Build an AI-ready business: give your AI a brain

You ask an AI assistant something about your own business. It fires back a fast, confident, well-written answer. And the answer is politely useless. Not wrong exactly, just generic, the sort of thing you would write for “a business” rather than yours. That moment is where most managers give up on AI. This course is about the step they miss.

It is blank, not stupid

The mistake is to read that generic answer as the AI being thick. It is the opposite. Think of a brand-new AI chat as the sharpest temp you have ever hired, on their first morning, who has read almost everything ever published but has never heard of you.

They can write a cracking email and summarise a contract in seconds. What they cannot do is know a single fact about your firm. Not your payment terms, not your lead times, not that you never take on contracts below a certain size, not that your tone with customers is warm and plain rather than corporate. So when you ask it to “draft a reply to this customer complaint about a late delivery”, it has no choice but to invent what you will offer, guess your policy on refunds, and sound sure of itself while doing it. Confident, fluent, and detached from your reality.

Before and after

Same assistant, with and without your business

Nothing about the AI changes. What it can see does.

Before

A blank chat

• Invents a refund policy that sounds plausible
• Guesses a lead time and sounds sure
• Writes like a call centre
• Forgets everything tomorrow

Write the business down once

What changes: Write the business down once.

After

With your business in view

1. 01Reads out the refund policy you wrote
2. 02States the lead time your team works to
3. 03Writes in the tone of your last good email
4. 04Starts next month with the same facts

Why it forgets between chats

There are two layers to the forgetting, and they are worth separating.

The first is that it starts blank. Out of the box it knows the world in general and nothing about you in particular.

The second is that, on its own, most AI forgets everything the moment you close the chat. You can spend ten minutes explaining your business today, get a great answer, come back tomorrow, and it is the blank temp on their first morning all over again. Nothing you taught it stuck. That is not a fault you did something wrong to trigger. It is simply how a plain chat works: each conversation is an island.

The gap that everything hinges on

That distance, between what the AI knows about the world and what it knows about you, is the context problem. It is the single reason AI feels like a party trick for so many managers and a proper tool for a few. The ones who get real value are not using cleverer AI. They have closed that gap, so the tool can see their business while it works.

Closing it is not technical and it is not expensive. It is mostly a decision to write your business down once, in a form the AI can read, and to store it somewhere the tool can reach again and again. That is the whole job of this course, and we go into what “a brain” means next. If you want the fuller background first, our guide Why your AI needs a brain covers the context problem in more depth.

“Giving your AI a brain” sounds like something from a film. It is far more boring and far more useful than that. Two plain words do all the work here: context and memory. Get those straight and the rest of the course is just doing the thing.

Context: what it can see right now

Context is simply the information you put in front of the AI before it answers.

Paste your actual returns policy into a chat and ask it to draft a reply to a customer who wants their money back. The generic guess vanishes, because now it has your rules and uses them. That is context: the facts the model can see while it works. No context, and it fills the gaps with plausible fiction. Good context, and it works from the truth. You have probably done this by accident already, any time you pasted an email or a document in before asking for help. That was you handing it context.

Memory: context that sticks

The trouble with pasting is that it does not last. Close the chat and it is gone, so tomorrow you are explaining yourself from scratch again.

Memory is context that stays put, so you do not have to re-feed the same facts every single time. Your business facts are stored somewhere the AI can reach them again and again, without you re-explaining who you are on every job. That is the difference between a one-off good answer and a tool that reliably knows the place.

Put them together and you have a brain

That is the whole idea. A business brain is context plus memory: your business written down (context), stored so it persists (memory), in a form the AI can read before it answers.

How it is built

Context on top of memory makes a brain

Two plain words do all the work. One is what the AI sees now; the other is where it lives.

1. 03

   What you get

   An answer from your reality

   A reply that uses your policy, your terms and your tone, not a plausible guess.

   The AI, with a brainChat

   Received: Can this customer return a part-used order?

   Sent: Within 28 days, unused only. Source: Returns policy.

2. 02

   Right now

   Context

   The information in front of the AI before it answers. Paste it in and the guessing stops.

3. 01

   Stays put

   Memory

   The same facts kept somewhere the AI can reach every time, so you never re-explain.

   Document

   Returns policy

   Unused items: returnable within 28 days.

   Refund to the original payment method.

