Where Does My Company’s Knowledge Live, And How Hard Is It To Get Out?

Knowledge is power. We all know that. But when it comes to a company's knowledge, most of it is scattered all over the place, locked away, and inaccessible. In fact, according to multiple studies, including IDC and Gartner, 80-90% of a company's data is unstructured and lives (or better, decays) in email threads, meeting notes, call transcripts, Slack or Teams channels, SharePoint sites, PowerPoint presentations, WhatsApp messages, private AI assistant conversations, ... you get the point.

But that's not the worst of it. The largest share of what a company knows isn't written down or recorded in any system: about 42% of the knowledge an employee needs is unique to that person [Panopto 2018]. To make matters worse, unstructured data grows 61% year over year. 

This is Scattered Knowledge: the knowledge, experience, expertise, judgment, standards, and definitions a company runs on, spread across people, systems, documents, formats, and locations that makes it hard or impossible for a new employee or AI to find and use. 

I have worked for over 25 years to extract, codify, and structure unstructured, inaccessible knowledge in B2B organizations and turn it into explicit knowledge used in sales enablement, content marketing, buyer education, and more. Over the years, I have developed a methodology to identify and uncover hidden knowledge, making it accessible and useful. With TrustLeader, I now use these processes to build AI Context Layers (I call them Trust Cortex), so founder-led companies become founder-independent and scalable. 

In this article, I will walk you through the ten most common places knowledge hides within your company. By the end of the article, you will know exactly which knowledge is hiding where, how valuable that knowledge is, and how difficult it is to extract, so you can put a structured plan in place to make the valuable context available to your AI so it can sound, decide, and act authentically like you. 

The 10 Most Common Places Company Knowledge Hides

Most of the valuable knowledge, decisions, judgment, etc. (let's call it context) is unstructured and scattered across ten common locations: people's heads, conversations that were never recorded, conversations that were recorded (transcripts) but just sit there untouched, inside the processes of your top performers, emails, Slack, Teams, or WhatsApp messages, completed client work, composite documents, CRM and project management tools, and shared drives like Google Drive or SharePoint. 

Below, I estimated the share of valuable knowledge per location, starting with the largest. Then I rated each location for value density (how dense the information's value is) and the difficulty of extracting and documenting that knowledge. Interestingly, there is almost a negative correlation between volume and value. Google Drives usually hold enormous amounts of data, but their value density is low.

Another interesting story emerges when you look at where things are shifting: more and more companies are recording their meetings. So #2 (unrecorded meetings)is losing importance, which shifts into #3 (recorded meetings and transcripts). Now, Conversation Intelligence solutions (like MiaRec) and AI-native CRMs with AI Notetakers built in (like Monday CRM and Day.ai), are starting to pull it out of rank 3 and into rank 9 (CRM and Project Management tools). This matters because many companies record meetings, but almost no one does anything with the recordings afterward. These recordings become an important part of what I call the "Context Engine," which automatically generates context based on your people's actions. 

These estimates are based on my experience from TrustLeader engagements and the sources cited, not measurements.

 Let's look at each of them in depth.

1. Inside People's Heads

The most valuable knowledge hides in your key people's heads: the founder, the best salesperson, and Sarah, who always knows where to find anything. This is also where company knowledge most often resides. This is where most of a company's implicit and tacit knowledge lives.

Implicit knowledge is the knowledge that can be articulated if someone asks the right question (the real value proposition, the company's point of view on the market, the pain points in customer language), while tacit knowledge cannot be articulated even when asked (the founder's instinct about fit before any qualification criteria, the behavioral standards senior people demonstrate in difficult situations). [Polanyi; Nonaka & Takeuchi]

In other words, the exceptions:  why customers get handled differently, the warning signs that a project is about to go sideways, and why the documented process says one thing and everyone does another. In a founder-led company, that's most of the standards, definitions, and judgment themselves. 

