This is a free course for owners of established service businesses with a team. You will build a personal operating system with AI to reduce recurring work on your plate and the questions that keep coming back to you.

What this is

This is the do-it-yourself version of the six-week Owner Leverage System. It follows the same method and order, but you do the work yourself. It is not complicated. The course makes AI integration easy enough to use in a real business, with precise steps designed from experience helping business owners make AI productive.

The goal

The goal is to integrate AI into your day-to-day work where it actually helps. We will give it better context, choose useful repeat work, and set up a few reusable ways of working so you spend less time repeating yourself and checking generic answers.

What you need

For now, all you need is access to the AI tool of your choice. Bring real work when a lesson asks for it.

What this course is not

This is not a course on how to use one specific tool. Ask your AI tool how its current interface works, then use this course for the business method and decisions the tool cannot make for you.

Use your tool for tool instructions

When you do not know where a setting or feature lives, ask the AI tool you are using. Say which app and plan you have, describe what you want to do, and check its answer against current official help if the setting affects privacy, permissions or data.

Where this comes from

I am Marta, from ValueLab. I designed this method for owners of service businesses who want AI to make their work more productive without adding another complicated project.

Four things change if you do the work in this course. Worth being plain about them before you spend an evening on it.

1. You get time back

The hours currently going into drafting, writing up, summarizing and re-explaining. Most owners who do this properly get somewhere between half a day and a full day back each week, depending on how much of the week was that kind of work to begin with.

2. Your team stops needing you for everything

The questions that arrive constantly, the approvals people wait on, the things only you know. Written down once, in a place they can ask, so work keeps moving when you are in a meeting or on a beach.

3. You get better at using AI

Not prompt tricks. The actual method: how to set it up so it knows your business, how to tell whether an output is any good, how to fix it when it is not. That applies to everything you use AI for from here, well beyond this course.

4. You learn where not to use it

The one nobody sells you. By the end you will have a clear view of which parts of your week should stay human, and that is worth as much as the automation, because it stops you wasting a month on something that was never going to work.

What you do with the time is your business

Some owners put it straight back into growth: more clients, better work, the projects that never get started. Others use it to stop working Sundays, or to actually go mountain biking on a Wednesday afternoon.

Both are fine. The point is having the choice, which you do not have when your week is full of work that could have been prepared for you.

What it is not worth

It will not fix a broken process. It will run a broken process faster, which is worse, because now you get bad output at scale and on a schedule.

It will not replace your judgment, and every tool that claims otherwise is one you should look at sideways. The point is to get more of your week pointed at the judgment, not less.

Before we start building anything, it is worth understanding what AI actually does. Not a computer science lecture. Just enough to know what it is good at, what it is bad at, and why we set things up the way we do in the modules that follow.

Type 2 + 2 into a calculator a hundred times and you get 4 a hundred times. A calculator follows fixed rules.

AI does not work like that, and it does not actually think. It is a prediction machine. It works out what words should probably come next, one piece at a time, based on patterns from an enormous amount of text. There is deliberate randomness in that prediction, which is why the same question can give you a slightly different answer each time. Sometimes better, sometimes worse, usually just different.

It can be genuinely smart. It can also be artificially intelligent in the least useful sense, confidently producing something that sounds right and is not, if you do not give it the right context. That word is going to come up a lot. Get used to it.

What follows from that

Four things, and they are the difference between owners who get real value out of this and owners who try twice and give up.

This course asks you to give AI real information about your business. Before you do, two minutes on which version of which tool you are typing it into.

Check the settings of the app you chose

Free and paid versions can have different data settings. If you are using a free version, you may be agreeing that anything you type or upload can be used to train models. Check the current settings and privacy policy of the app you chose before using real business information. If you are unsure, use fictional or redacted information until you know what is allowed.

Make a deliberate choice

You decide what is fine to share, what must be removed, and what stays out. Use only the information the task needs. If other people on your team use AI, write the boundary down and agree who handles uncertain cases.

Connecting other tools

Claude, ChatGPT and Gemini can offer connections to other tools, depending on the app, account and workspace. If you are on a business workspace, email, drives or calendars may already be available through an administrator-managed connection. Check rather than assume.

A connection can reduce copying and pasting. It may let AI find a meeting, summarize a thread, draft an email, update a record or search approved files. You decide whether that is useful. Start read-only if that is enough. Check exactly what the connection can see or change, and ask IT support or an administrator to approve it when required.

The rule that does not change

You stay accountable for the output. AI can prepare, summarize, compare and draft. You check anything that affects a client promise, legal obligation, hiring, pricing, financial advice, medical information or regulated work.

Your AI tool can become more useful when it can work with the places your information already lives. This is a decision about access, permissions and what you want it to do.

What a connection can do

Claude, ChatGPT and Gemini can connect to other tools depending on your app, account and workspace. Google Workspace users may already have access to Gmail, Drive and Calendar connections through their workspace settings. Check what is available in your account rather than assuming it is active.

A connection can reduce copying and pasting. It may help AI find a meeting, summarize a thread, draft an email, search approved files, or update a record. You decide which of those would actually help your week.

Choose the access deliberately

Before you connect anything, check which account and folders it can reach, whether it can read or write, and where its output will appear. Read-only access is often enough to start. Keep sensitive folders, client data and actions that send or change records out until you have a clear reason and approval.

If your workspace is managed, your IT support or administrator may need to approve the connection. Ask what the connection can access, how to turn it off and how to review its activity.

Start with one useful connection

Choose one narrow use case, such as finding the notes from a calendar meeting and preparing a follow-up draft. Test it with a non-sensitive example. Check the sources it used, the permissions it has and the draft before anything is sent or changed.

I use [AI app, account and workspace]. I am considering connecting it to [Google Workspace / Gmail / Drive / Calendar / another tool].

Help me check:
1. What this connection can read and whether it can write or send
2. Which account, folders or records it can access
3. Whether an administrator or IT approval is required
4. A narrow read-only test I can run first
5. How I can review activity and disconnect it

Suggest one useful, low-risk use case for my work. Do not invent interface steps; point me to the current official instructions for this app and account.

A good system can still feel uncomfortable when you are used to being the person who catches every detail.

Why owners resist good systems

When the business depends on your judgment, stepping back can feel like lowering the standard. You may worry that a written rule will flatten the nuance, that AI will make a confident mistake, or that the team will stop needing your expertise.

Those concerns are useful signals. They tell you where the review point, exception or escalation rule belongs. The goal is not to replace your judgment. It is to make your judgment available at the right moment, so you are not repeating it in every email, meeting or interruption.

Reframe the role

Think of AI as an amplifier. It can carry your context further, prepare the first version and surface patterns. You still decide what is true, what is safe and what deserves your name. A good system gives you more reach without asking you to give up the standard that made the business work.

Think about what happens when a good new hire starts on Monday. Capable person, right background, and still not much use on day one.

Picture what it is actually working with. Every industry, every opinion, every way of doing your job that anybody has ever written down. Millions of businesses that are a bit like yours and not one that is yours.

So when you ask a plain question, it answers from all of it at once. You get the average. On a good day that lands close to right. On most days it is a coin flip, and the annoying part is that the wrong half still reads confidently.

Give it a magnifying glass, then a microscope

That is what context does. It does not make AI smarter. It narrows what it is looking at.

The magnifying glass is your business. Who you serve, what you sell, how you talk, what you refuse to do. Now it is no longer answering from the whole internet. It is answering from firms like yours, for clients like yours. Already a different quality of answer.

The microscope is the job in front of you. The steps you actually follow, the judgment calls, one example of what good looks like. Now it is not writing a client update. It is writing your client update, the way you write it.

Same tool, same question, three very different answers. Nothing changed except how narrowly it was looking.

Why you cannot just rely on its memory

Most tools now remember things across chats, so it is not starting from nothing. That memory is the problem as much as the solution. It holds a price you quoted in March, a service you have since dropped, a client you no longer work with, and treats all of it as current.

You cannot see what it kept and you cannot see what it ignored. So you get a draft built partly on things you said once, months ago, in a conversation about something else. That is worse than a blank slate, because at least a blank slate is obviously blank.

The fix is a set of files you control, that it checks at the start of every task. A current answer to what this business is and how you work, rather than a pile of half-remembered conversations.

The part most owners get backwards

When a draft comes back wrong, the instinct is to rewrite the request. Sometimes that helps a little.

The bigger gain comes from what you gave it to work with. A plain question with real context beats a cleverly worded question with none, every time.

Module 1 is where you build the magnifying glass. Module 3 is where you build the microscope.