   Owner: customer service. Checked 1 October.

Stands on Your business, written down once

It is not a robot consciousness, and it is not a smarter or pricier model. A cheaper AI with a good brain will run rings around an expensive one that is guessing blind, because the deciding factor is not raw cleverness, it is whether the tool can see your business. Keep that framing. Everything else in this course, from what to write down to how to store it, is just building and looking after that one thing.

Next we get specific about the four kinds of facts worth writing down first.

A brain is mostly you deciding, once, to write down things you already know. The good news is that you do not need to write down everything. Most of the value comes from four kinds of fact. Get these on paper clearly and you have a brain worth feeding to an AI. Everything else is a bonus.

1. Your prices and packages

This is the one that stops the AI inventing figures, so start here. Write down what you charge and how you scope work: your rate card, your packages, what is included, what costs extra, payment terms, and the terms you hold with your main suppliers.

If your pricing depends on the case, write down the rule rather than pretending there is a fixed number. “Annual contracts get 30 days’ notice of any price change; one-off orders are priced at the rate card on the day” is a real, usable fact. A made-up flat price is not.

2. Your policies

The way you run the things that repeat. How a new starter’s first week works, who approves what, how a customer complaint is handled, your refund and cancellation rules, lead times, what you do and do not take on. These are the questions where a confident guess can cost you money or a customer, so they are exactly the ones worth pinning down.

3. Your voice

A few real examples of how you write to customers and colleagues. Paste in two or three of your own emails, good ones, and note what you want kept: warm and plain, direct, no jargon, sign off a certain way. This is what stops drafts coming out sounding like a call centre and gets them sounding like you.

4. Your common questions

The things people ask every week, with your actual answers. “Do you take on small contracts?” “How long until a new starter has their logins?” “Do you pay suppliers early?” Ten of these, answered the way you really answer them, covers a surprising amount of your day-to-day.

Taken apart

The first page of a business brain

Four kinds of fact, each written as you would brief a sharp new hire.

Brain: how we run things

1Annual contracts get 30 days' notice of any price change. One-off orders are priced at the rate card on the day.

2A new starter gets a laptop and logins on day one. Their manager books a first-week plan before they arrive.

3Warm and plain. First names. Short sentences. Sign off with the sender's first name and 'Customer Care'.

4Do you pay suppliers early? Standard terms are 30 days. We can go to 14 for a 2 per cent discount.

1. 1Prices, as a rule

   When pricing depends on the case, write the rule, not a made-up flat number.

2. 2A policy

   How you run onboarding, complaints or approvals, with the reason where it helps.

3. 3Your voice

   Two or three real emails sit beside this line so the AI can copy the sound.

4. 4A common question

   The question as people ask it, with the answer you really give.

Write it like you would brief a new hire

However you capture these, write them the way you would explain them to a sharp new employee on day one: plain, direct, with a line of reasoning where it helps. Vague inputs give vague outputs, every time. “We’re a bit flexible on terms” tells the AI nothing. “We hold our price for 30 days, then reprice” tells it exactly what to say.

You do not have to do all four today. Even the prices section alone, written clearly, will change the quality of what you get back. Next we look at how the AI reads a brain once it grows past what you can paste by hand.

At some point someone will tell you that you need “RAG”. It is one of those three-letter acronyms dropped into conversation to make a simple thing sound clever. It stands for “retrieval-augmented generation”, which is exactly the sort of phrase that puts managers off. Let us bin the acronym and describe what it really is, because the idea underneath is simple and worth knowing.

The one-sentence version

RAG means the AI looks things up in your documents before it answers, instead of answering from memory.

That is it. When a tool “uses RAG”, it has been wired so that, the moment you ask a question, it first goes and finds the relevant bits of your own files, reads them, then writes its answer based on what it just read. It is the difference between someone answering off the top of their head and someone saying “hang on, let me check the file” and coming back with the right page open.

How it moves

Look it up first, then answer

What a tool 'using RAG' does between your question and its reply.

Comes inA question from your team

1. 01

   You ask

   A plain question, the way a colleague would ask it.

   HR questionChat

   Received: How much notice do we give if a start date moves?

   Sent: Checking your files before I answer.

2. 02

   It searches your files

   It finds the relevant passages in your own documents.