Extracting this knowledge is extremely difficult because it's acquired almost subconsciously through lived experience and years, sometimes decades, of honing one's expertise. I often talk to the CEOs I work with about the curse of knowledge: they are so familiar with the knowledge that they assume others also know what they know about the topic. They not only can't explain why they made a judgment call a certain way (even though they instinctively know it was the right call), but they often don't even realize they made a judgment. I go into more depth on how to extract this type of knowledge in my book, Scattered to Scaled, but here's the key: implicit knowledge needs structured conversation (interviews, decision-forcing sessions). Tacit knowledge only comes out through observation and inference, reflected back until the person confirms or corrects it, and some of it never comes out at all. 

• 42 percent of job knowledge is unique to the individual [Panopto 2018].
• 92 percent of organizations fail to consistently capture knowledge from people about to retire [Deloitte / eGain 2026].
• Only 54 percent of Gen Z and 60 percent of millennials believe their team could maintain performance if a key expert left tomorrow [Deloitte Gen Z & Millennial Survey 2026].

2. The Conversations That Were Never Recorded

Imagine how much knowledge is shared in face-to-face sales conversations, team meetings, or in-person workshops that don't get recorded? Or in the informal hallway conversations, phone calls, or the twenty minutes with a client after the formal meeting ended. This is especially true for sales conversations, because that is often when objections come up, or difficult questions are asked. These conversations aren't recorded. So, the only way this knowledge gets captured is if someone writes a follow-up email, meeting notes, or a CRM note summarizing the conversation. But often, those capture the outcome, not the reasoning. The next person needs the reasoning.

While the knowledge is relatively easy to document right after the conversation, it's hard to extract over time and at scale because it exists only in the memory of the people in the room, and it decays within days. The only routes are immediate debriefs ("why did you do that?") and changing the habit so the next conversation is recorded.

Depending on your company, this type of knowledge might be considerable (if you have a lot of unrecorded, in-person meetings) or relatively small if you're fully remote and automatically record all meetings. Regardless, the share of valuable knowledge in this category is shrinking as more and more meetings are held virtually and recorded by AI Notetakers. Love them or hate them, they are very useful in capturing context. But it is also worth noting that this will never completely disappear. 

3. Meeting Recordings And Call Transcripts

Ahhh, we have all been in those meetings where there were as many AI Notetakers as there were people. We joke about them; we often find them intrusive (especially the ones that send you automated messages), but we (mostly) love them. 

This includes virtual meetings and webinars (e.g., Zoom) as well as meetings transcribed by AI Notetakers such as Fathom, Granola, and Plaud. It is the fastest-growing segment of where knowledge is hiding in your organization. According to Kolmogorov Law and Pollfish, one in three US workers has had an AI notetaker in a work meeting. Interestingly, another 22% cannot say whether they have been recorded. 

Depending on the meeting, you want to surface the decisions, changed priorities, the customer explaining in their own words what they were actually worried about, and the moment someone raised the concern that changed direction. Often, the note-taking app lets you apply a meeting template (e.g., customer call, kick-off meeting, sales call, or stand-up meeting) to surface the insights that matter to you, but the recording often goes unused. You recorded the meeting and, theoretically, captured the knowledge, but the insights are still inaccessible.

The good news is that the material is there and machine-readable. But now you'll have to detect patterns. The best way is to think about what you specifically want to surface from a certain type of conversation. For example, you could mine all sales conversations for customer questions (which can feed your editorial calendar or help you build a better self-help section on your website), sales objections and how to overcome them, and more. For example, I analyze some of my recordings for testimonial opportunities, customer milestone moments, potential problems, and churn risk. 

4. Your Top Performer's Process

Every company has a few key people who keep the place running. In the beginning, that might be only the founder. At some point, a salesperson will know exactly what to say if a prospect brings up price in the first conversation, and how to handle hard objections without flinching. These key people have a "knack" for getting things done no one else in the company can replicate. If they go on vacation or were to leave, these processes would have to be almost reinvented. 

The problem is that no one has a playbook for these processes. Your best salesperson outperforms the rest of the sales team because they tweaked the process to fit their style. It works like a charm for them. They can do it, they bring in the numbers, and because their numbers are good, they never have to explain it. 

