A second brain for your business: getting knowledge out of people's heads

Your company manual lives in someone's head

Picture three familiar scenes. Your bookkeeper gives notice, and a week later nobody knows how she reconciled card payments. Your most senior salesperson closes a quote in two minutes because he knows which discount each customer will accept, and nobody else does. And you, the owner, answer dozens of texts a day with questions you have already answered ten times.

In all three scenes the knowledge exists. It just doesn't belong to the company. It belongs to people. While they're around, things work. When they leave, get sick, go on vacation or are simply busy, the operation stalls or starts making mistakes.

That cost has been measured. A Panopto survey of 1,001 U.S. workers at companies with more than 200 employees, released in July 2018, found that 42% of institutional knowledge is unique to the individual and that each worker loses 5.3 hours a week waiting on information from colleagues or recreating something that already existed. The numbers are from 2018 and from larger companies, but the logic hits even harder at a 15-person business: the smaller the team, the more each head matters.

This isn't a new problem. A McKinsey Global Institute report from July 2012 estimated that knowledge workers spent about 28 hours a week writing email, searching for information and coordinating internally, and that making internal knowledge searchable could cut search time by up to 35%. More than a decade later, Microsoft's May 2023 Work Trend Index, covering 31,000 people in 31 countries, found that 62% still felt they spent too much time searching for information, communicating and coordinating.

Tacit vs. explicit knowledge: the difference that matters

Japanese management scholar Ikujiro Nonaka helped popularize this distinction in “The Knowledge-Creating Company”, first published in Harvard Business Review in 1991 and reprinted in 2007. In plain terms:

• Explicit knowledge is already in words or numbers: a price list, a contract, a written checklist. You can copy it and send it.
• Tacit knowledge is what someone knows how to do but never wrote down: how to calm an angry customer, the gut feeling that a quote will lose money, the right order for calling suppliers on Friday.

The article's classic example is a Matsushita team that couldn't get a bread machine to knead dough properly. The fix came when a developer trained with an experienced baker, watched how he worked the dough and turned that into a specification. The knowledge was there all along. Someone had to turn it into something others could use.

Your company works the same way. Most of what makes it run is tacit. Building memory isn't about inventing new knowledge. It's about getting the tacit out of people's heads and making it explicit, somewhere the team can find it when they need it.

From a personal second brain to a business second brain

The term “second brain” comes from Tiago Forte, author of Building a Second Brain (2022). He defines it as an external, centralized, digital repository for the things you learn and the sources they came from. His method has four steps: capture, organize, distill and express. It was built for individuals: get out of your head what you don't need to keep remembering, so your head is free to think.

Bringing it into a company changes the scale, not the principle. Instead of one person's notes, you store everyone's processes, decisions, customers and lessons. And instead of a folder only the owner opens, memory becomes something the team checks without interrupting anyone.

What goes into company memory

Don't start by trying to save everything. Start with whatever generates the most questions, mistakes or rework. In my view, six kinds of knowledge cover most of the dependency.

1. Processes

How each recurring task gets done: entering an order, sending an invoice, booking a visit, closing out the day. It doesn't need to be pretty. A ten-line checklist with who does it, when, and where each file lives beats a perfect manual that never gets finished.

2. Decisions and the why behind them

This is the part almost nobody saves and the one people miss most. “We stopped serving the north side” is a decision. “We stopped because delivery costs ate the margin and delayed other orders” is memory. Without the why, the next person either undoes the decision or follows it long after it stopped making sense.

3. Customers

Who they are, what they bought, what was agreed, how they like to be served, what they've complained about. When that lives only in a salesperson's phone, the customer belongs to the salesperson, not the company.

4. Pricing and sales rules

Price list, minimum margins, when a discount is allowed and how much, who approves exceptions. This is where “pricing from memory” becomes a rule anyone on the team applies the same way.

5. Standard answers

The 20 questions customers and employees ask most, with the answer the company considers correct. Delivery times, return policy, required documents, business hours. This takes the owner out of the help-desk role.

