In this article

Tuesday, second-line support at a telecoms operator. A customer has no internet after swapping her router, and Tom, the agent, has a feeling he has seen this before. He has: six months ago. Someone solved it and even wrote it up, but nobody remembers where. A good customer support knowledge base exists for exactly these moments: procedures and solved cases sit in one place, and you can find them before the customer runs out of patience.

In many teams this knowledge is scattered. Procedures live in one folder, service terms in another, and solutions in e-mail threads and in the heads of the most experienced people. It is one of the posts in our series "The AI-assisted archive in practice". How the assistant connects to the archive and how it is secured is covered in our post on AI document search. Here we show how to build a support team's knowledge base on that foundation.

Procedures and cases in one archive

A telecoms company's archive usually holds subscriber contracts, amendments, terms and conditions and complaints. Our page on solutions for telecom and media tells you more. The support team's knowledge can live right next to them, in CONTMAN Archive. Your company defines the categories and fields, and a simple layout works well for support:

  • Service procedures, with the fields "Service" and "Valid from". You add a new edition of a procedure as a new version of the file, and the earlier ones stay in the document as older versions. There is more on editions of one procedure in our post on employee onboarding.
  • Service terms, described with the same "Service" field, so that the procedure and the terms of one service can be found with a single filter.
  • Solved tickets, with the fields "Ticket no.", "Service" and "Problem type". A list field can hold up to 50 values, so the list of problem types stays compact. To each document you attach a short write-up: symptom, cause, solution.

It is worth marking the "Ticket no." field as one that links documents. The solution write-up, the customer's e-mail and the reply letter then become related, and the assistant can fetch them together. A unique key on the ticket number in that category makes sure one ticket does not end up with two solution write-ups. You can read more about unique keys in our post on keeping documents in order.

The archive extracts text from PDF, DOCX, XLSX, PPTX, TXT, CSV and MD files and from e-mail messages. A scan turns into text only with OCR switched on. It does not read e-mail attachments, so an important attachment is best saved as a separate document with the same ticket number.

The customer's words in the solution write-up

Search in the archive is full-text. It finds documents that contain the words you search for, not documents that say the same thing in other words. So it pays to write solutions in the language customers and devices use: the exact error message, the name of the service and the model of the equipment. In the application itself, an agent can also search for the exact phrase, which narrows error messages down to their precise wording.

Questions an agent can ask the assistant

Tom connects the AI assistant his company uses to the archive. During his shift he might ask, for example:

  • "Have we had a ticket where a phone would not register on the network after a SIM swap? How was it solved?"
  • "Find the current procedure for billing complaints and summarise the steps the agent takes."
  • "Show me all the documents related to ticket ZG-2026-0412."
  • "What do the terms of the Example Home Internet service say about suspending the service?"
  • "Which solved tickets for the Example TV service have no problem type filled in?"

The last question is more for the team lead who keeps the knowledge base tidy. The assistant has the tools for each of these: it searches for a phrase in document descriptions and content, narrows results by fields and by empty values, reads the text of the current file version and gathers the documents with the same ticket number. Tom, the ticket number and the service names are made up for this example.

The assistant's answer is a starting point, not a ready-made script for the customer. Before Tom passes a solution on, he opens the document it comes from and checks that the case described really matches her situation.

For the same kind of questions on a leasing company's hotline, see our post on the lease agreement archive.

Control over what the assistant reads and changes

A support knowledge base sits right next to customer data, so the boundaries have to be clear:

  • Agents see only their own categories. Permissions are granted in roles, separately for each category, and the assistant sees only what the person who connected it is allowed to see. The support team can read procedures and solutions, and subscriber contracts only if their role allows it. We cover roles and permissions in more detail in our post on secure document storage.
  • A customer's e-mail is data, not an instruction. The archive also holds messages from customers, and what they say varies. The description of every archive tool reminds the assistant that document content is data, not instructions to follow.
  • Every read leaves a trace. Whenever the assistant opens a document, its file or its text, the document's history shows it, marked as done by an AI assistant.
  • Changes only with consent. Reading is the only mandatory area. The assistant can change the archive only in the areas the agent has agreed to and an administrator has granted in the agent's role, and only as far as the agent's category permissions allow. Every change to a document is recorded in its history, noting that an AI assistant made it.

Customer support knowledge base: step by step

  1. Start with the procedures you use every day. Check that the current edition of each one is the latest file version in its document.
  2. Create a "Solved tickets" category. Decide on the fields, mark "Ticket no." as a field that links documents, and make the most important fields required. Required fields apply from the moment you switch them on, with no back-filling.
  3. Agree on a template for solution write-ups. Symptom, cause and steps, in the customer's words. A short write-up with a fixed structure is easier to find and easier to summarise.
  4. Switch on content processing. The assistant searches file content only in categories where it is switched on.
  5. Grant permissions. An administrator grants agents permission to use the assistant in their roles and, in the knowledge base categories, document read and file download and preview, without which the assistant cannot read the content. Each agent connects the assistant by signing in with their own account.

The knowledge base grows with every case you write up, and the assistant helps people use it while the customer is waiting for an answer. Book a demo, and we will show you how it works with procedures and tickets similar to yours.