According to a 2018 Panopto study, 42% of job knowledge is unique to the individual, and 51% comes from experience. This means each person holds a tiny volume (compared to the entire knowledge within, say, a 100-person organization), but the value density is the highest of all the items on the list. For example, how top performers qualify, sequence a deal, and answer objections is implicit and can be extracted through interviews. But the thousands of micro-calibrations they make when they are reading the room, adjusting the pitch, sensing when to push and when to listen, are tacit and have never been consciously mapped. This makes extraction very difficult. 

5. Email

According to McKinsey, your team is spending 28% of their workday managing emails. So, one can only imagine how much knowledge and context lives in your Outlook or Google Workspace inboxes: customer history, approvals, negotiated exceptions, concessions made, why scope changed, and promises someone made to a client. Email is, for the most part, where the outside world talks to your company. 

Email inboxes are personal. While organizations can monitor employee email, they don't do this to gather knowledge or understand context. So, the context stays between the sender and the receiver and will be archived when they leave. Privacy concerns and limited access make extraction difficult. Reading emails across inboxes is a governance decision before it is a tooling decision. However, with AI-native CRMs like Monday CRM, Day.ai, Ahoy, and Attio, connecting everyone's emails, calendars, note-takers, and other communication channels will become the norm, and extraction will become easy. In addition, more team and even customer communication is moving to other channels, such as Slack, Teams, and even WhatsApp.

6. Slack, Teams, WhatsApp, And Other Instant Channels

Instant communication channels are increasingly where founders and customers make real decisions, teams collaborate, and a huge amount of knowledge transfer happens. But the same feature that makes those channels so attractive (hint: instant collaboration) is exactly what keeps that knowledge from becoming accessible, persistent, and institutionalized. 

In an interview with The Wall Street Journal, BOX CEO Aaron Levie said: “Company cultures are going to have to get better at, like, writing down decisions, writing down best practices, writing down business processes.” Executives at Box, Zapier, and Salesforce are pushing internal decisions out of DMs and into channels their AI systems can reach, because context trapped in private conversations cannot be used by anyone else, while trying not to kill the candid conversations that happen in private [WSJ, Aug 2026].

Instant communication is increasingly where decisions are made and judgments formed. While it is difficult now to extract knowledge, AI within these platforms will make it easier, but it requires careful governance to preserve the platform's privacy and candor. 

7. Completed Client Work

Difficulty To Extract: Moderate
Question: What did this project teach us? Did that lesson change anything outside the heads of the people who were on it?

Think about the last project your team wrapped up. The deliverable went out, the invoice was sent, the client said thank you, and everyone moved on to the next project. But what did the team learn along the way? Which part of the approach worked better than planned? Where did the scope change, and why didn’t anyone see it coming? In most companies, the answers live in the heads of the three or four people who worked on it, and nowhere else. But if those people leave tomorrow, no one could replicate the project the same way.

Your process lives in completed client work, not in the process you documented. Good people get better with repetition, and they adjust the process slightly every time they run it. Those adjustments are improvements, and almost none of them get written down. Ask three project leads how onboarding works and you’ll get three correct, different answers. The documented process is the one nobody follows anymore.

There’s a mechanism behind this. Implicit knowledge forms when someone applies explicit knowledge (an SOP, a checklist, a framework) often enough to internalize it and stop referring to the source. At that point, they’re no longer following the process. They’re simply doing what they do. Great for them, not so great for the next person.

The good news: unlike your top performer's instincts, this knowledge is recoverable. The deliverables exist, the project files exist, retrospectives exist (where they happened), and the people are still around. Comparing how three teams ran the same type of project quickly surfaces drift, and AI is very good at that comparison once you hand it the project records and call transcripts. The differences between the runs are the knowledge.

8. Composite Documents

Difficulty To Extract: Moderate
Question (suggested): Which decisions inside your best composite document should every future version inherit, and where are they written down as decisions rather than buried in prose?

A composite document serves several purposes at once, and it’s usually created to solve one specific pain point. A new marketing agency comes on board and asks for the style guide, the brand guide, and the corporate voice. So someone on the marketing team compiles everything into one document: voice rules, company and product positioning, approved claims, examples, and sometimes visual guidelines. The sales playbook is another classic: process steps, customer definitions, and deal stories, all blended into one deck.