They do not know your clients. They do not know how you price, what you would never put in writing, or which corners you refuse to cut. So their first attempt at anything is a sensible guess at what a firm like yours probably wants.

So you onboard them. A few examples of good work, the rules that live in your head and nowhere else, a bit of correcting for a month or two. Then they are useful, and it is the same person who showed up in September. The only thing that changed is what they know about you.

This module is that onboarding, written down once. The difference is that you only have to do it once, and it never leaves.

Three short files. A Company Profile, a Personal Profile and a Tone of Voice guide. Together they are your Context Profiles.

What each one is

The Company Profile covers the business. What you do. Who your best clients are. What makes you different from the other option they were considering. How your brand sounds when it talks to a client. If your team uses AI, everyone loads the same one.

The Personal Profile covers you. Your background. What people come to you for specifically. What you actually believe about how the work should be done. How you want AI to work with you. This one is yours alone.

The Tone of Voice guide captures how you write: your sentence length, level of directness, preferred words and the patterns that make a message sound like you. It keeps the style consistent when AI drafts for you.

What you get out of writing them

Three things, and the third is why this module comes before everything else.

1. You stop retyping. No more three paragraphs of background at the start of every chat. You paste two files, or better still, you save them once and the tool keeps them. Module 3 shows you how.

2. Every answer gets better, on everything. AI stops guessing at an average business and starts working from yours. Drafts sound closer to how you actually write. Advice fits your real situation instead of a generic one. Suggestions account for the kind of clients you take and the ones you turn down.

This applies to every single thing you use AI for from now on, well beyond this course. Client emails. Job postings. Pricing questions. A hard conversation you are trying to word carefully. All of it gets better because the tool finally knows who it is working for. On its own, that is already worth an hour of your evening.

3. Everything later in this course is built on top of them. In Module 3 you will build a workflow that runs on these two files. In Module 4 your written decision rules get added to them, so your team gets your answer rather than a generic one.

These two documents are the foundation. Every system after this sits on them, which is why the time you spend here comes back many times over.

Most of what a stranger needs to know about your business is already on your website. Let AI read it first, and only ask you about the parts a website never says.

The prompt below does three things a blank template cannot.

It checks your website for facts instead of asking you to retype them. It uses whatever it already knows about you from the chat. And then it interviews you, one category at a time, about the parts a website never says: who you are really best for, what you refuse to do, how you want to sound.

It covers six categories. Five are required and the sixth, current operations, is optional but useful, because it feeds straight into Module 2. Expect it to take a few rounds of back and forth. That is the prompt doing its job rather than handing you a template to fill in alone.

Give it your website address when you start. If your AI tool cannot browse the web, paste in your home page and About page text instead and tell it to work from that.

You are a business strategist and AI implementation consultant. Your job is to help me build a Company Profile: a single reference document I can attach to any AI conversation so the AI understands my business without me re-explaining it every time.

STEP 1. START WITH WHAT YOU CAN FIND.
Check my company website for facts about what we do, who we serve, and how we are positioned. My website is: [paste your website address]
If you cannot browse the web, tell me, and I will paste the text instead.
Also use anything you already know about me and my business from this conversation.
Write a short summary of what you found. Clearly mark which parts are confirmed facts and which parts you are guessing at.

STEP 2. INTERVIEW ME FOR THE REST.
Ask me the following questions one category at a time. Wait for my answers before moving to the next category. Do not generate answers for me. Skip anything you already confirmed from my website, but tell me what you are skipping so I can correct you.

Category 1: Business fundamentals [REQUIRED]
- What is your business name and what do you do in one sentence?
- What industry are you in?
- Where are you based? Do you serve clients locally, nationally, or both?
- How long have you been in business?
- How many people are on your team, including you?

Category 2: Services and offers [REQUIRED]
- What are your core services or products? List each with a brief description.
- What is your average deal value or typical engagement size?
- How long does a typical client engagement last?
- What is your pricing model: hourly, project-based, retainer, subscription?

Category 3: Ideal client [REQUIRED]
- Who is your ideal client? Be specific: job title, business size, industry.
- What problem are they trying to solve when they come to you?
- What have they usually tried before they find you?
- What objections or concerns do they typically raise?
- What does success look like for them after working with you?

Category 4: Competitive positioning [REQUIRED]
- Who are your main competitors?
- What makes you different from those alternatives? Be specific.
- Is there anything you deliberately do NOT do that competitors offer?
- What do clients say about you that you are most proud of?

Category 5: Brand voice and communication [REQUIRED]
- How would you describe your communication style in three words?
- Are there words, phrases, or tones you never want to use?
- If your brand were a person, how would they speak to a client?
- Do you have examples of writing you like?

Category 6: Current operations [OPTIONAL]
- What tools and systems do you currently use?
- What are the biggest bottlenecks in your current workflow?
- What tasks feel like they could be faster?
- Are there compliance or regulatory standards you must follow?

STEP 3. CONFIRM BEFORE YOU WRITE.
Go back through every fact you pulled from my website and ask me to confirm or correct it. Do not include anything I have not confirmed. Do not guess at anything.

STEP 4. WRITE THE DOCUMENT.
Compile my answers into a clean, structured document with clear headings. Format it as a .md file using this structure exactly:

# Company Profile - [Business name]
Updated: [today’s date]
Load at the start of any AI session before any task prompt.

## Business fundamentals
[content]

## Services and offers
[content]

## Ideal client
[content]

## Competitive positioning
[content]

## Brand voice and communication
[content]

## Current operations
[content]

## Instructions for AI
When I attach this document, use it as background context for everything you produce. Match my brand voice. Reference my services accurately. Write for my ideal client. Do not contradict information in this document unless I specifically tell you to.

STEP 5. SAVE IT.
Give me the finished document as a downloadable .md file named company-profile.md. If you cannot produce a file, output it as plain Markdown in a code block so I can copy it straight into a text file myself, and tell me how to save it as .md on my computer.

AI cannot look this one up. Nothing on your website says what you actually believe about how the work should be done. That belief is the part that changes your drafts the most.

This prompt still starts with what the chat already knows about you. The real answers have to come from you.

It works through seven sections. Section 4, your first principles, is the one people rush. Take your time on that one. It asks what you believe about the work that other people in your field would argue with, and it is the section that changes your drafts more than any other.

You are going to help me build a Personal Profile: a structured .md file that gives AI tools persistent context about me as an individual. Who I am, how I think, and how I want to be assisted. This is separate from my Company Profile, which covers the business. This one is about me personally: my expertise, my voice, my positions, and my working style.

STEP 1. GATHER WHAT YOU ALREADY KNOW ABOUT ME.
Before asking me anything, review our conversation history and any context you already have about me as a person. Extract what you can reliably confirm: my name and role, my professional background, my areas of expertise, any opinions or positions I have expressed, my communication style and tone preferences, any constraints or banned language I have mentioned, and how I prefer to work.
Write a brief summary. Flag anything you are uncertain about.
Do not include company-level information. Focus on me as an individual.

STEP 2. INTERVIEW ME FOR THE REST.
Ask focused questions to fill the gaps. Cover all seven sections, one at a time. Do not move on until I have answered. If I say skip, note it and continue.

1. Professional identity: name, title, sector, location, years in the field.
2. Professional background: key past roles, domains of expertise, credentials.
3. Core skills and strengths: what I am genuinely good at, what people come to me for, where I outperform.
4. First principles and point of view: positions I hold based on experience, things I believe that are not conventional wisdom, contrarian views.
5. Voice and tone: how I write, banned words, what my writing resembles, formatting rules, one example sentence I approve of.
6. Unique positioning and differentiators: what makes my approach distinct, my methodology, what clients say I do that others do not.
7. Working style and AI collaboration rules: when to push back versus execute, how to handle ambiguity, my iteration preference, what done looks like.

Take section 4 slowly. Push me if my answer is something nobody would argue with. A safe answer there produces safe drafts.

STEP 3. WRITE THE PERSONAL PROFILE AS A .MD FILE.
Once you have all sections, produce the complete document using this structure exactly:

# Personal Profile - [My name]
Updated: [today’s date]
Load at the start of any session. This covers me, not my company. Load it alongside my Company Profile where relevant.

## 01 - Professional identity
[content]

## 02 - Professional background
[content]

## 03 - Core skills and strengths
[content]

## 04 - First principles and point of view
[content]

## 05 - Voice and tone
[content]

## 06 - Unique positioning and differentiators
[content]

## 07 - Working style and AI collaboration rules
[content]

STEP 4. SAVE IT AND TELL ME HOW TO USE IT.
Give me the finished document as a downloadable .md file named personal-profile.md. If you cannot produce a file, output it as plain Markdown in a code block so I can copy it into a text file myself, and tell me how to save it as .md on my computer. Then tell me how to load it at the start of a session in the tool I am using.

