Your business already has the answers. AI just needs the context.
Ten documents that make an AI tool useful for your business instead of generic. What each one is, what goes in it, and how to build it. No course, no gate on the guide.
Most owners have tried AI, gotten a generic answer, and quietly concluded it is not for them.
They are right about the output and wrong about the cause. Generic AI gives generic answers because it knows nothing about your business. It does not know what you sell, who buys it, what you charge, or why you turn certain jobs down. Handed a blank page, it writes what an average business would want to hear.
The fix is not a better prompt. It is context: a set of documents that describe your business accurately enough that a machine can reason about it, and plainly enough that a new hire could read them on their first morning.
Ten documents. You can do all of it yourself. It is a real project, mostly interviewing yourself, and the hardest part is that you cannot see your own blind spots.
Each one should be readable in four or five minutes. A tight reference beats a thorough one, for you and for the machine. Buried facts get missed by both.
Fragments are fine. Lists are better than paragraphs. If a line reads like prose, cut it down.
The order is not arbitrary. Each one supplies material to the next, and the last one needs all nine before it.
Source of Truth
Start here. Not because it matters most, but because it is the only one you can build without deciding anything.
Why first
Every other document asks you to conclude something. This one asks you to collect what already exists. That difference is why most people quit on document three: they opened a blank page and were asked to define their ideal customer cold. Collection has momentum. Contemplation does not.
What you are building
The canonical facts layer. Highest precedence for pricing, policies, claims, and constraints. When any other document disagrees with this one, this one wins.
Not a report. A fact file, dense enough that someone competent could answer basic questions about your business without calling you, short enough to read in four or five minutes.
What goes in it
- Identity. Legal entity, founding year, location, licenses, headcount, service area and its limits.
- Numbers you can prove. Jobs completed, customers served, reviews, years running. The countable things.
- Pricing as published. Every offer, current price, minimums, what fees apply and to what.
- Pricing principles. Fixed or negotiable. Discount policy. What you change instead of price.
- Approved and disallowed claims. Two columns. What you can say with evidence behind it, and what you must never say. The most useful block on the page.
- Constraints and open items. Capacity limits, paused channels, the things you already know are stale.
Three of these are skeletons of documents you will build later. Your offer list becomes the Offer Catalog. Your headcount becomes the Org Map. Your operating limits become the Operating System. Capture what is true today, not what it should be. You will come back.
How to build it
Two sessions. Not two weeks.
Session one is retrieval, ninety minutes. Open your bank account, payroll, calendar, accounting software, and your website. Copy facts out. Do not editorialize, do not fix what you notice is wrong, do not stop to think about strategy. Collect.
Session two is compression. Sort the pile into the blocks above and cut it to length. Where you find a gap, write the gap as a question rather than leaving it blank. Gross margin: unknown, need to separate materials from labor is worth more than an empty field. The gaps are the list of what you have been running without.
What done looks like
Someone who has never worked for you can read it in four minutes and get your prices, your claims, and your limits right. Boring to read. That is correct.
The failure mode
Writing it as marketing. The instinct is to describe the business you want. Resist it. If a sentence would look good on your website, it probably does not belong here.
You now have a customer list, a review corpus, and a set of prices sitting in front of you. That is exactly the raw material for deciding who this business is actually for.
Ideal Customer Profile
The first document that asks you to decide something. Start with the evidence you just collected rather than with your opinion.
What you are building
A precise definition of who to sell to and who to ignore. One persona, not four. If you genuinely serve two unrelated markets, write two, but be honest about whether the second is a real segment or a handful of exceptions you are reluctant to let go of.
What goes in it
- The persona in one line. Who they are, what situation they are in, and what they have already decided before they contact you.
- What they are trying to get done. Split three ways: the functional job, the emotional job, and the social one. The functional job is why they call. The other two are why they choose you over the next option.
- Buying context. The trigger that starts the search. Whether they are discovering or validating. When in the cycle they buy. Where they look. What they compare on.
- Objections and handling. The four or five things they actually say, with your actual answer. Written down, not improvised.
- Trust signals that convert them. Which specific proof moves them. Review volume, tenure, credentials, response speed.