These documents often contain some of the company's best thinking. Someone who knew the business well assembled it by hand, which makes it a snapshot of the company’s judgment at that moment. But the judgment behind it (why this positioning, why these claims, why this scope and not another) is stored nowhere. Only the output is. So the next version starts from the last one, and the versions drift.

Composite documents also hide the problem from you, and that’s what makes them tricky. Upload the brand guide into ChatGPT or Claude and the output gets noticeably better, so everything looks fine. The trouble starts when several workflows need to pull from the same file: the proposal generator needs the customer definition, the content workflow needs the voice rules, the sales team needs the approved claims. All of it is entangled in one document, and you can’t update one without touching the others.

Finding the document is easy, because it’s one click away. The work is untangling it: separating the customer definition from the claims and the voice rules, and deciding which version of each you stand behind. AI can do a decent first split. The deciding is still yours, which is why I rate it moderate.

9. Systems Of Record: CRM And Project Management

Difficulty To Extract: Difficult (moving quickly to Easy)
Question: If you exported every closed-lost deal from the last two years, could you tell from the records why each one was lost, or would you have to ask the salesperson?

Your CRM and project management tools are the closest thing most companies have to a single record of what happened: contacts, deals, stages, tasks, dates. Around those records, the notes fields and comment threads also hold some of the reasoning: why a deal was lost, what the client cared about, which estimate was off, and by how much. The records are structured. The reasoning usually isn’t, because it ends up in a free-text box or nowhere at all. Systems of record store what happened, not why.

One of my clients got 12 to 15 salespeople to document their win/loss reasons in the CRM. Good discipline. But the reasons went into a free-text field, and in just over a year, they had accumulated more than 134 different reasons. “Competitor Win,” “Competitor Won,” and “Client chose competitor” were the same reason. Nobody had decided the categories, so everybody typed whatever came to mind. Once we categorized the reasons and changed the field type, unstructured, inaccessible knowledge became usable context. The fix was a decision, not a tool.

That’s why I rate this difficult today. Exporting the data is easy, but the yield is low because we never captured the why. That’s also why I'm moving to easy: AI-native CRMs like Monday CRM, Day.ai, and Attio now connect emails, calendars, and AI Notetakers, and automatically assemble context from calls and deals. That’s the Context Engine at work, and it’s genuinely useful.

Just remember what it assembles: the average of what happened, including every off-brand email and every discount that should never have been given. A system that describes what happened can’t tell you what should have happened. Somebody still has to decide that.

10. Shared Drives

Difficulty To Extract: Easy
Question: For your twenty most important documents, can someone who isn’t the author tell which version is current, who owns it, and when it was last checked?

What’s hiding on your shared drive? Everything. That’s the problem. SharePoint, Google Drive, Dropbox... it doesn’t matter which. A drive holds documents, not knowledge, because it can’t tell you which document is true. Search for the onboarding process, and you find Customer-Onboarding.docx, Customer-Onboarding-NEW.docx, Customer-Onboarding-FINAL-v3.docx, and Customer-Onboarding-FINAL-v3-USE-THIS.docx. No owner, no last-reviewed date, no record of what changed between versions. I’m guilty of this myself.

Documentation in scattered companies tends to fail in four ways:

• Different versions of the truth: the value proposition appears in the voice guide, the sales deck, and the website copy document, slightly different in each, and nobody can say which one is canonical because nobody ever decided.
• Static artifacts: PDFs, decks, and exports that were uploaded once, never replaced, and went stale the minute they were created.
• Silo divergence: marketing’s ideal customer profile (ICP) isn’t sales’ ICP, and the claims marketing approved aren’t the promises sales is allowed to make.
• Unidentified gaps: implicit or tacit knowledge documented nowhere, invisible until a workflow needs it.

According to a 2025 systematic review published in ScienceDirect, knowledge management systems can reinforce outdated practice because obsolete content keeps surfacing as if it were authoritative. Your AI has the same problem. Give it access to the folder, and it will treat all four onboarding documents as equally true.