A weak answer in section 4 gives you a generic draft. A real opinion, the kind a competitor might argue with, gives you a draft that sounds like a decision you would actually make.

Three words about your tone will only get you so far. Your real writing has habits you cannot see, because you are too close to it.

You already write in a specific way. Sentence length, how you open, how you close, whether you soften a no or say it straight, the phrases you reach for without thinking. Describing that from memory is genuinely hard.

Showing is easier than describing. There are two ways to do it.

Option one: hand over your own emails

Go into your sent folder and pull out some emails you have written. Get a spread: clients, vendors, your team. Include a few where you had to say no, or deliver news somebody did not want. Those are the ones that show the most.

Eight is enough to see the pattern. Twenty is better. Do not let the bigger number stop you doing this tonight. Eight emails you actually paste in beats thirty you meant to get around to.

Then paste them in and run this:

Below are 8 to 12 emails I have written to clients, vendors, and my team. Work out how I actually write, so you can match it in future.

Read them all, then analyse and extract each of the following:

1. OVERALL STYLE SUMMARY
A short description of how I write, as if you were briefing a ghostwriter.

2. VOICE AND TONE
My default register. Warm or brisk, formal or plain, measured or direct.

3. TONE SHIFTS
How my tone changes depending on who I am writing to and what the message is about. Clients versus team versus vendors. Good news versus bad news.

4. RELATIONSHIP TO READER
How I position myself. Peer, expert, service provider, friend. Whether I write up, down, or across.

5. STRUCTURE PATTERNS
How I open. How I close. How I order information. Whether I lead with the point or build to it. Typical message length.

6. SENTENCE-LEVEL STYLE
Sentence length and rhythm. Paragraph length. Punctuation habits. Use of fragments, lists, questions.

7. WORD CHOICE
Words and phrases I reach for often. Words I clearly avoid. How plain or technical my vocabulary is.

8. RHETORICAL DEVICES
Any repeated moves. Contrast, repetition, understatement, humour, analogies, rhetorical questions.

9. PERSUASION PATTERNS
How I get agreement. Do I lead with evidence, with the relationship, with the cost of inaction, with options? How I handle a no.

10. UNCONSCIOUS HABITS
Two or three things I do repeatedly that I probably have not noticed.

Then write the whole thing up as a tone of voice guide in Markdown, under the heading "Tone of Voice", using those section headings. Keep it specific and usable, not abstract. Finish with a short list of rules for writing as me, and three example sentences taken word for word from my actual emails that best show how I sound.

11. WHAT TO AVOID SO IT DOES NOT SOUND AI-WRITTEN
Add a section of rules that stop drafts reading like a machine wrote them. Include at least these: no em dashes, use commas or full stops instead. No sentences built as a negation followed by a correction, the "this is not X, it is Y" shape. No scene-setting openers like "in today’s fast-paced world". No three-item lists used for rhythm rather than because there are genuinely three things. No words like delve, leverage, unlock, robust, seamless, elevate, transform, harness, testament to, navigate the landscape. No closing paragraph that summarises what was just said. No hedging like "it is worth noting that". Then add anything else you can see in my own writing that a machine would get wrong.

Then give me the finished guide as a downloadable .md file named tone-of-voice.md. If you cannot produce a file, output it as plain Markdown in a code block so I can copy it into a text file myself.

Here are the emails:
[paste them here]

Option two: connect your email directly

Most AI tools can now connect straight to Gmail or Outlook, so they read your mail without you pasting anything. It is faster and it works from a much bigger sample.

If you go this way, use this version of the prompt instead. It picks its own sample, which is usually a better spread than the one you would have chosen.

Before you connect anything

A connection like this usually reads far more than the 20 emails you had in mind. It can mean access to your whole mailbox: client information, contracts, anything else sitting in there.

Check what the connection is actually asking for before you approve it. Check whether your client contracts or your industry rules allow it. If you have a team, check what your own policy says.

Depending on your IT setup, this may not be yours to approve on your own. Plenty of businesses have their email locked down so that an outside app cannot connect until an administrator allows it. If you have an IT provider or someone in-house who looks after this, ask them first. It is usually a short conversation and a setting on their end.

If you do want to connect, ask your AI tool how to do it. Describe your setup, say which mail system you use, and it will walk you through the steps for that specific tool. That is faster and more current than anything written down here.

This is your call and your risk. If there is any doubt, option one gets you a very good result with none of the exposure. That is the safer default, and there is no shame in taking it.

You have access to my mailbox. I want you to work out how I actually write, so you can match it in future.

FIND THE RIGHT EMAILS YOURSELF:
Look through my sent folder over the last six months and pick 25 to 30 emails written by me. Choose a spread across clients, suppliers and my own team. Deliberately include several where I had to say no, chase something, or deliver news the other person did not want, because those show the most.
Skip anything automated, forwarded without comment, or one line long.
Tell me which emails you selected before you analyse them, so I can rule any out.

THEN ANALYSE AND EXTRACT EACH OF THE FOLLOWING:
Use the same eleven headings as the paste-in version of this prompt: overall style summary, voice and tone, tone shifts, relationship to reader, structure patterns, sentence-level style, word choice, rhetorical devices, persuasion patterns, unconscious habits, and what to avoid so it does not sound AI-written.

Write it up as a tone of voice guide in Markdown under the heading "Tone of Voice", keeping it specific and usable rather than abstract. Finish with a short list of rules for writing as me, and three example sentences taken word for word from my real emails that best show how I sound.

Do not quote client names, figures or confidential details in the guide itself. The example sentences should show style, not content, so swap any identifying detail for a placeholder.

Give me the finished guide as a downloadable .md file named tone-of-voice.md.

Three documents written in a chat window prove nothing. Find out whether they actually work before you build anything on top of them.

The test is simple. Give AI the same job twice. Once cold, once with your files loaded. Then read both.

Pick something real and slightly awkward. A follow-up to a client who has gone quiet is ideal, because tone matters and a generic version is obviously generic.

Write a follow-up email to a client who has gone quiet for two weeks on a project we are partway through. I want to nudge them without sounding annoyed or desperate.

Do not ask me any questions first. Just write it.

Here are three reference documents about my business and how I write:

[paste company-profile.md]

[paste personal-profile.md]

[paste tone-of-voice.md]

Now write the same email as before: a follow-up to a client who has gone quiet for two weeks on a project we are partway through. Nudge them without sounding annoyed or desperate.

Then tell me specifically which parts of my profiles changed how you wrote it, compared to writing it with no context at all.

What you are looking for

Put them side by side. The cold version is usually competent and completely anonymous. It could have come from any business on earth. There is a decent chance it contains the phrase "just circling back," which is the email equivalent of a soggy handshake.

The second one should sound closer to you. Right level of directness, right assumptions about the relationship, fewer things you would strip out before sending.

Read the last part carefully, where it explains what your profiles changed. That tells you what is actually landing.

If the second one is barely better

That is useful information rather than a failure. It usually means one of three things:

• Your profiles are too vague. "We provide quality service" tells AI nothing it had not already assumed.
• Your personal profile has no real opinion in it, so there is nothing to write from.
• Your tone of voice file describes your writing in adjectives rather than rules.

Go back and sharpen whichever one it is. Ten minutes now beats finding out in Module 3, when you are building on top of these.

If the voice is still off, fix the file

Read the second draft and be specific about what is wrong. Too formal, too chatty, opens the wrong way, uses a phrase you would never say. Then tell AI exactly that and ask it to update the file rather than just the draft:

That draft is still not quite me. Here is what is off: [list what you did not like, as plainly as you can]. Update tone-of-voice.md so it captures this, add whatever rule was missing, and give me the revised file back.

Run the same test again. Two or three rounds and the file gets genuinely accurate, because you are correcting it against real work rather than describing your own writing from memory.

Keep that habit. Every time a draft sounds wrong, spend the extra minute updating the file, or you will be correcting the same thing forever.

Right now your two profile files only help if you remember to paste them. That lasts about two weeks. Then you stop bothering, and you are back to generic answers.

Right now your two profile files only help if you remember to paste them. That lasts about two weeks. Then you stop bothering, and you are back to generic answers.

Every major AI tool has somewhere to save files and instructions permanently, so it pulls them in on its own. The names differ. The idea is identical: teach it once, use it forever.

