Getting Your Business Knowledge AI-Ready: What to Document First

Here’s the thing nobody mentions in the AI sales pitch: the assistant doesn’t know anything about your business. It knows language. Everything specific — your prices, your policy on late cancellations, whether you deliver to the next town — has to come from documents you provide. Give it nothing and it will either say “I don’t know” or, worse, guess.

So the honest first step isn’t buying software. It’s writing things down. The good news: it’s a smaller job than it sounds, and it pays off even if you never deploy an AI assistant at all — the same pages train new staff and answer customer emails.

Start with the questions you already answer

Don’t try to document your whole business. Spend twenty minutes listing the questions you and your team actually get — from your inbox, your phone notes, your DMs. In most small businesses, a surprisingly short list covers the overwhelming majority of real enquiries. That list is your documentation plan. Everything else can wait.

The seven documents to write first

1. Your top customer FAQs. Twenty to thirty questions with real answers, in the words customers actually use. Not marketing copy — answers.

2. Pricing and what’s included. What things cost, what’s bundled, what’s extra, and when you quote instead. Vague pricing is the single biggest source of wrong AI answers.

3. Hours, locations and contact routes. Including holiday closures and which channel gets a human fastest.

4. How to book, order or buy. The actual steps, including deposits, lead times and what happens after payment.

5. Your policies. Returns, cancellations, warranties, deposits, guarantees. Write the exceptions too — the “unless” cases are exactly where assistants get people into trouble.

6. Your product or service catalogue. What you offer, who each thing is for, and what it isn’t suitable for. That last part prevents a lot of mis-selling.

7. The internal procedures your team keeps asking about. How to process a refund, how to handle a complaint, how to open or close. Guidance on retrieval-based AI suggests documented internal procedures like these are among the highest-return items to capture, because they turn repeated interruptions into self-service.

Write it the way AI reads it

This is where most people go wrong, and it costs nothing to get right. AI assistants don’t read your handbook cover to cover — they retrieve fragments and answer from those. Industry guidance on preparing documentation for retrieval-augmented systems lands on the same handful of rules:

One topic per section. Don’t combine deposits, cancellations and gift-card rules in one block of text. Split them, so the right piece can be found on its own.

Make each section self-contained. Say what it covers, who it applies to and any limits inside the section. A paragraph that only makes sense after reading the previous three pages will be retrieved alone — and misread.

Use clear, literal headings. “Cancellation policy for appointments” beats “A note on flexibility.”

Use your customers’ vocabulary. If they say “delivery” and you say “fulfilment logistics”, the match gets weaker. Write both if you must.

Document the exceptions. Happy-path-only documentation is the most common failure mode: the assistant is confident right up until reality gets complicated.

Plain text, a shared doc, or simple web pages are all fine. Formatting sophistication doesn’t help; clarity does.

Decide what stays out

Being AI-ready is also about what you don’t feed it. Staff records, customer personal data, supplier contracts, anything commercially sensitive — leave it out of a customer-facing assistant’s knowledge unless there’s a specific reason and proper controls. A useful test: if you wouldn’t email it to a customer who asked nicely, don’t put it in the knowledge base. We wrote more on the guardrails in keeping your business data safe when you adopt AI.

Give it an owner, or it rots

Documentation goes stale faster than anything else in a small business. Prices change, a service gets dropped, opening hours shift for summer — and now your assistant is confidently wrong in public. Assign one person, put a fifteen-minute review in the calendar monthly, and add a rule to your team: if you answer a question that isn’t in the docs, add it. That one habit compounds. Where the answer lives in a live system rather than a document — stock levels, order status, next available appointment — the better fix is connecting the assistant to that system through automated workflows, so it reads the truth instead of a snapshot.

What “done” looks like

You’re ready when someone new to your business could answer your ten most common questions correctly using only your documents. That’s the bar — and it’s a bar you can clear in an afternoon, not a quarter. From there, an AI assistant trained on your material can handle the repetitive tier of enquiries while your team takes the ones that need judgement, the same pattern we described in cutting response times without hiring.

And if you go through this exercise and find your knowledge is already tidy, well-structured and current? Then you’re further ahead than most, and you should be talking about what to automate rather than what to write. We’ll tell you that plainly — it’s a better use of your money than a project you don’t need.