Skip to content

An internal knowledge assistant that answers staff from your own documents.

An internal knowledge assistant answers staff questions using your own policies, procedures and documents, and shows which document each answer came from. It only sees what each person is allowed to see, and says so when the documents do not cover a question rather than guessing.

What questions does it answer?

Every business has questions that get asked again and again: how to process a refund, what the policy is on annual leave carry-over, which form a new client signs, where the latest price list is. The answers exist somewhere, in a handbook, a shared drive or one experienced person’s memory. An internal assistant lets staff ask in plain words and get the answer from the right document.

It is most useful for new starters, for staff spread across sites and for procedures that are done rarely enough that nobody remembers the steps. It is not a replacement for the people who own the policies; it saves them from answering the same question twice.

For example, in a hypothetical practice with three sites, a receptionist asks how to handle a patient who wants a copy of their records. The assistant answers from the subject access procedure, links to the form and names the person who signs off requests. It does not interpret the law; it points to what your own procedure says.

Is internal knowledge assistant a job for an agent?

Fit check

Tick what is true of your internal knowledge assistant work.

Tick the statements that apply and the suggestion appears here: an AI agent, ordinary automation, or a person with software support.

Which sources should it use, and who keeps them current?

The assistant is only as good as the documents behind it. Old versions, duplicates and drafts in the same folder will produce confident answers from the wrong file. Before building, decide:

Decisions about the knowledge sources
DecisionWhat to settle
Which sourcesA named set of folders, wiki spaces or pages, not “the whole drive”.
Which version winsHow current documents are marked, and what happens to superseded ones.
Who owns each areaA named person for HR, finance, operations and so on, who approves changes.
How often it refreshesWhether new and edited documents are picked up hourly, daily or on publish.
What is excludedPersonnel files, legal correspondence, anything in draft.

Sources can live in Google Workspace, Microsoft 365 or Notion. The assistant reads through each tool’s API with an account limited to the agreed folders.

Why every answer should cite its source

An answer with a link to the paragraph it came from can be checked in seconds. An answer without one has to be taken on trust, and staff quickly learn whether that trust is deserved. Citations also show you which documents are used most and which are causing confusion.

How much time could internal knowledge assistant give back?

Estimate

Put your own numbers in.

The starting values are an illustrative example, not a benchmark or a neu figure. Use your own volume, the time each one takes today and the share you expect the system to handle without a person.

Hours back each month20
Value of that time€600
Over a year€7,200

Time saved is only part of the case: response times, missed work and errors matter too. The ROI guide covers how to set a baseline, and the cost guide what the build and running costs include.

How do permissions work?

The safest design respects the permissions you already have: a person can only get answers from documents they could open themselves. That means the assistant checks who is asking and filters sources before it answers. Building one assistant with access to everything and hoping it will not repeat a salary figure is not acceptable.

Also decide what is logged. Questions can reveal personal matters, such as someone asking about sick leave or a grievance process. Keep logs only as long as you need them and limit who reads them. Our AI agent security guide covers access, logging and prompt injection in more detail.

Where staff ask their questions

Put the assistant where people already work: a Slack or Teams channel for office teams, or WhatsApp for staff on the road or the shop floor. A separate website that nobody bookmarks will not be used.

Over time the assistant can do more than answer, such as filling in a leave request or logging an incident, which turns it into an AI operations assistant. Start with answers first; actions need more controls. Where staff ask questions also affects permissions: a WhatsApp number used by many people needs a way to confirm who is asking before it answers anything restricted.

Building a first version

  1. Collect the questions staff actually ask, from emails, chats and the people who answer them.
  2. Choose one area, such as HR policies or operating procedures, and tidy its documents first.
  3. Test the assistant against the real questions and check every answer and citation.
  4. Launch to a small group, with an easy way to mark an answer as wrong.
  5. Review flagged answers weekly with the document owners, and fix the document, not just the assistant.

Measure questions answered without a follow-up, answers marked wrong and repeat questions to managers. If your documents are short and well organised, a good search tool may be all you need; see chatbot vs AI agent for the difference in effort.

Common questions

Is our data used to train the AI model?

That depends on the provider and the terms of the account used. Business API terms from major model providers generally state that customer data is not used for training by default, but check the current terms for the specific service and plan, and record the answer in your data-protection documentation.

Can it read scanned PDFs and old Word files?

Usually, but scans need text recognition first and the quality depends on the scan. Old files often hold outdated policies, so this is a good moment to decide whether they should be included at all. Tables and forms are harder to read reliably than plain text and are worth testing.

How is this different from searching the shared drive?

Search returns files; the assistant returns an answer with the relevant passage, even when the question uses different words from the document. It can also combine information from two documents. Search remains useful, and the assistant’s citations link back to the files so people can read the whole thing.

What happens when a policy changes?

The owner updates the document in its usual place and the assistant picks up the new version on its next refresh. The old version should be removed or clearly archived, or the assistant may still find it. That is why each area needs an owner and a clear rule about superseded files.

Bring us the task.

Tell us what happens today, which tools are involved and where it goes wrong. On a free 30-minute review, we’ll work out whether an agent, an automation or neither is the right fix, and what a sensible first release would be.

Start a project.

Two people on a sofa with laptops, planning a project beside a whiteboard of sticky notes

Book a call or send us a message. We read everything.

tash@neu.ie
Working with teams in Ireland and worldwide.

Get in touch

Free 30-minute AI review.