The useful way to think about this is a written set of working instructions. The kind of thing you would hand a new person so they can do a job properly without asking you how it goes. That is exactly what you are building here, except your AI is the one reading it.

Claude calls these Skills. ChatGPT calls them Custom GPTs. Gemini calls them Gems. Whatever the label, it is a saved bundle of instructions and files the tool reaches for when the job matches.

Do this now, before you build anything. Once your Context Profiles are loaded in, everything you build afterwards can use them without you pasting a thing. That includes the workflow in Module 3 and the decision rules in Module 4.

The buttons move. Ask your tool instead.

These menus get redesigned every few months, so any instructions written today go stale. Rather than hunting, ask the tool itself. Paste this in:

i If your tool does not offer Skills, Projects or Gems yet, keep pasting the three files in manually. It still works, with one extra step. Ask your AI where to put the files and how to load them for a new conversation.

I want to set up something like a Claude Skill: a saved bundle of reference files you pull in automatically whenever a task calls for it, so I never paste them in again. Think of it as a set of working instructions that you read rather than my team.

I am using [name your AI tool and plan, for example ChatGPT Plus, Claude Pro, or Gemini Advanced].

The files are:
- company-profile.md, background on my business
- personal-profile.md, background on me and how I want you to work with me
- tone-of-voice.md, how I actually write

Tell me, step by step, exactly where in this tool I should put them, using the current interface. Skills, Projects, Gems, custom instructions, whatever this tool actually calls it today. If there is more than one option, tell me which is best for files I want available across many different kinds of task, and why.

Then tell me three more things:
1. What should I name it, so it is obvious what it is for later?
2. How do I check it actually worked?
3. How do I update these files later without breaking the setup?

Follow what it tells you. Then test it: start a fresh chat and ask something about your business without pasting anything. If the answer shows it already knows who you are, it worked.

Review them every three months

Put a recurring reminder in your calendar now, once a quarter. Your services change, your best clients change, your team changes, and your opinions change. A profile that is eighteen months out of date makes every answer slightly wrong, and it is hard to spot because nothing obviously breaks.

The review takes ten minutes. Open both files, read them, fix what has moved on.

Share the company one with your team

Your Company Profile should not live only with you. Anyone on your team who uses AI for work should be loading the same file, otherwise you get four people producing four slightly different versions of your business.

Your Personal Profile and your tone of voice file stay yours. How you share the company one depends on your setup, and there are two situations.

Whichever you are, the test before you send it round is the same: are you comfortable with everything in that file if it were not kept private? A Company Profile usually passes easily, since most of it is on your website already. If one or two lines would bother you, take them out and share the rest.

Start using them today

You do not have to wait for the rest of the course. Next time you open AI for anything at all, a client email, a pricing question, a job posting, make sure your context is loaded first. Compare what comes back to what you normally get.

That habit alone improves everything you get from AI from here on.

You have the context files. Now you need to know what to point them at, and that starts with an honest look at where your week actually goes.

Most owners skip this and go straight to picking something. They grab whatever annoys them most this week, which is how people end up spending a Saturday automating something they do twice a year.

This module runs in three steps. Map where your week goes, filter that list down to what is worth building for, then sense check what survives before you commit to anything.

What you end up with

A list of places AI or a written rule could take work off you. Most owners find five to ten. Some find twenty.

Do not panic when you see it. You are not doing all of them. You are doing one, in Module 3, and getting it properly working. The rest is your roadmap: when the first one is running and you want another, you open the list and take the next, instead of sitting there trying to remember what was annoying you in March.

That list is the real output of this module. The one task you pick at the end is just the first item off it.

Two ways to do this

There is a lazy version and a proper version. Both work. They give you different things, and it is worth knowing what you are trading before you pick.

Fair warning on the proper version before you commit to it: it needs two working days of elapsed time, because you are tracking your week as it happens. It is the one place in this course where you cannot just carry straight on to the next lesson tonight. If that does not suit right now, take the lazy version, keep going, and come back and do the proper one later. That is a real option, not a consolation prize.

The lazy version, about five minutes

Answer one question honestly: what am I spending the most time on that is worth the least?

Not the hardest thing in your week. Not the thing you like least. The thing that eats hours and would not change much if somebody else did it, or if it were half as good.

Most owners already know. The answers usually sound like this:

• Writing up notes and follow-up actions after every client call
• Answering the same handful of client emails, worded slightly differently each time
• Putting together the weekly or monthly client update from scratch
• Rewriting the middle of a proposal using an old one as a starting point
• Being the person who answers the same team question over and over
• Turning a messy set of numbers into something readable for somebody else

If one of those made you wince, write it down and go straight to the next lesson. You have enough to build something today, and something built today beats a perfect map you never finish.

The proper version, about an hour spread over two days

The lazy version gets you one task. The proper version gets you the whole picture, and that picture is the thing you actually need.

Here is why it is worth the hour. Everything else in this course, and everything after it, runs on knowing where your week goes. The task you build in Module 3 comes from it. The questions your team keeps asking in Module 4 show up in it. When you finish the first workflow and want a second, the map tells you what it is instead of you sitting there trying to remember. Owners who skip this build one thing and stop. Owners who do it end up with a list they work through for a year.

It also catches what your memory deletes. The lazy version gives you the thing that annoys you most. That is not always the thing costing you the most, and the gap between those two is usually where the real hours are hiding.

Track your week in 30-minute blocks

Here is the part nobody enjoys. Go through your working day in 30-minute blocks and write down what you actually did. Yes, it sounds like homework, and yes, I am aware I said this course would save you time. It is also the most useful hour you will spend on any of this, because you cannot hand over work you have never named.

The trick that makes this survive past lunchtime: set a repeating 30-minute timer on your phone. When it goes off, write one line about what you just did, and reset it. Ten seconds each time. Without the timer you will remember at four in the afternoon and reconstruct the day from memory, and memory deletes exactly the interruptions you are trying to catch.

Your calendar will not do this for you. Calendars record meetings. They do not record the forty minutes you spent rewriting a proposal, or the six times somebody caught you on the way past your desk.

What a block actually looks like

Write them plainly. This is the level of detail that works:

• Answered four emails about project timing
• Reviewed a proposal before it went out
• Rewrote the middle section of a proposal using an old one as a starting point
• Call with a client about a change to scope
• Sorted a scheduling clash between two crews
• Answered a team question about which supplier to use
• Wrote up notes after a meeting and sent them round
• Chased two unpaid invoices
• Fixed a mistake on an invoice
• Half an hour I genuinely cannot account for

That last one is fine. Write it down anyway. Most owners have several.

A template you can copy

Four columns. Paper, a note on your phone, or a spreadsheet, whichever you will actually keep up with for two days.

The last two columns do the real work. The interruption column shows you what your calendar has been hiding. The final column is the one that stings, and it is where most of your answers are.

Two full days is enough to see the pattern. A full week is better. Do it as you go rather than from memory at the end. Memory deletes the interruptions.

Once you have your blocks, run the prompt below. It will not sort anything yet. It just makes sure you have captured the week honestly before you start filtering it.

You are helping me capture an honest picture of where my working time actually goes. Do not sort or analyse anything yet. This step is only about getting the full picture.

USE MY SAVED CONTEXT:
Use my saved setup, the one holding my company profile, personal profile, and tone of voice. If you cannot see it, tell me and I will paste the files in.

I am going to attach a breakdown of my last two working days in roughly 30-minute blocks. It shows what I did in each block, whether it was planned or an interruption, and whether somebody else could have done it. If I paste it in as text instead of attaching a file, work from that.

Read the whole thing before you say anything substantial.

Then ask me these follow-up questions, one at a time. These catch the time a calendar never recorded:
1. What interrupted you that never made it onto the list? Messages, quick questions, things you said yes to on the way past your desk.
2. Which meeting in the last two days could have been a short written update instead?
3. What did you do that you would struggle to explain to somebody else, and why?
4. Was there anything you avoided or postponed because you did not have the headspace?

When we are done, give me back a clean list of every distinct task, grouped by the ones that repeat. For each repeating task, estimate roughly how much time it takes me per month.

Keep that list. I will use it in the next two steps.

You have a list of repeating tasks, or one task you already know is the problem. Most of what people put on that list is not worth building anything for, and this filter tells you which.

The temptation is to start with whatever annoys you most. That instinct is how people end up spending a Saturday automating something they do twice a year.

If you took the lazy version

You have one task rather than a list. That is enough to carry on with. Run it through the three questions below on your own, in your head, and if it passes all three you have your candidate and you can go straight to the next lesson.