- Anti-profile. Who to turn away, named as specifically as the profile. The section most owners skip and the one that saves the most money.
How to build it
Do not start from imagination. Start from your last twenty customers.
Pull the list. Mark the ten best, meaning profitable, pleasant, and likely to return or refer. Mark the five worst. Then find what the ten have in common that the five do not. It is rarely demographic. It is usually situational: a trigger event, a budget already allocated, a decision already made before they called.
For objections, read your own sent mail. The answers you have typed five times are the real objection handling, and they are better than anything you would write fresh.
For the anti-profile, look at the five worst and name the trait that predicted it. Then ask the harder question: does your marketing currently invite that person in?
What done looks like
You can read a new inquiry and know within two lines whether it is a fit, and you could hand that judgment to someone else on your team.
The failure mode
Writing the customer you wish you had. The profile has to describe who actually pays you today, or every document downstream inherits the fiction.
You have just written down how this customer talks about their own problem. That vocabulary is the foundation of how your business should sound.
Brand Voice Guide
Now that you know who you are talking to, decide how you sound talking to them.
What you are building
A written system for how your business sounds, specific enough that someone else could write in your voice and you would not need to rewrite it.
What goes in it
- Voice pillars. Three or four named qualities, each with a one-line definition. Named, not adjectives in a list. “Refined warmth: sophisticated but never stiff” is usable. “Professional, friendly, trustworthy” is not.
- Tone by context. A short table. Website, email, social, proposals, phone. The voice holds; the register shifts.
- Prefer and never. Two columns of actual words. The never column is more valuable and takes more nerve to write.
- Grammatical person. Do you say I or we. Do you address the reader as you. Consistency here does more work than word choice.
- A hard test. One repeatable check anyone can apply in thirty seconds.
- Approved phrasing bank. Five or six lines you have already written that are exactly right. These get reused forever.
How to build it
Extraction, not invention. Gather five pieces of your own writing that felt right when you sent them. A proposal that landed, an email a customer complimented, a page you did not hate.
Read them together and mark what repeats. Sentence length and rhythm. Words you reach for. Words you avoid. Whether you lead with the benefit or the detail. That pattern is your voice. It already exists; you are transcribing it.
For the never list, gather three pieces of writing from your category that make you wince. Competitor copy works well. Name what is wrong with it in specific words, not in general disapproval. Those words go in the never column.
What done looks like
You hand it to a contractor or paste it into an AI, get a draft back, and you are editing rather than rewriting.
The failure mode
Aspiration. A voice guide describing how you wish you sounded produces writing that sounds like nobody. Extract from what already works.
You know who you serve and how you sound. The next question is who else is talking to that same person, and what they are promising.
Competitive Analysis
Your customer is comparing you whether you look or not. This document is what they see.
What you are building
A clear read on the field. Who you actually compete with, how they position and price, where they win, and the gap you can own that they cannot easily copy.
What goes in it
- The real competitive set. Usually three to five. Include the alternative that is not a company: doing nothing, doing it themselves, or the cheap substitute.
- How each one positions. Their claim in their words, not your characterization of it.
- Pricing, where visible. What they charge and what is included. Note where you had to guess.
- Where each one genuinely wins. If you cannot name a real strength for every competitor, you have not looked hard enough.
- Your gap. The thing you do that they cannot copy quickly, stated in one sentence.
- The comparison your customer actually runs. Which two or three criteria decide it. Rarely the ones you would pick.
How to build it
Shop them. Fill out the contact form under a real name, read the auto-response, note how long a human takes to reply and what they say. That sequence tells you more than their homepage.
Read their reviews, especially the mediocre ones. Three-star reviews name the gap between what a competitor promises and what they deliver. That gap is your opening, and the language in those reviews is the language your customer already uses to describe it.
For your own gap, apply one test: could a well-funded competitor copy it inside ninety days? If yes, it is a feature, not a position. Keep looking until you find the thing that would take them years or a different business model.
What done looks like
You can state in one sentence why a customer picks you over the closest alternative, and back it with something a competitor cannot claim.
The failure mode
Building the flattering version. A competitive analysis where you win every row is not analysis, it is reassurance. The document is only useful if it can tell you something you did not want to hear.