6. Lessons learned

What went wrong and what changed because of it. A supplier who was late, a campaign that didn't sell, a customer who didn't pay. Every lesson on record is a mistake that doesn't need to happen again.

Find out how much of your company still depends on you.Take the free index

How AI changes things

Wikis and shared drives have been around for decades. They usually die for the same reason: writing things down takes effort, and finding them takes even more. AI tackles both sides.

• Lighter capture. Instead of asking someone to write a manual, they record a voice memo explaining how they do it, forward an email thread or upload a spreadsheet. AI transcribes, organizes and drafts the process for that person to review.
• Plain-language search. Nobody needs to know which folder holds the file. People ask the way they'd ask a coworker, “what's the maximum discount for a repeat customer?”, and get the answer.
• Answers with sources. Good company memory doesn't answer from thin air. It shows where the answer came from: the document, the decision, the date. The team can double-check, and the owner can see whether the rule is current.

The technique behind this has a name: RAG, short for retrieval-augmented generation. In one sentence: before answering, the AI searches the company's documents for the passages on that topic and writes its answer from them instead of making things up. The approach was described in a 2020 paper by researchers at Facebook AI and several universities, and it now underpins most assistants that answer from a company's own documents.

The practical takeaway: answer quality depends on memory quality. If the price list is out of date, AI will give the wrong price, just faster. That's why the cycle below includes a step many people skip: review.

That principle sits at the Core of the Business AIOS: the company's memory, rules and data in one place, feeding every area and every AI agent. For the full structure, read what an AIOS is.

What to leave out (or include only with controls)

Company memory isn't a dumping ground. Some information needs care, and privacy law is explicit about it.

Brazil's General Data Protection Law (Law 13,709/2018) treats data on racial or ethnic origin, religion, political opinion, union membership, health, sex life, genetics and biometrics as sensitive personal data, and it sets a necessity principle: process only the minimum needed for the purpose. U.S. businesses face their own patchwork of federal and state rules, but the same common sense applies. In practice:

• Health data about patients or employees (common in clinics and HR) goes in only with access controls, a clear purpose and a legal basis. When in doubt, talk to whoever handles your compliance first.
• Passwords, API keys and bank details stay out of general memory. They belong in a password manager or their own system.
• Salaries, reviews and personal matters stay in a restricted area, not in the knowledge base the whole team searches.
• Customer data goes in only as far as you need to serve people well. Don't keep ID numbers, addresses or history you have no use for.

The general rule: each person sees what their job requires. Well-built memory has access levels, logs who looked at what, and lets you delete what shouldn't be there anymore. That protects customers, employees and the company.

Who writes and who reviews the memory

A company memory that nobody looks after ages in months. The simplest rule is to separate two roles. Whoever does the work records it, because they know what really happens. A responsible person reviews it, so the record uses the same vocabulary and does not contradict itself.

The philosopher Michael Polanyi summed up the challenge in 1966: we know more than we can tell. That is why recording works best when it starts from a short conversation and a concrete example, not from a blank form. Asking “how did you decide this last time?” yields more than asking “describe the process”.

• Record right after doing. What is written the same week is more faithful than what is reconstructed months later.
• Mark the date and the responsible person. Information without a date cannot be known to still hold.
• Review by use, not by calendar. What the team consults often is what most needs to be right.

A one-week plan to get started

You don't need a new tool for the first step. You need one week of attention. This plan works with what you already have: a phone to record voice memos and a shared folder.

At the end of the week, run a test: next time someone asks something that's already in the folder, reply with the link instead of the answer. It seems small. It's how a team learns to check memory before checking with you.

Next step

Before picking a tool, it helps to know the size of the problem. The Dependency Index (IDD) is a free quiz that scores, from 0 to 100, how much your company depends on you. You'll find it on the contact page, and it takes a few minutes.

If you're still deciding where to begin, read where to start with AI in your company. The logic is the same as here: understand how the company works first, then choose the technology.

Cesar Augusto Carlos

Founder, NoOne Systems

Has been building several businesses with digital marketing since 2008. Today he designs systems that bring data, processes and AI together so owners can see the whole company and decide what matters.