If it fails one, that is genuinely useful to know before you spend an evening building it. Write down two or three more things you do regularly, run those through as well, and pick whichever passes. Then carry on.

Run every task on your list through three questions. It needs to pass all three.

1. Repetitive

Does it happen often? Daily, weekly, or monthly. Monthly reports, standard client questions, routine cleanup. A one-off is faster to just do.

2. Rule-based

Is there a usual way to do it? Same kind of information going in, same kind of result coming out. If you can describe it as "when this happens, we do that," it will work. Tasks that need gut feel or reading a room will not.

3. Return

Is it worth the effort? Do not spend forty hours automating something that takes two minutes a week. Go after the tasks where one afternoon of setup saves hours every month.

Passes all three, worth exploring. Fails one, just do the thing manually. Some tasks do not need a robot. They need Tuesday afternoon and a coffee.

The question that comes before all three

There is a fourth question, and it should really be the first one. Your tracking table had a column asking whether somebody else could have done the task. Go back and look at what you wrote in it.

If the answer was yes, look at delegation before automation. A task that a capable person on your payroll could do should usually be delegated with a clear handoff and a written rule. It may still be a candidate for AI later, but do not automate a task simply because it has not yet been handed to the right person.

This matters more the bigger your team is. If you have a practice manager and four managers, most of what lands in your week is not work that needs a partner. It is work that has never been handed over, usually because handing it over once took longer than doing it. Those tasks want a person and a written rule, which is Module 4, not a workflow.

So sort your list into three, not two:

• Should go to a person. Somebody on your team could do this with a bit of instruction. Park it for Module 4, where you write the instruction once.
• Should go to AI. It is drafting, summarizing, formatting or comparing, and AI can prepare it for a human review. This is your Module 3 shortlist.
• Stays with you. It genuinely needs your judgment, your relationships, or your signature.

What your list looks like once it is marked up

Take the task list from lesson 2.1 and add five columns. Do it yourself, quickly, on instinct. Ten minutes with a pen is fine.

Read down the last column and the three piles fall out on their own. Rows one to three go to AI. Rows four and five go to a person, with a written rule where one is needed. Rows six and seven stay yours.

That is the sorting done. The prompt below does not redo this work, it tidies what you wrote, argues with you where your marks look wrong, and hands back a clean ranked list.

Only the AI pile carries into the next lesson. If that pile is thin and the person pile is fat, that is a genuinely useful thing to have found out in twenty minutes, and it is a better problem to have.

One thing worth saying if you have a real team. The method in this course works just as well run by your practice manager on their own repeat work, or by each of your managers on theirs. Four people each getting three hours back is worth more than you getting four, and none of it lands on your desk. You are doing it first because the judgment in Module 4 lives in your head. The rest of it does not have to.

Where owners get this wrong

• Keeping a task with missing inputs. If the information does not exist anywhere you can hand over, it stalls on day one.
• Keeping something that actually needs a business decision made first. AI cannot decide your pricing policy for you.
• Calling something rule-based when the rule lives only in your head. If you cannot write it down, it is not a rule yet. That is Module 4.

Here is my own first pass at sorting my repeating tasks. I have already marked each one up myself, so do not start from scratch and do not silently rescore everything. Your job is to tidy this up, challenge me where I have got it wrong, and turn it into a clean list I can work from.

MY MARKED-UP LIST:
[paste your table: task, roughly how often, roughly how long per month, who could do it, and your R / R / R marks]

WHAT THE MARKS MEAN:
- Repetitive: does it happen often enough to be worth building for
- Rule-based: is there a usual way to do it, or does it need gut feel each time
- Return: would the time saved beat the time it takes to set up
- Who could do it: me only, someone on my team, or nobody needs to do it

DO THIS:
1. Tidy it into one clean table, keeping my marks. Where I have written something vague, ask me rather than guessing.

2. Challenge me on anything you think I have marked wrong, one line each, with your reasoning. Push back specifically on: tasks I marked rule-based where the rule probably lives only in my head, tasks I marked as needing me where you suspect that is habit rather than necessity, and anything where my monthly time estimate looks optimistic.

3. Flag any task where the information needed probably does not exist anywhere I could hand over, or where a business decision has to be made before anything can be automated.

4. Split the list into three groups, using my "who could do it" marks as the starting point:
   - GOES TO A PERSON: someone on my team could do this with instruction
   - GOES TO AI: nobody needs to do it, and it passes all three of the other marks
   - STAYS WITH ME: genuinely needs my judgment, relationships or signature

5. For the GOES TO AI group only, rank them by monthly time saved and give me the ranked list with the time estimate next to each.

Ask me about anything unclear before you produce the final version. Keep the ranked list, we sort it in the next step.

You have a shortlist. Before you spend an evening building something, one more pass over it, using a grid that catches the mistakes the last lesson cannot.

This one comes from Dan Martell. I like it, I use it, so I am passing it on.

You may now have five or more tasks you could potentially automate. This framework maps which workflow to start with. Give is the usual first place to test: AI can prepare it and a human can check it quickly.

Sort the shortlisted AI use cases by how suitable they are for an AI workflow. In plain language: how well can AI prepare this kind of work, and how confidently can you check the result? This is a prioritization tool, not a test of whether a computer is “easy” or “hard.”

Those two questions make four boxes that help you choose what to automate first, what to explore later, and what to leave alone. The initials spell GAIN, which is either a helpful memory aid or a slightly convenient coincidence, depending on how cynical you are feeling.

What each box is for

Give removes admin drag. The work is easy for you and easy for a computer, so there is no good reason it is still eating your Tuesday.

Accelerate speeds up slow work. Things that take you ages but a computer handles quickly, like reading through options or spotting a pattern in a pile of numbers.

Integrate improves whole workflows. Harder for everyone, so AI becomes one step inside a bigger process rather than the whole thing.

No AI protects judgment and trust. Easy for a human, genuinely hard for a computer. Telling a long-standing client their project is over budget belongs here and always will.

Build the whole map, then build one thing

Let me be straight with you about something. You probably already have a hunch about which task to start with, and there is a decent chance your hunch is right. So why bother sorting everything?

Because the sorted list is worth more than the winner. What you end up with is a library: fifteen or twenty tasks, each labelled with what kind of help it needs and roughly what it costs you every month. When you finish the first build and want another, you open the list and take the next one instead of starting from a blank page and a vague feeling of being busy.

Then the warning that goes with it. Owners who start six things finish none of them and conclude AI does not work for their kind of business. Owners who finish one usually have five running within a few months, because each gets faster to build than the last. Build one properly first.

The goal here is better work, not more AI.

Use the axes literally: “Easy for a computer” means routine, structured and pattern-based. “Hard for a computer” means the task needs nuance, relationship judgment or context the tool may miss. Place your shortlisted AI use cases on the matrix, then use the position to prioritize what to test first.

Take the survivor list from the last step and sort it using the GAIN Matrix.

GAIN sorts work by two questions: how suitable each shortlisted use case is for an AI workflow, based on AI’s ability to prepare it and your ability to review it. That gives four boxes:

- GIVE (good candidates for AI preparation with a clear human review). Routine work with no real judgment in it. AI does the first pass, I review. Examples: meeting notes, summaries, first-draft emails, basic reporting, turning rough notes into structure.
- ACCELERATE (good at heavy analysis, but potentially harder for you to verify). Slow or effortful for me, fast for a machine. AI does the heavy lifting, I check the thinking. Examples: research, data analysis, comparing options, finding patterns across a lot of information.
- INTEGRATE (hard for humans, hard for computers). Multi-step work where AI becomes one step inside a bigger process rather than the whole thing. Examples: client follow-up systems, onboarding journeys, recurring reporting workflows.
- NO AI (work where human judgment should stay central). Judgement, relationships, trust. AI can help me prepare, but I decide. Examples: leadership, persuasion, coaching, sensitive conversations, final professional advice.

Put each task in exactly one box with a one-line reason.

Then recommend my first build, using these rules in order:
1. Prefer something from GIVE, even if an ACCELERATE task scored higher on the three filter questions. Give work is easy for me to check, so I learn faster from it.
2. Among the Give options, prefer the one that saves the most time per month.
3. Rule out anything where a bad first draft could reach a client, cost money, or create a legal or compliance problem.

Tell me your recommendation and why. Then describe, in three or four sentences, what a good first version of it would actually do. If you think I should pick something outside Give, say so and explain what makes it worth the exception.

Read the recommendation, then use your own judgment. The model is working from what you told it, and you know things about your business you did not think to mention.