   Search resultsScreen

   Onboarding policy

   Section 3, start dates

   Offer letter template

   Clause 2

   Holiday policy

   not relevant

3. 03

   It reads what it found

   The answer is built from those passages, not from memory.

4. 04

   It answers and shows the page

   You can open the source and check it in a glance.

   A person decides here

   Document

   Answer

   Give at least five working days' notice in writing.

   Source: Onboarding policy, section 3.

Goes outAn answer with its source

Gilt ring: a person makes the call at this step

Why “read first” changes everything

Remember the core weakness: on its own, AI states things with total confidence whether or not they are true. Ask it about your returns policy and it will happily invent one. RAG is the main fix. When the tool fetches your real documents first, its answer stops being a guess and becomes closer to a citation. It is reading your policy and telling the customer what it says.

Two things follow. First, the making-up drops sharply, because the AI is anchored to your text and has far less room to wander off. Second, and this matters more than people expect, you can check its work: a good RAG setup shows you which document it pulled the answer from, so you can glance at the source instead of trusting a black box.

When you need it

Here is the part that saves you money, because not every business needs to rush out and buy this.

If your business knowledge is small enough to paste into a chat by hand, you may not need RAG yet at all. Dropping your one-page onboarding policy into a conversation and asking for a reply does the same job manually, and it works perfectly well when the knowledge is small and you are the only one using it.

You start to need RAG when one of these is true:

• There is too much to paste. Once your knowledge is dozens of documents, hundreds of past contracts or years of policies, you cannot keep hand-feeding it.
• The answer must be exactly right, every time. Customer replies, supplier terms, HR and policy questions. Anywhere a confident guess could cost you money or trust.
• Other people need it, not only you. The moment your team asks the AI questions, you cannot be the one pasting context each time.
• Your information changes. Update the document and the answers update. No retraining, no starting again.

If none of those apply, do not let anyone sell you a complicated setup. Start with pasting, get comfortable, and reach for RAG when you feel the pain of doing it by hand. That order saves money and teaches you what you need. Our guide RAG without the jargon walks through the same idea with a couple more examples if you want them.

Next: the tool-agnostic way to build your first brain this week, without buying anything.

Here is the part where a lot of courses would send you shopping for software. We are not going to. The first version of a business brain is not a system you buy, it is a document you write, and you can have a working one by the end of the week. Fancier storage comes later, and only if you need it.

Step 1: open one document

Anywhere you already work is fine. A Google Doc, a Word file, a note. One document, one place. Do not overthink the container. The value is in what goes inside, not the software around it.

How it unfolds

A first brain in one working week

Small daily steps, no software to buy, and a working version by Friday.

1. Monday

   Open one document

   A Google Doc, a Word file or a note. One place, no tidying.

2. Tuesday

   Write the four sections

   Prices, policies, voice and common questions, half a page each.

   Brain: sections2 of 4

   • Done: Prices and packages
   • Done: Policies
   • To do: Voice: add two real emails
   • To do: Common questions: ten of them

3. Wednesday

   Paste it into a real job

   Draft a complaint reply or a supplier response with the document in front of the AI.

   Reply to a complaintChat

   Sent: Here is our brain document. Draft a reply to this complaint.

   Sent: Uses your refund rule and your tone.

4. Friday

   Note what you reached for

   Which facts you used all week, and which you wrote and never touched.

Step 2: write the four sections

Give it four headings, the ones from lesson three:

• Prices and packages. What you charge, how you scope, payment terms and supplier terms.
• Policies. Onboarding, approvals, lead times, what you take on, complaints, cancellations.
• Voice. Two or three real customer emails of yours, plus a line on how you want to sound.
• Common questions. Your ten most-asked, with your real answers.

Do not aim for perfect or complete. Aim for true and clear. Half a page per section, written the way you would brief a new hire, beats ten pages of waffle. You are writing for a sharp reader who knows nothing about you yet.

Step 3: put it in front of the AI

This is the step that makes it a brain rather than a document. When you next ask an AI to draft a complaint reply, a supplier response or a message to a new starter, paste the relevant section, or the whole thing if it fits, into the chat first, then ask your question. Watch the difference. The generic guess is replaced by something that uses your policies, your terms and your voice.

That paste is you giving the tool context by hand. It is not clever and it is not permanent, but it is the real thing working, today, for free.

