Skip to main content
Tillr Systems← Back to Site

Knowledge in one person's head

When everything the company knows lives in one person's head.

It works right up until that person is on holiday, or busy, or leaves. The knowledge is not missing. It is sitting in an inbox, a group thread, and five years of replies nobody can search.

A searchable memory built from what your company has already written down, wherever that currently is. It gives you:

  • Plain-English questions answered from your own documents, not from a general model
  • Every answer citing the file, message or thread it came from
  • A system built to say it does not know, rather than guess

The knowledge stays yours, and stops depending on who is available.

You'll recognise this if

  • The same question reaches the same person several times a month, and they answer it from memory each time
  • One person's inbox is, in practice, the company archive
  • A new hire takes months to become useful because the answers are not anywhere they can look
  • You have thought about what happens if that person leaves, and stopped thinking about it
  • The documentation project has been started twice and abandoned twice
  • Finding a past decision means asking someone who was there rather than looking it up

How we'd approach it

  1. 1

    We find where the knowledge already is

    Usually four or five places: an inbox, a shared drive, a group or community thread, a folder of manuals, and one person's memory. We index what exists rather than asking anyone to write documentation first, because the documentation project is the thing that never gets finished.

  2. 2

    We make whose answer it is part of the system

    In most archives the expert's answer sits next to three guesses from other people. We make authorship part of retrieval so the authoritative answers are the ones that surface, and the guesses stay out of the results.

  3. 3

    We wire it to how people actually ask

    A search box nobody opens is worth nothing, so it goes where the questions already get asked. Answers cite their source document, which is what makes someone trust the second answer after they have checked the first.

What changes

  • The answer is in someone's inbox

    The answer is one search away, with its source attached

  • New hires ask a person and wait

    New hires ask the system and check with a person

  • Five years of replies are effectively unreachable

    Five years of replies are the most valuable asset you own

  • One person leaving is a business risk

    One person leaving is a hiring problem

We've done this before

Online education & technical training

Five years of answers, finally searchable

An instructor's entire body of knowledge was spread across years of email replies, a course community, and social groups, alongside a library of about 4,700 manufacturer service manuals nobody could search by model number. We built one index across all of it, with her answers treated as the source of truth.

Read the case study

Common questions

How is this different from asking ChatGPT?
A general assistant knows what is on the public internet and nothing about your company. Ask it what you quoted a particular customer for phase two and it will either refuse or invent something plausible. This searches your documents only, cites the one it used, and is built to say it does not know when the answer is not in there.
Our files are a mess. Do we need to organise them first?
No, and organising first is usually how these projects die. Messy is the normal input. Folder structure matters far less than you would expect because the system searches contents rather than filenames, and the cleanup that genuinely helps becomes obvious once you can see what is in there.
What stops it giving a confidently wrong answer?
Two things. It answers only from documents it can point at, so an answer without a source does not get shown. And it is configured to return nothing rather than reach, which feels worse in a demonstration and is what makes it usable in front of a customer.
Does our information get used to train an AI model?
No. We configure the providers so your content is not used for training, and we tell you exactly which providers are involved before anything is connected. The index lives in your own account, and you can export or delete the whole thing whenever you want.

This is one way into the custom business software we build. Most projects start with whichever piece is costing the most and grow into the others, so it is worth seeing the whole picture before deciding where to begin.

Other places people start

Different ways into the same system. Most projects begin with whichever of these is costing the most right now.

Start with thirty minutes.

Tell us what's slowing you down and we'll tell you what we'd build. How much scoping it needs before a quote depends on the job, and we'll be straight with you about that.

Or see how pricing works →