You picked one task in Module 2. Before you build anything, write down how you actually do it, then let AI argue with you about whether that is a good way to do it.

Here is where this module is going, so you can see the shape of it:

1. Describe the process and improve it. This lesson.
2. Run it end to end with AI. Refine until the output is genuinely good.
3. Turn it into a skillWhat is a skill?
   A saved bundle of instructions and files your AI tool loads automatically whenever the job comes up. You set it up once and after that you bring only the new information, instead of re-explaining the job every time.
   
   Claude and ChatGPT call these Skills. Gemini calls them Gems. ChatGPT also has Projects and Custom GPTs, which do the same thing. Different names, same idea: repeatable instructions that stay put.
   
   You already built one in Module 1 to hold your context files. This is the second.. So next time you bring only the new information.
4. Put it on a schedule. Optional, and only for some jobs.

First, write down the steps

Explain the job the way you would to a capable new person in their first week. Three or four steps is usually enough. Include the judgment calls, not just the mechanical parts, because those are the bits that make the output right or wrong.

Remember the talk-first hack from Module 1.2. Use it whenever you are describing a process from memory.

Then check whether the process is any good

Most processes in a business were never designed. They grew. Somebody did it that way once under time pressure, it worked well enough, and eleven years later it is how the job gets done, including the step that exists because of a client who left in 2019.

Automate that as it stands and you have made an unexamined process permanent and faster. Same avoidable rework, now on a schedule, and harder to change because it is written down.

Ten minutes here saves you from that.

Before we automate this process, help me check whether it is worth keeping as it is.

PROCESS:
[paste the steps you actually follow]

Ask me only these questions, one at a time:
1. What outcome is this process meant to produce?
2. Which steps are repeated, and which are judgment calls?
3. What information is needed at the start, and where does it come from?
4. What can be removed, combined or moved earlier?
5. What must a human check before the result is used?

Then return: a shorter improved process, the inputs it needs, the human review point, and one thing I should test before saving it. Do not redesign the whole business or assume AI is the answer.

Take what it gives you with judgment. It does not know why you moved that step, and some suggestions will be things you already tried. But it also does not know that a step is sacred, which is exactly why it is worth asking.

If the process is fine, good, that took ten minutes and you now know. If it turns up one change worth making, make it before you build anything, because changing the process later means rewriting the instructions too.

Now do the job once, in an ordinary chat, and keep going until the output is something you would actually use.

Not a saved setup yet. A normal conversation, the same way you tested your tone of voice file back in Module 1, and for the same reason: you are finding out whether this task works at all before you commit anything to memory.

Skipping this is the most common reason first attempts go nowhere. People save a setup on day one, it produces something mediocre on day four, and they never open it again. If you cannot get one good result by hand, saving it just means you have made something that does not work repeatable.

What to give it

• Your saved context from Module 1, so it starts from your business
• The steps you cleaned up in the last lesson, including the judgment calls
• One real case to work on
• A description of the output you want. Be specific. How long, what sections, what order, who reads it, what it must never say. Vague in, vague out.
• One real example of your own finished work. Optional, and it is the single biggest improvement you can make. One example does more than three more paragraphs of instructions.

Most people underdo this. "Write me a client update" is a category, not an instruction. "A one page update, four sections in this order, addressed to a non-technical client, opening with the most significant change, never committing to a future date" is an instruction.

Then refine, the same way you refined your tone

Read what comes back against your real example, not against what you pictured. Then be specific about what is wrong:

• Say what is off, plainly. Too long, wrong order, missed the thing that mattered, sounds like a stranger.
• Ask it what it had to guess at, because that tells you which instruction is missing.
• Fix the instruction, not the output. Rewriting the draft by hand fixes today. Fixing the instruction fixes every time after.
• Ask it whether the output itself could be better. Something like: forget my format for a moment, is there a better way to present this for the person reading it, and what would you add or cut? You will not always agree, and every so often it names something obvious that you stopped seeing years ago.

Two or three rounds is normal. You are done when you get something you would send with light editing.

Three examples of what this looks like

The three first workflows that show up most often in professional services firms. Each takes a different kind of input and produces the same kind of thing: a structured draft you review.

Drafting correspondence is the single most common professional use of AI, by a distance, which makes the middle one the safest first build. The update saves the most time, because you do it once per client per month.

None of these make a decision. None of them send anything. The workflow gets you to a good draft and you stay the one who says yes.

You have a process worth keeping and an output you would use. Now save it, so next time you bring only the new information.

This is the move from doing it every time to having a reusable setup. It may be called a Skill, Custom GPT, Gem, Project or something else in your tool. The name matters less than having the instructions, files, example and review step saved together.

Stay in the same chat while you build it. Do not open a new one. The conversation you just had contains everything worth saving: the steps, the corrections you made, the output you approved. Starting fresh throws all of that away and you end up describing the job again from memory, worse than you just did it.

Do not write the instructions yourself either. Ask AI to write them from the conversation, read what it produces, correct it, and only then tell it to save.

If your tool does not offer a saved feature yet, keep the approved instructions and files together and attach them manually. It still works, with one extra step.

Stay in this same conversation. Do not start over.

We have just worked out how this job gets done, and the last output is one I would actually send. Now turn what we did into something reusable.

STEP 1. Write out the instructions.
Based on this whole conversation and the output I approved, write the working instructions for this job as if for somebody who has never done it. Cover:
- what the job is and what triggers it
- what information you need from me each time
- the steps, in order, including the judgment calls I made along the way
- what the finished output must look like: length, sections, order, tone, who reads it
- what to do when something I normally give you is missing or contradicts itself: flag it and stop, rather than filling the gap
- anything I corrected you on during this conversation, written as a rule so you do not make the same mistake again

STEP 2. Tell me what is still thin.
Before I approve this, tell me which parts of these instructions you are least confident about, and what you would still have to guess at if I handed you a new case tomorrow. Be specific. I would rather fix it now.

STEP 3. Wait for me.
Do not save anything yet. Let me read the instructions and correct them.

STEP 4. Once I say I am happy, save it.
Turn the approved instructions into a reusable setup in this tool, whatever it is called here: a Skill, a Gem, a Project, a Custom GPT. Include the instructions, the example of my finished work from this conversation, and a pointer to my saved context files. Suggest a name based on the job itself.

Then tell me, in plain steps, how to run it next time and what I will need to give it.

Step 2 is the one that matters. It tells you which parts of the instructions are thin before you commit them, which is the difference between a workflow that survives a real month and one that does not.

Read what it wrote as if you were the person who has to use it next month. Where it is vague, say so and have it rewrite. Only then let it save.

What goes in the skill

• The working instructions
• One example of finished work
• A pointer to your context files, so it starts from your business
• A line telling it what to do when something is missing: flag it rather than filling the gap

Name it after the job. "Monthly client report" beats "My GPT 2" when you are hunting for it six weeks from now.

You now have two skills saved. One says who you are. One says how a particular job gets done. Every job you add from here follows the same shape, which is why the second takes half as long as the first.

Some jobs happen on a predictable day. Those can run without you remembering to start them. Plenty of jobs should not, and it is worth knowing which is which.

Where a schedule works

• A morning email drafter, running after approved messages arrive, landing as drafts ready to review (requires an email connector)
• A daily meeting summarizer from the day’s calendar meetings, ready after each meeting (requires a calendar connector)
• A month-end client update that pulls together the numbers you always end up chasing, two days before you need them
• A quarterly reminder that hands you your context files and asks what has changed

The pattern: predictable timing, and the output is a draft for you rather than something that goes anywhere on its own.

Where it does not

• Anything triggered by an event rather than a date. Proposals arrive when they arrive.
• Anything you have run fewer than three or four times. It has met one kind of month so far.
• Anything where nobody would notice if the output was wrong that week.

You stay in the loop

A schedule changes who remembers to start the job. It does not change who is responsible for the output. Every scheduled workflow should land as a draft you read, not as something already sent.

If you ever find yourself not reading one, that is the signal to turn it off rather than trust it.

How to set one up

Every tool puts this in a different place and calls it something different, and they move it around. Ask your AI tool how to schedule the skill you just built. Tell it which tool and plan you are on and it will walk you through it, and its answer will be more current than anything I could write here.

Native scheduling support varies by tool and plan. A calendar reminder to run the workflow manually is a fine fallback.

I have a reusable workflow called [name]. I use [AI tool and plan]. Tell me whether this setup supports scheduled runs. If it does, give me the current steps to schedule it for [time or trigger].