Why start manual

You might wonder why we are not wiring this into something automatic straight away. Two reasons, both learned the hard way.

First, doing it by hand for a week or two teaches you what your brain needs. You find out which facts you reach for constantly and which you wrote down and never used. That is priceless before you spend on anything.

Second, the manual version is useful on its own. Plenty of teams never need more than a well-kept document they paste from. Only when the pasting starts to hurt, too much to paste, other people needing it, do you graduate to the setup we cover later. Start here. Feel the value. Then decide.

Next, a short but important lesson: what is safe to put in a brain, and what should never go near one.

A brain is only as safe as what you put in it, and this is the lesson where a bit of care saves you real trouble. The rules here are simple and they are not scary. This is general guidance to keep you sensible, not legal advice, and if you handle a lot of personal data it is worth a proper look at your obligations.

What belongs in a brain

The good news: the things that make a brain useful are also the things that are safe. General, reusable business facts.

• Your prices, packages and terms.
• Your policies and how you run an order or a new starter’s first week.
• Your voice and example messages, with any customer’s personal details stripped out.
• Your public FAQs and standard answers.

None of that is secret or personal. It is the stuff you would happily tell any customer who asked. That is a good rough test for what belongs.

What never goes in

Some things should stay well out of a brain, no matter how convenient it seems:

• Passwords, card details and bank details. A brain is a reference the AI reads, not a vault. These do not belong in one, ever.
• Customers’ personal data in bulk. A spreadsheet of names, addresses, phone numbers or anything sensitive. Do not pour your customer database in to “save typing”. That is exactly the kind of data you have a duty to look after.
• Anything confidential to someone else. A supplier’s private pricing, another firm’s documents, staff records. If it is not yours to share, it is not yours to feed in.

The one-line rule

When in doubt, leave it out.

If you would hesitate to email a fact to a stranger, do not put it in a brain. Nearly every safe decision falls out of that single question, and it keeps you on the right side of the line without needing to memorise regulations.

Before and after

Keep the rule, drop the record

For each thing you are tempted to paste, there is a safer version that does the same job.

Before

Tempting to put in

• Every customer's name and order history
• Card numbers for refunds
• Last year's signed supplier contract
• Sarah's sickness record for the policy
• The accounts login for the AI to use

Keep the rule, drop the record

What changes: Keep the rule, drop the record.

After

Put in instead

1. 01The complaints policy and example replies, names removed
2. 02The refund policy and its time limits
3. 03Your standard terms and the clauses you always ask for
4. 04The sickness absence policy, no names
5. 05Nothing: the AI does not need a login

A word on where it lives

If you stay with the manual, paste-from-a-document version from the last lesson, your brain sits in your own files, on your own account, which keeps things simple. When you move to a more automatic setup later, the same care applies: know where the data is stored, who can reach it, and keep anything personal out of the general reference layer. We come back to how we handle this in our own setup in the next lesson.

Next: how we run our own brain, including the part nobody warns you about, keeping it true once your business moves on.

Most guides on this topic are written from the outside, by people describing a thing they do not run. This one is not. We run a persistent context system of our own, a second brain for the business, and it sits behind the tools we build. So rather than tell you what a brain is in theory, here is what it is like to run one, warts and all.

Two notes first. This is our own setup and how we use it, not a product we sell you. It is also one of several: one of our knowledge bases holds 4,535 notes, 5.9 million words and 1,023 reusable skills. And nothing here needs a big budget or a technical team. The principles are identical whether your brain is a polished system or a well-kept folder of documents. The value is in the discipline, not the software.

What it is

Strip away the name and ours is a single, organised, always-available store of everything the AI needs to answer and act as our business would. The same four kinds of fact you have been writing down: our prices and how we scope work, our policies and how we run a job from enquiry to sign-off, the way we write to people, and the questions that come up again and again with the answers we give.

When we ask the AI to draft something or help with a task, it works from that, not from a blank slate. It is the difference between briefing a colleague who knows the business and briefing a stranger every morning. The “persistent” part is the point: it does not forget between conversations. What we taught it last month is still there this month.

