Tacit knowledge capture
In most organizations, 20% of people write 80% of the documentation. Timmy and Susie always update the wiki. Jared, John, Terry, and Carl never bother because Timmy and Susie have it on lock. And even that coverage isn't enough, because the real knowledge lives in conversations, habits, and workarounds that never make it to a page.
In Part 1, we covered what tacit knowledge is and what happens when it walks out the door. Now the question is: how do you actually capture it?
The Four Capture Methods (and Why Each Falls Short)
Most organizations rely on some combination of four systems, and each one has a blind spot.
Ticketing systems like Jira or ServiceNow generate thousands of hyper-specific tasks loaded with internal jargon. They are useful for audit trails but useless for knowledge transfer. A new hire would have to read thousands of tickets just to build basic historical context.
Wikis like Confluence, SharePoint, and Notion sound great in theory. Everyone contributes. But in practice, the same Pareto split applies: a small handful of contributors carry the load while everyone else assumes it is handled. Add in access barriers, SharePoint admin gatekeeping, and the insecurity some people feel about publishing to a shared space, and you get a wiki that looks comprehensive but has massive blind spots.
Templates and process docs in Word, Excel, or Google Docs occupy the fuzzy middle. They are too specific for the wiki and too vague for the ticket. Inputs and outputs are easy to document, but the messy transformation in between is where everyone gives up. And the real problem? Version sprawl. The old version lives alongside the new version because approval cycles lag behind reality. Navigating between these documents is the tacit knowledge.
Chat messages are where the real transfer actually happens. A senior engineer explaining in Slack why a particular problem is consequential, why the fix should be durable and, ideally, anti-fragile. But those conversations are ephemeral, unsearchable at scale, and buried under noise. The most valuable institutional knowledge disappears into scroll history.
Why Speech Beats Writing
Writing is a secondary form of information transmission. Speaking is the one everyone is born with. If we already know that most people will never voluntarily write documentation, the answer is to stop demanding writing and start capturing speech.
Walking-and-talking is the oldest knowledge transfer method humans have. Low friction, high output. A person can zone out, enjoy the air, and have a conversation that surfaces process details they could never force themselves to type into Confluence.
The semi-structured interview takes this further. Periodic prompts, whether AI-generated or human-led, draw out process knowledge naturally, the way a good conversation does. The key is that talking removes the blank-page problem. People who freeze when asked to "document your process" will talk for an hour when asked the right questions.
And it is not just your top performers who matter here. Even the employee on a performance improvement plan holds valuable knowledge. Their struggle often reveals gaps in the system itself. Did they have the authority to act? Were they given guidelines? Is it a person failure, or a process failure? That distinction is critical, and it only surfaces through conversation.
What to Do with Extracted Knowledge
Capturing knowledge is not the finish line. Once you have it, you need to make it useful.
Query it. Feed extracted knowledge into a searchable system so anyone can ask questions of the combined knowledge base.
Manipulate it. Transform raw conversation into new forms: visualizations, simplified documents, relational maps that show how different processes connect.
Relationally link it. Connect knowledge across people, departments, and roles. Lindsay from Part 1 held cross-departmental understanding that no single document could reconstruct. Linking knowledge relationally is how you build that picture without depending on one person to hold it all.
Resolve discrepancies. Where two people's accounts align, you have validated process. Where they diverge, you have an investigation. Is the difference correlated with differences in outcome? If so, find out why and fix it. If not, it may simply be a personal style that does not affect results.
Keep it living. Knowledge that gets archived and forgotten is no better than knowledge that was never captured. The system must support continuous updates, not one-time extraction.
Next Up
Capturing knowledge is only half the battle. In Part 3, we will show how to turn extracted knowledge into visual models, run experiments on your processes before disrupting your team, and build AI agents that actually know what they are doing.