You know the field and your gap in it. Now structure what you sell so the gap is obvious in the buying decision itself.
Offer Catalog
You listed your offers in the Source of Truth. This is where you decide whether that list is doing its job.
What you are building
Your services structured into clear offers, each with a reason to exist, arranged so the right choice is obvious to a buyer in ten seconds.
What goes in it
- The ladder. Every offer from entry to premium, in one visual sequence, with price and minimum. Seeing them ranked reveals gaps and redundancies you cannot spot in a list.
- Per-offer specs. For each: who it is for, what is included, and explicitly who it is not for. The “not for” line prevents more bad fits than any qualifying question.
- Fees and add-ons. What applies, to what, and how you explain it.
- The packaging rule. Your standing policy when a buyer wants a lower number. Change scope, hold price, or discount. Decide it once here so you are not deciding it live under pressure.
- Routing logic. Given what a buyer tells you, which offer they should land on. Written as rules, not instinct.
How to build it
Lay every offer on the ladder first and look at the spacing. Most catalogs have one of two problems: a gap so wide that buyers who cannot afford the top tier have nowhere to go, or rungs so close together that nobody can tell them apart. Both suppress revenue quietly.
For each offer, write the “not for” line before the “for” line. It is harder and it clarifies faster.
For routing, take your last ten inquiries and write the rule that would have routed each correctly. If you cannot write the rule, the catalog is not yet clear enough.
What done looks like
A stranger reads the ladder and picks correctly without asking you a question.
The failure mode
Cataloging what you do instead of what someone buys. A list of capabilities is not a set of offers. Every rung needs a reason someone would choose it over the rung below.
The offers are structured. Now trace what actually happens between a stranger finding you and that offer being sold.
Customer Journey Map
The offers are right. This is where you find out how many of them you lose on the way to the sale.
What you are building
The full path from first contact to repeat customer, stage by stage, with the friction named at each step.
What goes in it
- The stages. Usually five or six: discovery, first contact, quote or proposal, decision, delivery, after. Use your real stage names.
- What happens at each stage. What they do, what you do, how long it takes.
- What they need at each stage. The question they are actually asking. It is usually not the one they typed.
- Where you lose them. Named specifically, with whatever evidence you have.
- The moment of decision. The single point where it tips. Most owners guess wrong and optimize the wrong stage for years.
- After the sale. What happens next, or an honest admission that nothing does.
How to build it
Walk your own funnel as a stranger. Search for your service the way a customer would, find yourself, submit an inquiry, and time every response. Most owners discover a dead stretch they did not know existed.
Then get real numbers on the biggest gap. How many inquiries turn into quotes, and how many quotes turn into jobs. Two ratios. If one is far worse, you have found your constraint and the rest of the map is detail.
For the decision moment, ask five recent customers what made them choose you. The answer is often something you consider minor: a fast reply, one photograph, a question you asked that nobody else did.
What done looks like
You can point at the one stage where fixing friction would produce the most revenue, and say why with numbers.
The failure mode
Mapping the ideal path instead of the real one. If your map has no dead ends or delays, you have drawn the brochure version.
You have mapped what you think happens. Your customers have already written down what actually happened.
Voice of Customer
Your best marketing copy is already written. Your customers wrote it.
What you are building
A structured pull of what customers actually say, mined for the phrases, pains, and objections that repeat.
What goes in it
- Recurring phrases. The exact wording that shows up again and again, quoted verbatim. Never paraphrased into your language, which defeats the purpose.
- What they say they bought. Often different from what you think you sell.
- The pains they name. In their words, ranked by frequency.
- Objections in the wild. What they hesitated over, from emails and reviews rather than memory.
- What they warn other people about. Reviews often contain advice to future buyers. That advice is your positioning, pre-written.
- The gap. Where your marketing language and their language diverge.
How to build it
Gather everything: reviews across platforms, testimonials, thank-you emails, and the messages where someone explained why they chose you. Fifty items is plenty. Thirty is workable.
Read for repetition, not for praise. You are looking for the phrase that appears in eleven reviews, even if it is mundane. Mundane and repeated beats glowing and unique, because repeated means it is the actual reason.