What we learned putting things in

The first lesson was about what belongs and what does not. Good context is specific and stable: prices, policies, product details, tone, standard answers. Things true across many jobs that do not change by the hour. The temptation early on is to pour everything in, on the theory that more must be better. It is not. A brain stuffed with one-off details and half-finished thoughts that contradict each other makes answers worse, because the AI cannot tell what still matters. We got more value from a smaller, cleaner brain than a bigger, messier one.

The second lesson: write it the way you would explain it to a sharp new hire. The clearer we wrote the source, the better the AI performed, every single time.

Side by side

A bigger, messier brain against a smaller, cleaner one

What we learned the hard way: more going in does not mean better answers coming out.

	Pour everything inMore must be better	Keep it specific and stableFewer, true facts
What it looks like	Document Pricing notes (draft 3) Rush rate 20 per cent. Or 25? Check. Ignore the above, see Dan's email. Old tiers: keep for now.	Document Pricing policy Rush jobs: 25 per cent on top. Deposit: 30 per cent on order. Owner: operations. Checked 12 Oct.
What it holds	Against it: One-off details, half-finished thoughts	In its favour: Prices, policies, tone, standard answers
Contradictions	Against it: Common, and the AI cannot tell which wins	In its favour: Rare, because each fact has one home
Answer quality	Against it: Worse as it grows	In its favour: Better and steadier
Upkeep	Against it: Nobody knows what is stale	In its favour: One owner, a date on each note
The verdict	More material, worse answers	Smaller and cleaner wins

in its favour mixed against it

The part nobody warns you about

If there is one thing to take from how we run ours, it is this. Setting a brain up is the easy, satisfying part. Keeping it true is the real job, and it never fully ends.

A business changes. Prices move, policies get revised, you drop one service and add another. Every one of those has to make it back into the brain, or the AI keeps confidently answering with yesterday’s facts. And a wrong answer delivered with total confidence is worse than no answer, because you might not catch it before a customer does. A stale brain is a real liability, not a neutral one. We treat maintenance as a standing habit, not a project with an end date. We go deeper on exactly how in the next lesson.

Why we bother

Once the context was solid, new tools could draw on the same source instead of being given the business background from scratch. That applies to the content line, outreach drafting and smaller helpers alike. The practical difference is structural: the source is maintained once and reused wherever it is relevant.

Later tools can inherit the same shared understanding of prices, policies and voice. A good brain is more than a better chat assistant, it becomes a common foundation for the rest of the AI work. That is why we use this structure. If you want this told as a standalone piece, it is written up in How we run our own AI brain.

We said it in the last lesson and it is worth its own: setting a brain up is easy, keeping it true is the real work. This lesson is the practical side of that. The aim is not to make maintenance a chore you dread, it is to make it small, routine and mostly automatic in your own head, so the brain stays trustworthy without eating your week.

Why a stale brain is worse than none

It is tempting to think an out-of-date brain is at least no worse than nothing. It is worse. With no brain, you know the AI is guessing and you check everything. With a confidently wrong brain, the AI states last year’s terms or a dropped policy with total certainty, and because it sounds right, you are far more likely to let it through to a customer. Trust in a stale source is exactly how the mistakes escape. So freshness is not tidiness, it is the thing that keeps the whole system safe to rely on.

Make updating part of the change

The single habit that matters most: when something changes in the business, updating the brain is part of making that change, not an afterthought.

How the loop runs

How a brain stays true

Two habits keep it current: update it with the change, and read it once a month.

1234

Something changes

Update it in the same sitting

Read it monthly, ten minutes

Fix what drifted

The brain

Always matches how you work today

1. 1

   Something changes

   A price moves, a policy is revised, you drop a service.

2. 2

   Update it in the same sitting

   Edit the brain as part of making the change, not as an afterthought.

   brain-changes.xlsxSheet

   NoteWhy it changed

   Complaints policyRefund window now 28 days

   Matched: Every answer now uses 28 daysMatched
3. 3

   Read it monthly, ten minutes

   As a new customer would: right prices, dropped services, new questions.

   Monthly check2 of 4

   • Done: Prices still right
   • Done: Nothing here we no longer do
   • To do: Policy changes we forgot
   • To do: A new common question to add
4. 4

   Fix what drifted

   Correct the slow drift before it reaches a customer.

