tells you which documents are missing and which exist but nobody finds. In a service business it usually includes:

• the support policy and what each plan includes;
• the procedures for cases outside the rule, such as an urgent visit or a change of scope;
• the steps to set up a new client;
• who approves what: discounts, expenses, payment exceptions.

In AgentticCRM you upload PDF, Word (.docx), Markdown or plain text, up to 25 MB per file, and sort them into collections such as Internal procedures or Support policies. A PDF scanned as an image has no text to read, so upload a version that contains text. The full detail is on the knowledge base page.

Search for the exact passage or ask for a written answer

There are two ways to look things up. Library search shows the matching passages, highlighted, and can be narrowed to one collection. It searches by meaning, so it finds the passage even when your words differ from the document's, and even when you ask in Spanish about a document written in English. It works without connecting an AI provider, and search does not send your documents to any outside service.

Knowledge Chat is for when you would rather get a written answer. You ask in your own words, it answers from your team's documents and lists the sources it used, with a link to each one. To write the answer, it sends the passages it found to the AI provider your organization connected. The chat only looks at documents; a question about a client, an invoice or a project is a job for the Assistant.

The same library feeds other work. A ticket's analysis draws on your procedures and stays as an internal note in support tickets, and the WhatsApp inbox prepares reply drafts from the collection you pick; your team decides whether to send them.

When the question is who handles it

Many repeated questions are looking for a person rather than a rule: who approves the discount, who a complaint gets escalated to. In Positions you describe each role in your organization with its responsibilities, decision rights, escalation path and the people who hold it.

When someone searches the library for a topic that appears in a position, the results include a "Who does this" section with the position's title, its org chart unit and the people who hold it, or "No one assigned yet". That last case is useful too, because it exposes a responsibility nobody owns.

Keeping it trustworthy so it does not become another dead folder

A knowledge base gets abandoned the day someone finds an old rule in it. To prevent that, AgentticCRM shows the status of every document: indexed or failed, and the reason it failed. You can open a document and check the text that was read and how many searchable passages it has before you rely on it. When a policy no longer applies, you archive it: it stays on the list but stops appearing in searches and answers.

The chat helps you see what is missing. When your documents don't cover a question, it tells you so and suggests uploading a related document. If someone reviews those unanswered questions every so often, they already have the list of what to write next.

Permissions complete the picture. Viewing, uploading, editing and deleting documents are separate permissions, and so is using the chat; by default a Member can view, upload and edit, but not delete. Giving each collection one person who keeps it current works well.

If you are thinking about assembling these pieces yourself with the help of AI, first look at what it costs to maintain a homemade CRM. And to see how your ChatGPT or Claude saves what it researches into the CRM, read the article on your CRM as your company's memory.

Frequently asked questions

Can my clients look things up in the knowledge base?

No. It is a tool for your team and your clients don't see it. What reaches them are better-prepared replies, such as the WhatsApp drafts your team reviews before sending.

Do I need to connect an AI provider to use it?

Not for search: library search works from the first document. For the chat to write answers, your organization connects its provider in Settings, under Assistant model.

What do I do with a policy that changed?

Upload the new version and archive the old one. The archived one stays on the list for reference, but stops appearing in searches and in chat answers.