Then compare against your own website copy. Where your customers consistently use a word you never use, that is a free conversion improvement waiting.
Read the negative and neutral reviews last, and separately. They name the objection your happy customers overcame silently.
What done looks like
You have a list of phrases you can paste directly into a landing page, and you know at least one thing your marketing says that no customer has ever said back.
The failure mode
Collecting compliments. This is not a testimonial file. It is a language file, and the useful entries are often unflattering.
You know how customers describe the experience. Now write down how the work actually gets done, so the experience holds when you are not the one delivering it.
Operating System
You listed your systems in the Source of Truth. This is where you write down how the work actually gets done.
What you are building
The documented way the business runs. The core procedures and standards that turn how I do it into how we do it.
What goes in it
- The standards. Non-negotiable quality bars, stated as absolutes. Absolutes are enforceable; preferences are not.
- The core procedures. Five to eight, not fifty. Only the ones where doing it wrong costs real money or reputation.
- Decision rules. What someone is allowed to decide without asking you, with the dollar or risk threshold stated.
- The handoffs. Where work passes between people. This is where most quality is lost.
- Policies that touch customers. Deposits, cancellations, change windows, refunds. Written once, applied identically.
- What is not documented yet. An honest list.
How to build it
Do not write procedures from scratch. Narrate them.
Next time you do the task, talk through it out loud while you work and record it. Then have that transcript turned into steps. Fifteen minutes of narration produces a better procedure than an hour of writing, because you describe what you actually do rather than what you think you do.
Prioritize by cost of error. The procedure worth documenting first is the one where a mistake costs you a customer or a day. Ignore anything where a mistake costs five minutes.
For decision rules, look at the last twenty decisions someone routed to you. Most of them had an obvious right answer and only reached you because nobody knew the threshold.
What done looks like
Someone competent could deliver your work to your standard without you in the room, and you would not wince at the result.
The failure mode
Documenting everything. A fifty-procedure manual is unmaintainable and nobody reads it. Eight procedures that cover the expensive failures beat fifty that cover everything.
The work is documented. The remaining question is who owns each piece of it.
Org Map
Procedures describe the work. This describes who is accountable for it.
What you are building
A clear map of who does what, where the gaps and overlaps are, and what to hire for next.
What goes in it
- Roles, not people. List the functions the business needs, then assign names. Two roles held by one person is a fact worth seeing; a person with an ambiguous title is not.
- Single accountability per function. One name per role. Shared accountability is no accountability.
- What each person owns end to end. Stated as an outcome rather than a task list.
- The gaps. Functions the business needs that nobody currently owns. Usually the honest answer is you, by default.
- The overlaps. Where two people both think they own something, or both think the other does.
- The next hire. Which role, and the trigger that says it is time.
How to build it
List functions first, before any names. Sales, delivery, scheduling, billing, marketing, hiring, customer follow-up, vendor management. Then write a name next to each.
The exercise reveals itself immediately. Roles with your name on them that should not have your name on them are your real bottleneck. Roles with no name are the things quietly not happening.
For the next hire, do not start from what you can afford. Start from which unowned function is costing the most, then work out what it would take to cover it. Sometimes the answer is a contractor or a tool rather than an employee, and you only see that if you started from the function.
What done looks like
Everyone can name the one thing they are accountable for, and when something falls through, it is clear whose it was.
The failure mode
Drawing a hierarchy instead of a map of accountability. Who reports to whom matters far less than who owns what, especially under ten people.
Nine documents describe how the business works. The last one asks whether it works profitably.
Money Model
Last, and hardest. Everything above feeds this, which is why it cannot come first.
What you are building
Your money model answers one question: how does this business actually make money? Not how much it takes in. How it makes money, which is a different question with a less comfortable answer.
The profit picture. Where money comes from, what each line nets after real costs, and the levers that move take-home rather than revenue. Same shape as the nine before it. Short, dense, and mostly numbers.
What goes in it
- The target and the math. Your revenue goal and the unit economics under it. What one customer pays, what they cost to serve, what is left.
