Sample Business
Why this works, and what you are actually building.
Most people meet AI the same way. They open a chat window, describe their business in a paragraph, ask for something useful, and get back something that could have been written about any company in their industry. The output is fluent and generic, which is the worst combination, because generic writing that reads well is harder to spot than generic writing that reads badly.
The problem is not the model. The problem is that you gave it a paragraph.
This document describes the alternative: building a structured, permanent representation of a business that an AI reads before it does anything. Do it once, properly, and every task after it starts from knowledge instead of from a guess.
Not a folder of documents. A memory.
The distinction matters. A folder of documents is passive, and its natural end state is a graveyard nobody opens. A memory is something a system consults before it speaks. The structure in this pack is designed for the second thing: small focused files, each answering one question, each named so a machine can find it and a human can maintain it.
The test of whether you have built one is simple, and it comes at the end of this process. You ask the AI five questions only someone who works there could answer, and you see whether it can.
A business owner will tell you things in conversation they would never write down.
They will tell you which type of customer they quietly refuse. They will tell you the real reason they raised prices last year. They will use a word for a piece of their work you have never heard and will not find on their website. They will describe a competitor in one sentence that is sharper than anything in their marketing.
None of that survives note taking, because while you are writing the first thing down they are saying the second thing. So do not take notes. Record the call, ask permission on the recording so consent is captured in the same file as the conversation, and listen properly.
Then transcribe it, within a day, while your memory can still repair the places where the audio is unclear or a name is garbled.
That transcript is the single highest value asset in the whole build. Everything the business sends you afterward was written to be read by strangers. The transcript is the only source where they are speaking normally.
There is a reason the first form in this pack is called the source inventory rather than the questionnaire.
Roughly half of what you need to know is already visible from the outside. Whether a website exists and what state it is in. Whether the Google Business Profile is claimed. How many reviews there are and what the bad ones say. Which social accounts exist and whether anyone has posted to them since 2021. What the industry associations list about them.
Answer all of that yourself before the second conversation. Two things happen when you do.
You stop wasting the client's time on questions you could have answered alone, which is the fastest way to look like you are paying attention. And the gaps you find become the most useful part of the whole exercise, because "you have 34 reviews and have never replied to one" is worth more than any answer they could have given you.
Do not guess. A field marked "do not know yet" is honest and gets resolved. A guess becomes a fact nobody ever checks.
The forms in this pack are separate on purpose.
Send all of them to the owner and you get one document back, late, half filled, in one voice. Send the money form to whoever does the books, the licensing and safety form to whoever actually owns that, and the vocabulary form to the person who has done the work longest, and you get answers from people who know.
You also get something you did not ask for, which is a picture of who understands what. That is usually worth knowing.
This is the step people skip, and skipping it is why most knowledge bases fail.
You will finish gathering with a pile: transcripts, filled forms, a price list as a spreadsheet, some photos, an insurance certificate, a safety manual from 2019. None of that is usable yet. It is raw material in the shape it happened to arrive in.
Normalizing means moving the information out of that shape and into files organized by what the information is about. The forms in this pack each declare which files they fill, so the work is mechanical rather than a judgment call. Work the mapping table, form by form, until every file is either written or explicitly marked as not applicable with a reason.
Keep the raw sources. Do not throw away a transcript because you extracted from it. You will read it again in six months and find something you missed.
Connect the AI to the folder and it will feel finished. It is not finished, and the way to find out is to interrogate it.
Ask it five questions only an insider could answer. Not "what does this business do," which it can answer from the first file. Ask what a specific customer type actually cares about, or what the business refuses to do and why, or how a job goes wrong most often, or what the busiest month is and what breaks during it.
If the answers are specific, you built it right. If they are fluent but general, a file is thin, and the hedging tells you which one. Go back, fill it in, and ask again.
This step takes twenty minutes and is the difference between a system people trust and a system people quietly stop using.
Everything above describes a build with an end. The real value has no end.
Every call after this one gets recorded and transcribed and folded back in. Every complaint teaches the service file something. Every quote that was won or lost teaches the pricing file something. The knowledge base in month six knows things the owner could not have told you in month one, because it has been paying attention continuously and no human does that.
That is the actual pitch, and it is worth saying plainly. Most systems are worth the most on the day they are delivered and decline from there. This one is worth the least on the day it is delivered, which is exactly why it is worth building.
Open 01-build-checklist.html and work the phases in order. It covers the same ground as this document, but as a sequence of things to do rather than an argument for doing them.
The structure you are filling is in business-md-system/. The questions that fill it are in business-md-system/intake-forms/.