Before you suggest a setup, explain:
- what files, connectors and permissions the scheduled run can access
- where the draft result will appear
- how I will notice a failure
- how to pause or remove the schedule

The result must stay a draft for my review. If scheduling is not supported or the access is unclear, tell me how to set a calendar reminder to run it manually instead. Do not invent interface steps.

Someone on your team asked you a question this week that they had already asked before. You answered it the same way you always do. Neither of you wrote it down.

The answer already exists. It just lives only in your head, which is the single most common reason a business cannot run without its owner.

Where this module ends up

Right now you are the lookup service. Every question comes to you, including the ones you have answered eleven times, including the ones that arrive while you are on a beach.

By the end of this module your team has somewhere else to ask first, and only the genuinely new questions reach you. Each new one you answer becomes the next entry.

Two kinds of answer go into it

Some questions have one answer, every time. Those become a rule, in three parts:

• The situation. When does this apply?
• The rule. What do we do?
• The exception. When does the rule not hold?

Skip the exception and the rule breaks the first time real life does not match your example. Then the person comes back to ask you anyway, which was the thing you were trying to stop.

Some questions have an answer that depends. On the client, the size of the job, how long you have worked together, whether it is a first time or a fourth. A single rule cannot carry that, so it becomes a small grid: the conditions written out, and what to do in each case.

That grid is the more valuable of the two and the more work. It is the one your team genuinely cannot guess at, and usually the thing that follows you on holiday.

Why the manual never got written

Most owners have tried this once, usually by blocking out a weekend to write a full operations manual. Most of those weekends produce three pages, a cold coffee, and a lasting sense of guilt about the other forty.

One rule at a time is different. Small enough to finish in the time it takes to answer the question anyway, and you can tell right away whether it worked: does it stop the next person from having to ask you?

Before you write anything, go and find out what your team actually asks you. The evidence is sitting in your mailbox and your messages, and it is more accurate than your memory.

Ask most owners what their team keeps asking and you get three examples off the top of their head. Go and look properly and you find eleven, and the top three are rarely the ones they named.

Where to look

If you have connected your AI tool to your mailbox, it may be able to analyze messages and extract repeated questions and common answers. Check the connection scope and permissions first. Otherwise, paste a redacted sample of questions into the prompt below.

The last 60 to 90 days, across wherever your team actually reaches you. Email, Slack or Teams, WhatsApp. Do not skip WhatsApp because it feels informal. Informal is exactly where the interruptions live.

Search for the shapes questions take rather than topics:

• "quick question"
• "can you confirm"
• "are we ok to"
• "what should I"
• "is it fine if"
• "just checking"
• "sorry to bother"

That last one is the most useful search you will run today. People apologise when they know they are interrupting you, and those messages cluster around exactly the gaps in what has been written down.

The holiday test

Now the part a search will miss, and the part that matters most.

Think about the last time you were properly away. What still got through? What did somebody phone about, because the business could not move without an answer from you specifically?

A question that follows you onto a beach is a question nobody else can answer. Those are your highest priority items regardless of how often they come up. A question asked twice a year that stops everything is worth more than one asked weekly that costs ten minutes.

Two more worth adding to the list:

• Decisions somebody made alone while you were away, and got wrong. Nobody asked, because they did not know there was a decision to ask about.
• The weekday version. In a normal week, what comes to you that stops somebody else working until you answer?

You probably already have half of this written

Before starting from a blank page, go and look for what exists. Most businesses have more written down than the owner remembers: an onboarding doc from two years ago, a half-finished process guide, a training email you wrote once and reused, notes from the last time somebody left.

Gather what you can find and hand it over with your search results. Some will be out of date, which is fine, you correct it. Starting from a stale document beats starting from nothing, and it usually reminds you of three things you had forgotten you decide.

Sort what you find into three piles

• Answered the same way every time. These become rules.
• Answered differently depending on the situation. These become grids. More valuable, more work.
• Genuinely needed you. Leave them. A business where nobody ever asks the owner anything has usually stopped thinking.

I want to find out what my team actually asks me, so I can write down the answers once instead of repeating myself.

USE MY SAVED CONTEXT:
Use my saved setup for background on my business and team.

I am going to paste in the questions my team has asked me over the last two or three months, taken from email, Slack, WhatsApp, wherever they reach me. I have written out just the question each one was asking and left the client details behind, so these will be short and the formatting will be rough. Ignore the formatting.

Strip out any client names, personal details, or confidential specifics as you go. I only need the pattern.

When I have finished pasting, do this:

1. Group the questions by what they are actually about, not by the words used. "Which supplier for the Henderson job" and "can I use the other supplier here" are the same underlying question.

2. For each group, tell me: how many times it came up, roughly how long a full answer takes me, and whether the person seemed to be blocked while waiting.

3. Sort every group into one of three piles:
   - RULE: I almost certainly answer this the same way every time.
   - MATRIX: the answer depends on the situation, so it needs conditions rather than a single rule.
   - KEEP: this genuinely needs my judgment each time.

4. Rank the RULE and MATRIX groups by how much of my time they are costing me per month.

5. Tell me which single one you would write down first, and why.

Be direct if you think something I would call KEEP is actually a MATRIX I have not bothered to define.

I have connected this AI tool to my mailbox. Before you analyze anything, tell me what mailbox, date range and message scope you can access. Do not read or quote messages outside that scope.

Find the questions my team asks me repeatedly over the last [60 to 90 days]. Group similar questions, identify common answers, and show the evidence count for each group. Separate questions that block work or have material consequences from frequent low-impact questions. Remove names, client details and confidential information from your summary.

Return:
1. Repeated question
2. Count and date range
3. Common answer, only where the messages support it
4. Exceptions or conflicting answers
5. Whether this should become a rule, a grid or stay with me

Do not invent an answer. Flag anything I need to review.

You have your piles. Turn the top one from each into something written, starting with the easy one.

Start simple. Getting the shape right on an easy rule teaches you what a hard one needs, and the first always takes longest. The fourth takes about four minutes.

Talk this through rather than typing it. Explain the situation the way you would to somebody in their first week, including the caveats and the annoying exception. The prompt below will push you on that exception until the rule actually holds up, which is the part that decides whether anyone can use it.

I want to write down one decision rule from my business, in this format:
"When [situation], we [rule], except when [exception]."

MY CONTEXT:
[paste your Company Profile and Personal Profile, or skip this if you already have them]

STEP 1. Help me find the right one.
Ask me these, one at a time:
- What question does your team ask you more than twice in a normal week?
- Think about the last time you were on holiday. What still got through to you? What could not wait until you were back?
- While you were away, did anyone make a decision on their own and get it wrong? What was it?
- Is there a situation where you have caught yourself giving different answers to different people, and you are not sure why?

Then tell me which of my answers would save me the most interruptions if it were written down, and why.

STEP 2. Push on it.
Once I have named the situation and the rule, do not just accept it. Ask me:
- Is there a real exception to this that is not obvious yet?
- What would happen if someone on my team applied this rule literally in a case I have not thought about?
- Is any part of the wording vague enough that two people could read it differently?

STEP 3. Tighten it.
Help me reword it until someone on my team could apply it correctly without asking me first. Then write the final version in Markdown so I can save it as a .md file.

FINALLY, GIVE ME THE FILE.
Write the finished rule out as Markdown, ready to drop into a file called how-we-decide.md. Use a plain heading for the topic, then the situation, the rule and the exceptions as separate labelled parts so nobody has to infer them. Put a named owner and a last-updated date at the top, and a line at the bottom telling my team to ask me if the answer is not in there, so I can add it.

If I already have entries in this file, add to it rather than replacing it.

Then the one where the answer depends

Now the top question from your second pile. These are the ones where a single rule falls over, because the right answer genuinely changes with the situation.

What that needs is a small grid: the conditions that change your answer written out, and what to do in each case. It takes longer and it is worth more, because it is the one your team cannot guess at. Every one of these you write is a question that stops following you on holiday.

I want to turn a judgment call I make regularly into a decision matrix my team can use without me.

USE MY SAVED CONTEXT:
Use my saved setup for background on my business.

THE DECISION:
[describe one of the questions where your answer depends on the situation]

STEP 1. Find the conditions.
Interview me to work out what actually changes my answer. Ask one question at a time. Push me on things like: the size or value of the job, how long the client has been with us, whether it has happened before, timing, who is asking, what it costs us to say yes.

Do not accept vague answers. If I say "it depends on the client", ask me what specifically about the client changes it.

STEP 2. Build the matrix.
Write it as a table: the conditions down one side, what to do in each case, and a one-line reason so my team understands the thinking rather than just following instructions.

