Your company has a second database: its leader’s head. Map it before AI.

The company’s second database

Let us run a realistic thought experiment. Take one company leader, one business and one question: what does this company rely on that AI cannot learn from its documents, yet would make its answers dangerously incomplete if missing?

Small companies often have a database nobody talks about. It is not the CRM. It is not Google Drive. It is not the accounting system. It is the company leader’s head.

It holds clients whose full story was never written down. Old decisions whose reasons only the company leader remembers. Exceptions a new employee would not know to ask about. Proposals shaped differently by five years of lessons. Warning signs the company leader senses before being able to put them into a spreadsheet.

Now someone proposes: “Let’s adopt AI.” But what exactly will that AI adopt? If the company’s real logic lives in its leader’s head, uploading documents does not give AI the company. It gives AI only the company’s shadow.

Audit before the tool

We do not begin with an IT project or a “let’s map every process” marathon that ends in a colourful diagram and the same old working day. We choose one part of the work — for example, the initial assessment of an incoming client enquiry — and ask what someone must know to make the next good move. The company leader may say: “All the essentials are documented.” The basics may indeed be documented, while what matters is scattered across conversations, habits and decisions whose reasons have not been explained in years.

Why do we never send an immediate quote to certain clients? Which old decision may still appear in the documents even though it no longer applies? Where has the company leader said yes too quickly and later paid for it in time, money or energy?

The company folder may not answer those questions. The company leader can. A memory audit does not prove that the leader knows everything; it separates their knowledge, assumptions and old decisions.

A company leader’s memory is not mystical

I dislike treating a company leader’s judgement as a mystical superpower. Sometimes it contains excellent intuition. Sometimes it contains an outdated fear, a generalisation born from one bad client or a decision nobody has reviewed for years. That is why a company leader’s memory audit must be correctable, not reverential.

The audit would not simply record “the company leader knows”. It would distinguish:

what the company leader knows as fact;
what they believe based on experience;
which decision was once right but may be outdated;
what AI must not invent;
and which real task will test whether AI understood.

That is the difference between the fantasy of a “digital twin” and an AI Partner fit for real work. AI does not need to copy the company leader. It needs to help the company carry one part of the work better.

Repeated explanations are work assets

One of the most valuable findings might be the sentence: “I have explained this to new people at least ten times.” It means important knowledge lives in the company, but not yet in a shareable working method. When the company leader explains the same thing ten times, there may be an undocumented rule, a confusing process or a decision point that AI must not take over but could help prepare.

For example, the audit might uncover a client rule like this: “When a client begins with price, it does not always mean they are a bad client. Sometimes it means they cannot yet describe the problem. The difference becomes clear with a second question.” A generic AI might write a polite reply about price. A better AI Partner would ask: “Do we have enough information before quoting to tell whether this is a price comparison or a real problem?” That is a small change with potentially large consequences in sales.

The same logic applies elsewhere: which jobs look good but consume the company, or when a phone call must come before a written reply. For AI, these may be facts, warnings, old decisions or the company leader’s hypotheses. Put them all into one file and AI may sound intelligent while doing very poor work. The first question is therefore not “Where do we connect the company’s data?” but “What kind of knowledge is this?”

A small audit framework

I would begin the company leader’s memory audit with six sections.

1. Repeated explanations. What have you had to explain to your team, partner or client too many times?

2. Critical exceptions. When does the standard rule not apply?

3. Old decisions. What was once right but may be wrong today?

4. Client judgement. Which signals do you notice before they reach a spreadsheet?

5. Decision boundary. Where may AI help prepare, but must not decide for a person?

6. Acceptance test. Which 5–10 real tasks will prove that AI understood this part of the work well enough?

The point of the audit is not to pour the company leader’s entire head, inbox and client history into a machine. Only the knowledge needed for the chosen part of the work goes onto the map. Trade secrets, employee information and client data require a reason, permission and protection; otherwise they stay out. Done honestly, the company leader may see the business from a new angle by the end of the first hour — not because AI is magical, but because the company’s tacit knowledge has finally reached the table.

Why this is an AI project

“Isn’t this just consulting?” Partly, yes. And that is a compliment. An AI project becomes weak when the technology is built before anyone understands the work it will touch. A company leader’s memory audit becomes an AI project when it produces a work card that AI can use and that people can test. In our thought experiment, it might look like this:

Part of the work: initial assessment of a client enquiry
AI role: helps distinguish price comparison from a real problem and asks for missing information
Confirmed knowledge: for example, four enquiry types and their signals
Company leader’s experience-based hypotheses: signals that may indicate a difficult client
Boundaries: AI does not issue the final price or reject a client on a person’s behalf
Acceptance test: ten old enquiries — does AI suggest the right next question?

This work card can be tested. Once it can be tested and corrected, a chatbot starts becoming a working partner.

An invitation to company leaders

If you are a company leader or run a small business, ask yourself: if you were away for two weeks, which part of the company’s real decision logic would remain invisible to everyone else? This is not about replacing you. It is about ensuring the company does not depend on how many times you still have the energy to explain the same thing. An AI Partner can start small: within one part of the work, make visible what you know, what you assume, what you no longer believe and where a human must decide.

If this resonates, write to me in one sentence: “Too much of our company lives in my head.” Add which part of the work this affects most. That is where we can begin.