The loop ends when the business stops changing, which it never does

You raise your prices, you update the prices section in the same sitting. You drop a service, you remove it from the brain before you move on. You revise the complaints policy, the new wording goes in there and then. Done this way, the brain never drifts far, because it is corrected the moment reality moves. It is the afterthought version, “I’ll update it later”, that lets a brain rot, because later never comes.

A light regular check

On top of that, a quick standing review keeps the slow drift in check. Once a month, or whatever fits your rhythm, spend ten minutes reading the brain as if you were a new customer:

• Are the prices still right?
• Is anything in here that you no longer do?
• Has a policy changed that you forgot to update?
• Is there a new common question that deserves an answer?

That is it. Ten minutes catches the things that slipped through, and it is far cheaper than one wrong figure sent to a customer in good faith.

The upside is real

Here is the reward for the discipline: maintenance is cheap once it is a habit, and it pays off everywhere at once. You change one document and every future answer changes with it. No retraining, no rebuild. Whether you paste from a document or run a full system, keeping the source of truth current means every tool that reads it simply stays correct. You do the small thing once, and correctness flows out to everything downstream.

Keep it true and the brain earns its trust. Let it rot and you will stop relying on it, which wastes the whole effort. In the last lesson we pull it together into a real system, and point you at what to do next.

You have the pieces now. You know why AI forgets your business, what a brain is, what to write down, how the AI reads it, how to start this week, how to feed it safely, how we run ours, and how to keep it true. This last lesson is about the arc from here: how a humble pasted document becomes something real, and what to do next.

The path is one direction

You do not decide up front how sophisticated your brain needs to be. You let the pain tell you. The path runs one way and you stop wherever it stops hurting:

1. One document you paste from. Free, manual, useful. Plenty of teams never need more.
2. A tidy, organised set of documents as it grows, still pasted or dropped in per job.
3. A setup that looks things up for you (the RAG idea from lesson four), for when there is too much to paste, other people need it, or answers must be exactly right every time.
4. Tools built on top of the brain, where complaint replies, onboarding or reporting draw on the same context automatically.

How it unfolds

Four steps up, each taken only when it hurts

You do not choose your brain's size up front. The pain tells you when to climb.

1. Step 1

   One document you paste from

   Free, manual and useful. Plenty of teams never need more.

   Document

   Brain: how we run things

   Returns: unused within 28 days.

   New starters: laptop and logins on day one.

   Suppliers: standard terms 30 days.

2. Step 2

   An organised set of documents

   Policies, terms and examples split by topic, still pasted per job.

3. Step 3

   A setup that looks things up

   For when there is too much to paste or others need it. Answers cite their source.

   Ask the brainScreen

   Question

   Notice for a moved start date?

   Answer

   Ok: Five working days, in writing

   Source

   Onboarding policy, section 3

4. Step 4

   Tools built on top

   Complaint replies, onboarding and reporting all draw on the same context.

Each step is only worth taking when the one before it starts to hurt. Do not leap to step four because it sounds impressive. Earn your way up.

What it gives you

The value is structural. Once policies, terms, voice and common answers are written down, each new tool can start from the same approved context instead of rebuilding it from memory. The benefit still has to be measured in the real workflow, with review time and maintenance included.

That is the shape of the whole thing. The first tool is the hard one. Once the brain exists, everything you build after it starts from a running position, because it already knows the business.

Your next steps

You do not need anything fancy to get the benefit. Start with the truth and start small:

• This week: open one document and write the four sections, prices, policies, voice, common questions. Paste it in front of the AI next time you draft anything.
• Keep it true: update it the moment something changes, and give it ten minutes once a month.
• Move up only when it hurts: reach for a lookup setup or built-on tools when pasting stops being enough.

That, kept current, is a business brain. Everything fancier is just a better way of storing and retrieving the same thing.

Take it further

To get going today, grab the AI Readiness Checklist from our free resources. It is the one-page check for what to write down first, and the fastest way to turn this course into action.

If you would rather bring the process you suspect is costing the most to us, a free 30-minute fit call tests whether it is specific enough to audit. The business-wide AI Constraint Audit is £2,000 over 14 days. If you would rather have a brain built for your own business, our departments show where we build.

To see a full-scale version, read How we run our own AI brain. One of our knowledge bases holds 4,535 notes, 5.9 million words and 1,023 reusable skills, and it is one of several brains we run. It is how we work, not a product we sell, and the evidence page shows the systems behind it.