- Where the money comes from. Every revenue line ranked by margin quality per hour of your team's time. Not by revenue. By what it nets for the effort it consumes.
- The constraint. One sentence naming what actually limits growth: demand, capacity, cash, or your own hours. Most owners guess wrong here.
- The profit levers, in order. Three or four, ranked by impact, specific enough to act on this quarter.
- Gaps. What you tried that did not work, and what you still cannot answer.
What you need before you start
This is why it is last. Done properly it needs your real offer list, your actual customer profile, your competitive position, and your cost structure. If you built the other nine, you have the source material. If you skipped here, you are guessing with confidence.
How to build it
Write a line for every offer you sell. Price, cost to deliver, what is left. Include your own time at a real rate. Most owners discover their most popular offer is their worst one, and that they have been working hardest at the thing that pays least.
Then answer three questions honestly.
One. If you raised prices fifteen percent tomorrow, who leaves? Not “would anyone,” which is a fear. Which specific customers, and would remaining revenue be higher or lower. Most owners cannot answer this and price defensively as a result.
Two. What does it cost to get one customer, and how long until they pay it back? Two numbers, not one. First, add everything you spent last year acquiring customers, including ads, commissions, and your own selling time at a real rate, then divide by customers gained. That is your cost. Second, take what an average customer nets you per month after delivery costs and divide the first number by it. That is how many months you carry them before they turn profitable. If that is longer than your average customer stays, you are paying for the privilege of being busy.
If you sell one-off purchases rather than ongoing relationships, payback is a single-transaction question instead of a duration one: does one sale clear its own acquisition cost, and if not, what has to be true about repeat business or referral for the math to work?
Three. Which of the three levers is actually available right now: more customers, larger average sale, or more frequent purchase? All three sound available. Usually one is, and the other two are a year of work.
What done looks like
You can name the single change that would most improve profit next quarter, and defend it with numbers instead of instinct.
The failure mode
Optimizing revenue. Revenue is the number you say at parties. Plenty of businesses scale a model that quietly loses money on half of what they sell, and scaling makes it worse rather than better. Growth applied to a broken model is a faster way to find the bottom.
Why this one is genuinely hard
The other nine ask you to write down what you know. This one asks you to see something you are standing inside of. You set these prices. You built these offers. You have reasons, and the reasons feel like facts.
There is also a data problem. A working money model runs on numbers most owners do not have on hand: true margin by offer line, blended margin across the mix, acquisition cost per real customer, actual capacity headroom. Getting them is the work. It is unglamorous, it is where the exercise stalls, and no template solves it for you.
The most common outcome of a solo money model is a document confirming what the owner already believed, which is worse than not doing it, because now the belief has a citation.
Two things fix that. Real numbers, which the first document started you on. And a reader with no stake in your prior decisions.
If you get here and the numbers do not resolve, that is not failure. It is the line between what an owner can do alone and what needs an outside read. It is the one section where I would tell you to consider help, and I would tell you that whether or not the help came from me.
Go back to the first one.
You have ten documents. Now return to the Source of Truth and reconcile it.
The offer list you sketched in document one is now a real catalog. The headcount is now an org map. The operating limits are now a documented system. Update the first document so it points at the other nine rather than duplicating them, and it becomes what it was always meant to be: the index, and the tiebreaker when two documents disagree.
That is the loop. Ten documents that describe the business, and one that governs them.
Then use them. Paste the relevant ones into whatever AI tool you use, at the start of the conversation, before you ask your question. That is the whole trick. The answers stop being generic because the machine finally knows something.
Three shells to start from.
Source of Truth, Ideal Customer Profile, and Brand Voice Guide. The three where a structure saves the most time, and the first three you would build anyway.
Source of Truth, Ideal Customer Profile, and Brand Voice Guide. Fill-in shells, no course, no sequence.
Or have it built.
If you would rather not spend the weekends, this is what I do. The same ten documents, built from your business, with the analysis on top that the documents make possible.
The honest version: the value is not the documents. It is the extraction, getting what is in your head onto a page, and the outside read on what the numbers actually say. That part is hard to do for yourself, which is the reason this page exists rather than a sales pitch.
See the advisory track