It gets worse when different teams keep their own piles. A client of mine wanted to consolidate their customer proof: testimonials, quotes, reviews, and case studies. They found it in nine different forms, scattered across SharePoint sites, PDFs, Word documents, Trustpilot, and testimonials that existed only on web pages. Some belonged to marketing, some had been collected by sales and never shared, and neither team knew what the other had. Finding the one right quote for a proposal was nearly impossible, so the company’s best proof simply never got used. I rate this one easy because you have full access, the content is machine-readable, and it has the lowest value density on the list. The work isn’t extraction. It’s deciding which version is true and retiring the rest.

Start With What’s Already Captured, Then Decide What Only You Can

You now know the ten places, roughly how much valuable knowledge sits in each, and how hard each one is to get out. At this point, the instinct is to start a documentation project and ask everyone to write down what they know. Please don’t. That instinct is how the shared drive got the way it is.

My recommendation is to extract, not assemble. Almost everything you need already exists in the ten places, so nobody has to write anything new. Start where the material is already captured and machine-readable: your meeting recordings, composite documents, CRM, and shared drive. Then go to the people. Your key people, your unrecorded conversations, and your top performers’ processes need structured interviews, decision-forcing sessions, and debriefs, and what you found in step one tells you what to ask.

Then comes the part where most companies get stuck. Extraction produces a pile, and the pile is full of contradictions, overlaps, and competing versions of the same answer. The Decision Gap is the gap between having knowledge and having a decided position on it.

And the most important knowledge isn’t in any of the ten places: who you won’t serve, what you won’t sell, what claim you’ll never make. Nobody writes down the deal they didn’t chase. That can’t be extracted. It can only be decided.

One last point. The connections between all these places aren’t an eleventh hiding place. They’re the answer: a single source of truth that every workflow, tool, and person draws from without having to go find it. Knowledge in a document people have to remember to open is still Scattered Knowledge, just with a new address.

Not sure where your company stands? Take the Scattered to Scaled Assessment and see where it sits between Scattered and Scaled.

1. IDC, Data Age 2025 (2018) / Gartner: 80–90 percent of enterprise data unstructured; ~61 percent annual growth.
2. Panopto, Workplace Knowledge and Productivity Report (July 2018), 1,001 US workers. https://www.panopto.com/company/news/inefficient-knowledge-sharing-costs-large-businesses-47-million-per-year/
3. McKinsey Global Institute, "The Social Economy" (July 2012). https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-social-economy 
4. Deloitte Insights with eGain, "The $9 Trillion Knowledge Exodus" (June 2026). https://www.deloitte.com/us/en/insights/topics/talent/knowledge-management-plan.html
5. Deloitte, Global Gen Z and Millennial Survey 2026. https://www.deloitte.com/global/en/about/press-room/deloitte-2026-gen-z-and-millennial-survey.html
6. Forrester / SiriusDecisions: 60–70 percent of B2B content unused by sales; 62 percent of orgs say reps look in six or more places. https://www.forrester.com/blogs/its-not-content-its-a-lack-of-buyer-insights-thats-the-problem/
7. Kolmogorov Law / Pollfish survey (July 2026), 500 employed US adults. https://kesq.com/stacker-business-economy/2026/07/16/ai-notetakers-have-sat-in-on-1-in-3-us-workers-meetings-but-only-a-third-say-they-were-asked-first/
8. Wall Street Journal, "Why CEOs Have a Problem With Workplace DMs" (August 2026). https://www.wsj.com/business/why-ceos-have-a-problem-with-workplace-dms-bc22bf8a
9. Atlassian Teamwork Lab, State of Teams 2026. https://www.atlassian.com/blog/state-of-teams-2026
10. ScienceDirect, 2025 systematic review on KM systems preserving obsolete knowledge. https://www.sciencedirect.com/science/article/pii/S2468227625004168
11. Michael Polanyi, The Tacit Dimension; Nonaka & Takeuchi, The Knowledge-Creating Company.
12. Hannah Eisenberg, Scattered to Scaled: How CEOs Build an AI Advantage That's Authentically Theirs (Nova Polaris Press, 2026).