Start with documents, not AI

You decide this is the quarter you finally build a knowledge base. So you open pricing pages. You book two demos. You start a trial.

Stop there. The tool is not what is missing.

What is missing is that almost nothing your team knows has been written down somewhere another person can find it. Fix that first, and fix it with no AI in the room.

Series: Knowledge That Keeps Up — Part 1 of 3. Write it down first · Make it a governance layer · Keep it current

The bottleneck is not the tool¶

Search tools search what exists. Retrieval systems retrieve what was written. If the answer lives in one person's head, or in a chat thread from March, nothing indexes it.

Buying software first means paying to index an empty shelf.

The uncomfortable part is that this is not a technology gap. It is a writing gap, and no vendor closes it for you.

Start with the questions your team already asks¶

You do not need a content plan. You already have one, spoken out loud, every week.

For two weeks, log every question one team member asks another. Desk interruptions. Direct messages. The "quick question" calls. Answer them exactly as you always would, then write the question down.

Do not write answers yet. Just questions.

At the end you have a list sorted by how often each one repeated. Start with the count. The questions asked five times are your first five documents.

Give every topic one home and one owner¶

One topic. One document. One named person who is responsible for it.

Not a shared folder that everyone can edit and nobody maintains. Not a committee. A name.

That is the core of procedure documentation with a named owner: every document says who answers for whether it is still correct.

This is the same rule that governs automation work, for the same reason. An unowned document decays quietly, and you find out when it is already wrong. No owner, no automation makes the wider case.

What a good enough first pass looks like¶

A 34-person insurance brokerage logged 61 questions in two weeks. Nineteen of them repeated three times or more.

They wrote nineteen documents. Most ran under 400 words. Each one had a title, an owner's name, a last-reviewed date, and the answer. They lived in one folder.

That was the entire first pass. It took nine working days.

Good enough means the answer is correct, someone owns it, and you can tell how old it is. Nothing more.

Why this pays for itself without AI¶

Assume a question gets asked six times a month and costs eleven minutes each time to chase down. That is one hour a month, for one question. You just wrote nineteen of them.

Then there is onboarding, which stops being a shadowing exercise. And the resignation risk, which stops being a single point of failure.

None of that requires a model, a vendor, or a budget line.

What to deliberately leave for later¶

Leave the tooling decision alone. Leave permissions, taxonomy, and migration alone.

Also leave formatting for machines alone. How you structure a document so an AI system can read it cleanly is a real question with a real answer, and it is covered in AI-readable documentation pays off twice. It is the wrong problem to solve on day one.

Write the answers down first. Decide where they live later.

Keep exploring¶

This is Part 1 of Knowledge That Keeps Up, which continues with governing those documents and then keeping them current. If you want a clear read on where your written process baseline actually stands, start the AI Readiness Audit or contact FIT.