STEP 3. Find the holes.
Give me three realistic situations my matrix does not cover, and ask me what I would do in each. Add the answers.

STEP 4. Set the escalation line.
Tell me where the matrix should stop and someone should come to me instead. Every matrix needs one of these. Write it as a clear condition, not "use your judgment".

Write the final version in Markdown so I can save it as a .md file.

FINALLY, GIVE ME THE FILE.
Write the finished grid out as Markdown, ready to drop into a file called how-we-decide.md. Use a plain heading for the topic, then the situation, the rule and the exceptions as separate labelled parts so nobody has to infer them. Put a named owner and a last-updated date at the top, and a line at the bottom telling my team to ask me if the answer is not in there, so I can add it.

If I already have entries in this file, add to it rather than replacing it.

Put both in one file

One document, named after what it is for. Something like how-we-decide.md. Every rule and grid goes in it, with a named owner at the top and the date it was last touched.

It starts thin, with two entries. Then you add the next one. Within a few months it covers most of what people interrupt you about. The work is small each time and the effect builds, which is the opposite of the weekend-long manual that never got finished.

A file only you can see has solved half the problem. Your team still asks you, because they do not know the answer has been written down.

Two ways to share it, and you probably want both

Markdown is the right format for a machine to read, not for a person to open on their phone, which is why the two routes want different files.

Start with the document. It works for everybody immediately. Add the skill once you know the content is right, because there is no point making a wrong rule easier to reach.

How sharing works depends on your setup

Two situations, and they are quite different.

Then the same test: are you comfortable with everything in that file if it were not kept private? If yes, share it and get on with it. If one or two entries would bother you, leave those out. Nine rules shared beats eleven kept in your head.

Either way, this is usually three or four people rather than everyone. Start with whoever asks you the most questions.

If people are on free tiers, remember lesson 0.5. You have just written down how your business makes decisions, and anything pasted alongside it on a free tier is fair game.

Tell them properly

An announcement is not a rollout. Three things make the difference:

• Show them once, live. Ten minutes, one real question, so they see it answer something they actually asked last week.
• Give them the ground rules. What to put in, what not to, what to bring to you instead. Your Company Profile from Module 1 is worth handing over at the same time, so their answers sound like your business.
• Say what happens when it does not know. They ask you, and then you add it. That is the deal, and saying it out loud stops people treating a gap as proof the thing does not work.

Keeping it alive

Two habits, both small:

• Somebody asks you something already in there. Either they did not know it existed, or the answer was not clear enough. Both take two minutes to fix.
• Somebody asks you something new. Answer them, then add it. Answering takes ten minutes either way. Adding it takes two more and means you never answer it again.

Owners who keep this up end up with something covering most of what their team used to interrupt them about. Owners who do not end up with three rules in a file nobody remembers.

A few evenings ago you had an AI tool you used occasionally and a week that only moved when you touched it. Worth looking at what actually changed.

Before and after

That is a real change. Now the honest part.

Check it properly

Go back through what you built. For each piece, ask one question:

If someone on my team used this tomorrow, with no help from me, would it hold up?

Be strict about it. The question is not whether it is a promising start. It is whether it would hold up unsupervised, on a real case, on the day it is inconvenient.

Most owners get one or two clean yeses out of four. That is an accurate result for what this course asked you to do: four separate pieces, built alone, in single sittings, without anyone checking your work.

What breaks next, and how to fix it

Three things go wrong for almost everyone in the weeks after a build like this. Here they are, with the fix, so you are not finding out the hard way.

Notice what those three have in common. Each is a piece that learned something and had nowhere to put it. That is the difference between four things in a drawer and a system: whether the parts feed each other, and whether there is a moment in your month where that happens on purpose.

The fixes above cost you nothing and will get you considerably further than most people get alone.

None of which makes today a waste. A drawer of half-decent tools beats what was there before, which was a browser tab you kept meaning to get back to. It just means you are four weeks from finding out whether you are the sort of person who fixes the one line or closes the tab.

Two honest options from here, and one of them is free.

Keep going on your own. You have everything you need to build the next workflow and the next rule, using the same prompts. Most of the value in this course comes from doing that four or five more times. If you want to get sharper at the wording itself, prompts.valuelab.ca is free and covers exactly that.

Some people will do this. Most will not, and I say that without any judgment, because the reason is structural rather than personal. Building this alone means being your own expert, your own tester, and your own deadline, on top of running the business. The fourth workflow never gets built, not because you lost interest, but because nothing was scheduled and nobody noticed.

Or build the whole thing properly, with help. That is what the six-week program is for.

What is different when it is built with youWho I am, briefly
I am Marta, and this course is the free version of what I do at ValueLab. The whole approach is value first, AI second: I am more interested in why the work keeps reaching you in the first place than in which tool you use.

Two ways to work together. An 5-Hour Owner Sprint, a single working session mapping where your week actually goes and what to build first. Or the Owner Leverage System, six weeks building the whole thing with you.

I take on a small number of owners at a time, because it is one to one and it only works if you are willing to test things between sessions and change how you work. If you want somebody to do AI to your business while you carry on as before, I am the wrong person.

More at valuelab.ca

Four things stop owners getting from what you have now to something the business actually runs on.

And there is a pile of practice that does not fit in a free course

Which tools suit which jobs, and where each one falls down. How to structure files so AI finds the right thing instead of the nearest thing. What to do when output drifts over time, because it does. How to handle the compliance and data questions properly for your industry rather than generically. What to build second, and third, and what to leave alone. Six weeks of that, applied to your business rather than an example.

What you built, and what a complete system adds

You can build the first layer alone. The Audit and full Owner Leverage System help you pressure-test it against real cases, connect it safely to your existing tools, get your team using it and keep it accurate as the business changes.

One last thing

You read a free course to the end, which puts you in a small and slightly strange minority. Click the little star in the footer five times. It is not a lesson. Call it payment for finishing.

Who the six-week program is for

Three habits make everything else in this course work better once you have your context files and first workflow.

Talk to it, not at it

You have already seen this once, back in Module 3. I probably sound like a broken record, but it is the single best hack I have, because it is the one habit people forget to keep using once the course is over.

You talk three times faster than you type. When you talk, you ramble in useful ways. You mention the exception, the annoying client, the reason you actually do it that way. All the things you trim out when typing, and all the things that make the output good.

Typing gives AI a thin three-line brief. Talking gives it the version with the details still in it.

Use it for anything you are describing from memory: a process, a decision, an answer you have given a hundred times, an email you are about to write. Anywhere in this course that asked you to explain something you already know, that was the moment to talk instead of type.

Before you agree with anything I have just said, argue against it.

Take the position of someone who thinks this plan, decision or draft is wrong. Be specific: what would a smart, skeptical person in my industry say is the weak point? What am I assuming that might not hold? What would have to be true for this to fail?

Do not soften it and do not immediately agree with yourself afterward. Give me the strongest version of the objection, then stop.

Argue with it before you agree with it

By default, AI is agreeable. Its job is to make you happy, so it will find a way to like whatever you propose, which is exactly the wrong quality in the one advisor you are talking to most.

You do not want a cheerleader. You want something that pokes holes before a client, a bank, or a bad quarter does it for you.

Use the prompt above on anything that matters more than a first draft: a pricing decision, a hiring call, the plan you built in Module 2, or a decision rule before you hand it to your team. You do not have to agree with what comes back. You do have to hear it before you commit.

Do this now: pick one thing you built earlier in this course and fix the one thing the prompt finds.

I am about to send the attached draft to a client. You did not write it and you have no context on how it was built. Read it cold, the way the recipient will.

Tell me:
1. Anything that is factually questionable, or that you cannot verify from what is in front of you
2. Anything that reads as vague, generic, or like it was written by an AI rather than a person
3. Anything you would flag before hitting send if this were your name on it

Be blunt. I would rather hear it from you than from the client.

Get a second opinion before it goes out

AI can be confidently wrong. Lesson 0.3 already told you that. The fix costs nothing and takes two minutes: before anything genuinely high-stakes goes out, put it in front of a second, fresh set of eyes. That can be a different AI tool, or the same tool in a brand new chat with no memory of how the draft was built.

The reason a fresh chat works is that the tool that wrote the draft is invested in it. A second set of eyes has no attachment to the first answer.

Reserve this for work where being wrong actually costs you something: a client-facing number, a proposal, a decision rule your team will apply without you in the room. Do not bother running it on a routine internal note. That is what the review step in Module 3 is already for.

Do this now: next time you are about to send something genuinely awkward to get wrong, run it past a second AI or a fresh chat first. Note what it caught, if anything.