Use case · Customer support

An assistant that answers from your documents

The answer exists somewhere in a manual, a policy or an old email, and finding it takes longer than asking a colleague. We build an assistant that searches your own material and answers with a link to where it found the answer.

Book a free process analysis

Knowledge assistant

What is the return period for a custom-made item?

Custom-made items cannot be returned unless they arrive damaged or differ from the confirmed order. Damage must be reported within 8 days of delivery.

Returns policy, section 4 Terms of sale, article 12

How you know you have this problem

New colleagues ask the same questions for months
Two or three people are interrupted all day because they know where things are
Documentation exists but nobody can find anything in it
Customers get different answers depending on who replies

What we build

How it works once it is running

01
Gather the material
We connect the folders, wiki pages and help articles you choose, and agree what must stay out.
02
Index it
The content is split and indexed so that the relevant passages can be found for any question, and re-indexed when documents change.
03
Answer with sources
The assistant answers only from the passages it found and shows them. When the material does not cover the question, it says so.
04
Put it where people work
It lives in your chat tool, on an internal page or in the helpdesk, and unanswered questions are collected so you know what to document next.

What you get

An assistant in chat, on a web page or in your helpdesk
Answers with links to the source document
Access rules, so people see only what they may see
A list of questions the material could not answer

What we need from you

The documents, in whatever state they are
A decision on who may see what
A person who knows the material to review test answers
Agreement on where the data may be processed

Tools we typically use: Claude or GPT models, Slack, Teams, Google Drive, SharePoint, Notion, Confluence
Services involved: AI integrations, Custom software

See it built

Illustrative exampleAn online store with a 3-person support teamA support assistant that knows the order and the policyCommon questions get a drafted answer built from order data and the store’s own help pages; an agent approves anything sensitive.AI integrationsBusiness automationOpen example

Questions about this use case

Will it make things up?

The assistant is restricted to the passages retrieved from your documents and instructed to say when they do not contain the answer. That reduces invented answers a great deal but not to zero, which is why every answer shows its source.

Our documentation is out of date. Is it worth doing?

Partly. The assistant will faithfully repeat outdated content, so we usually start with the part you trust and use the unanswered-questions list to decide what to fix first.

Can customers use it, or only staff?

Both are possible. We normally start with staff, where a wrong answer is caught by a person, and open it to customers once the answers have been reviewed for a while.