Your company knows more than it can find
A new colleague joins your team. Smart, motivated, ready to go. Three weeks in, they ask you how returns work for enterprise clients. You know the answer, but only because you sat next to Lisa during a support call in November. Lisa left in January.
That answer is not in your wiki. It is not in your help centre. It exists in your head, and now you are the only person who has it.
This happens in every company, every day. Not with one piece of knowledge, but with thousands. The right answer exists somewhere in your organisation. Someone said it in a meeting. Someone typed it in a Slack thread. Someone explained it on a call with a customer. But when the next person needs it, they cannot find it. So they ask around, or they guess, or they figure it out from scratch.
The numbers are worse than you think
A 2012 McKinsey Global Institute report estimated that knowledge workers spend nearly 20% of their workweek just searching for internal information or tracking down colleagues who can help. That is roughly one full day per week spent looking for things that should be available. A 2022 Forrester study found similar numbers: 30% of a knowledge worker’s week lost to hunting for data. Not surprising when large enterprises juggle hundreds of different apps and systems.
But the real cost is not the searching. It is what happens when people stop searching and start guessing. Wrong answers to customers. Repeated mistakes. Decisions made without context. The same problem solved from scratch by three different people who did not know someone had already cracked it.
And then someone leaves. A 2018 Panopto/YouGov survey of over 1,000 US workers found that 42% of the knowledge people use in their role is unique to them and not shared with coworkers. When that person walks out, their knowledge walks out with them. In the US alone, more than 11,000 Baby Boomers turn 65 every single day during the current peak period, and that wave continues through 2029.
The wiki is not the answer
Most companies try to fix this with a wiki. Or a knowledge base. Or a shared drive with a folder structure that made sense to whoever set it up in 2019.
The problem: these tools only capture what someone deliberately sits down and writes. And in my experience, that is a small fraction of what your company actually knows. The vast majority (the processes, the workarounds, the decisions, the reasoning behind the decisions) lives in conversations that nobody transcribes, in meetings that nobody summarises properly, in email threads that disappear into archives.
A wiki is a publishing tool. It captures what people choose to document. It says nothing about the vast majority of knowledge that never makes it to a page.
I have seen companies with beautifully maintained Notion workspaces where the search still returns nothing useful for half the questions people ask. Not because the wiki is bad, but because the knowledge was never in the wiki to begin with. It was in a call with a customer, in a decision made during a standup, in a Slack thread that scrolled off the screen three weeks ago.
The real problem is the gap between what you know and what you can find
Every organisation has two knowledge layers. The first is explicit knowledge: documented procedures, help articles, policies. The stuff that lives in your systems.
The second is tacit knowledge: experience, intuition, context, judgment. The stuff that lives in people. How your best support colleague knows, from the tone of the first sentence, that a customer is about to churn. How your lead developer knows which part of the codebase is fragile. How your sales director knows that this particular prospect needs to hear the pricing last, not first.
Tacit knowledge is where the real value sits. And it is, by definition, hard to capture. You cannot ask someone to write down their intuition in a Notion page.
But you can capture the conversations where tacit knowledge surfaces. The meeting where someone explains why they made a particular decision. The support call where a colleague works through a problem nobody has documented. The Slack thread where your team debates the right approach and arrives at an answer. That is tacit knowledge becoming explicit, in real time. The question is whether anyone is paying attention.
What a knowledge system actually needs to do
The traditional approach to knowledge management is: hire someone to maintain a wiki, nag people to document things, and hope for the best. It is basically a New Year’s resolution for your company. Works for a month. Falls apart at 50 people. Completely broken at 200.
Unless you make it part of your culture. A few companies have pulled this off. Voys runs an open handbook and an internal oracle where every process, decision, and policy is documented and publicly available. GitLab built an entire company around a handbook-first approach. Their handbook runs to over 2,000 pages. Basecamp has written extensively about defaulting to written communication over meetings. These are real examples and they work. But even in these companies, there is always a gap between the documentation being in place and the ease with which you can actually find what you need.
A system that actually works needs to do three things:
Capture knowledge where it naturally appears
Not in a separate documentation workflow, but from the conversations, meetings, and interactions that are already happening. Support calls contain problem-solution pairs. Meetings contain decisions. Chat threads contain micro-decisions that never make it to any document. A good knowledge system extracts this automatically, without asking people to change how they work.
Understand what it has, and what it is missing
When users ask questions and the system cannot answer well, that is a signal. When the same question comes up five times in a month and the answer is not in the knowledge base, that is a gap. A good system detects these gaps, groups them by topic, and tells editors: here is what is missing, here is how often it comes up, here is where a new article should go.
Get better over time without requiring constant maintenance
Every question that gets answered well reinforces the system. Every gap that gets filled reduces future gaps. The detection, prioritisation, and re-indexing happen automatically. Humans decide what to write, not whether to write.
This is not a hypothetical. This is what we are building at Klai. We call it Klai Knowledge, and it is an organisational memory layer that learns from every interaction.
The 90/10 principle
But I have to be honest, because I have learned something uncomfortable while building this: you cannot fully automate knowledge management. And you should not want to.
AI can extract insights from transcripts. It can detect knowledge gaps. It can suggest taxonomy categories. It can draft articles. But at every step, about 10-15% of cases need a human to look at it. The extraction is ambiguous. The classification is uncertain. The draft misses context that only a domain expert has.
The goal is not to remove humans from knowledge management. It is to remove the 90% of drudgework so that humans can focus on the 10% that actually requires judgment. Write the article that only you can write. Review the classification that the system is unsure about. Validate the decision record before it becomes part of the organisation’s memory. Writing and checking this blog.
That 10% of human effort is what makes the system trustworthy. Without it, you get a knowledge base that confidently serves wrong answers. With it, you get one that improves every week.
What you can do today
You do not need to build or buy a knowledge system to start closing the gap between what your organisation knows and what it can find. Three things you can do this week:
Record your meetings and summarise them. Not all of them. Start with the ones where decisions happen. A 30-second summary with the decision, the reasoning, and who owns the follow-up is worth more than a 90-minute recording nobody will watch.
When someone asks you something and the answer is not documented, document it. Right then. Not later. Later never happens. Spend two minutes writing the answer where the next person can find it. This is the single highest-leverage habit in knowledge management.
Audit where your knowledge actually lives. Spend 30 minutes mapping it out. Where do new hires go when they have questions? Where do your support colleagues look things up? If the answer is “they ask Sarah,” you have a problem. Sarah is a single point of failure with a notice period.
Knowledge management is not a software category. It is an organisational capability. The best tool in the world is useless if nobody feeds it. And the simplest system in the world is powerful if it captures what people actually know.
Your company already knows enough to serve every customer, onboard every new hire, and avoid every repeated mistake. The challenge is not creating knowledge. It is making it findable.
That is what we are working on. And in this blog series, I will take you through how we think about it and what we learn while building it. From the data models to the retrieval pipelines to the uncomfortable truth that taxonomy does not manage itself.
And we put all of it out in the open. Klai is open source. Our thinking is too. AI learned from all of us. The tools we build with it should benefit all of us.
Next up in this series: why organisations lose value every time they overwrite a wiki page, and what treating knowledge as something that evolves (rather than something to be cleaned up) looks like in practice. Read why you should never delete knowledge.