How to make AI know your business and stop starting from zero
For AI to know your business you need three things: the company context in writing, somewhere the AI reads on startup, connections to the sources where your data already lives (email, documents, spreadsheets), and the habit of operating it on real cases. Without all three, every conversation starts blank.
Written by Andrés Guevara, who runs his own business on this system. The numbers of that business are below, and so is what it costs to do this without help.
The chat has no memory of your company because nobody gave it one
A conversation with AI only sees what is inside that conversation. Open a new window and the previous one does not exist. This is not a flaw in the model: it is how it works. It happens to the best model on the market exactly as it does to the worst.
Here is the part that matters, and it is the opposite of what almost everyone assumes: the limit today is no longer how much the AI can read. Current models read the equivalent of hundreds or thousands of pages in a single conversation. The limit is that nobody ever loaded those pages.
So the problem does not get fixed by writing better prompts. A perfect prompt on top of an AI that does not know what you sell is still a stranger with a good ten-minute memory. It gets fixed by setting the context up once, somewhere it always gets read.
Loose chat versus a system with context
Same tool, same subscription, two results that look nothing alike.
| Loose chat | System with context | |
|---|---|---|
| Starting point | Zero. You explain your business every time | It already knows what you sell, to whom and how you charge |
| Where the data comes from | Whatever you paste into the chat | Your email, your documents and your sheets, live |
| Who defines the task | You, task by task | It can propose what is worth doing, because it sees the state |
| What you ask for | "Draft me an email" | "Go through the week and tell me which clients are waiting on a reply" |
| What improves over time | Nothing. Every session is the first | The context, because what you correct stays written down |
| Where it fails | On anything that depends on knowing your operation | On whatever you have not written or connected yet |
What exactly has to be built
The context in writing
Files the AI reads every time it starts: what the company sells, to whom, how it charges, who the clients and suppliers are, what was decided before and why. It is the step everybody skips because it does not look technical, and it is the one that decides the quality of everything else. A small company fits in a few pages.
The connections to your sources
Authorize access to what you already use: email, documents, calendar, spreadsheets. This is what lets the AI read the current state of the business instead of depending on what you tell it. You authorize from your own account and revoke whenever you want.
The habit of operating it
Use it on real cases, correct the context when it gets things wrong, and write down what you learn. This piece does not get installed, it gets picked up. It is also why one company's system ends up different from another's even when the method is the same.
You can do it alone. Here is what each road costs
None of this is secret and the tools are public. The difference between the two roads is not access, it is how long it takes you to work out which context matters and which of your processes to attack first.
Weeks of trial and error
The AI subscription runs somewhere between 20 and 100 dollars a month depending on the plan. The rest is your time: reading, installing, getting the context wrong, finding out in week three that you started with the wrong process. It is a legitimate road and plenty of people walk it.
One two-hour session
This is what I do, and it costs from $400. You leave with the system built on your operation and a case of yours solved inside the session. If in the first hour you see it does not apply to your business, we stop and I refund you.
// I put it this way because I would rather you arrive knowing the free option exists. If the information is enough for you, use it.
The business I am writing this from
I do not have a client case to show yet and I am not going to invent one. What I have is my own business, a bag brand selling online, running on exactly this system. These are the numbers, verifiable in the repository history:
a month I used to pay a marketing freelancer, today replaced by the system
a month in third-party tools, replaced by versions I built
automation systems running, across 652 commits of real work
I put it this bluntly for a reason: almost everyone selling this shows slides. I can show the system I use to make a living. It is the only proof I have, and it is harder than a testimonial.
// and it is also why I know where it hurts: I built it by getting it wrong first.
The number that made me take this seriously
This is not the opinion of an AI salesman. It is from the World Economic Forum's annual jobs report, which surveys employers worldwide:
- Employers expect 39% of workers' key skills to change by 2030.
- 63% of employers name the skills gap as the main barrier to transforming their business.
- 59% of the workforce will need training or reskilling by 2030. Out of every 100 workers, 11 will need it and will not get it.
- 77% of employers plan to train their people to work better alongside AI.
Source: World Economic Forum, The Future of Jobs Report 2025.
What I find relevant in those figures is not the scare, it is the 11 out of 100. Training will reach big companies first, because they have training budgets. The owner of a small business is on nobody's list. That is why this gets solved by building it yourself, or with someone who has built it before.
Is this not the same as a custom GPT or a ChatGPT Project?
It is the same problem attacked at a different level. A Project or a custom GPT stores instructions and a few files, and for a lot of things that is enough. What it does not do is read the current state of your business: it does not go into this week's email, it does not see which client has not paid, it does not cross your order sheet with your inventory. The practical difference is not how much information you store, it is whether the AI can query your live sources at the moment you ask.
Does my data have to leave my company?
The connections are authorized from your own account, on your own computer, and revoked whenever you want. Nobody else needs your credentials for this to work. What is worth deciding up front is which sources are in and which are out: that is a business decision, not a technical one, and it is easier to make before than after.
Can I do this on my own?
Yes. None of this is secret and the tools are public. What costs you is not the install, it is knowing what context to write and in what order, and working out which of your processes is worth attacking first. That part is learned by trying, and trying alone takes weeks of trial and error.
How much context does it take before it starts being useful?
Less than people think. What you sell, to whom, how you charge, and the state of your three or four main processes already changes the quality of the answers in an obvious way. The context grows later, with use, and only if you write down what you keep correcting.
What if my business information changes all the time?
That is exactly the argument for connecting the sources instead of copying and pasting. What changes every day, like email, orders or the calendar, gets queried live. What barely changes, like your business model or your pricing rules, gets written once and corrected when it needs to be.
Does it work if my business is small?
It works better. In a small company the entire context fits in a few pages and one person knows the whole operation, so there is no negotiating with five departments to write it down. Big implementations are expensive precisely because the context is scattered across departments that do not talk to each other.
~/company $ next_step
A 20-minute call, free.
You tell me your case and I tell you straight whether this is useful for you or not. If you are better off doing it alone, I will tell you that too.
Rather write first? [email protected] · Or see the formats and pricing.