# How to Keep Know-How When People Leave Your Company

> Staff turnover takes know-how with it. Learn how to turn experience into versioned, human-approved skills and onboard new hires from a living playbook.

Published 2026-10-19 · Updated 2026-10-19 · MOSY AI Editorial Team · https://mosyai.com/blog/keep-know-how-when-people-leave/

Knowledge retention is the practice of keeping the methods, judgment and context that experienced employees carry, so the company can still use them after those people change roles or leave. The most durable approach writes that experience down as reusable, versioned skills, approved by senior staff, that both people and AI agents can apply to everyday work.

## How much does employee turnover cost?

According to [Gallup (2019)](https://www.gallup.com/workplace/247391/fixable-problem-costs-businesses-trillion.aspx), replacing an employee costs between half and twice their annual salary. Gallup estimates that voluntary turnover costs US businesses about $1 trillion a year.

These figures cover recruiting, onboarding and lost output. They do not fully capture what leaves with the person: the shortcuts, the client history and the reasons behind decisions.

## How much time do teams lose looking for knowledge?

Knowledge loss also happens while people are still employed. According to a [Panopto and YouGov survey (2018)](https://www.prnewswire.com/news-releases/inefficient-knowledge-sharing-costs-large-businesses-47-million-per-year-300681971.html), knowledge workers lose 5.3 hours a week waiting for information or recreating knowledge that already exists. The survey put the cost for a large US business at $47 million a year.

Panopto sells video and knowledge-sharing software, so treat the exact numbers with care. The pattern will still be familiar to most managers: the same questions go to the same few people, and the answers are lost each time.

## What kind of know-how is most at risk?

Not all knowledge is equally fragile. Documents and system records usually survive turnover. The following usually do not:

- **Methods:** the sequence of steps an expert follows to produce a good proposal, audit or report.
- **Judgment:** which exceptions matter, which numbers to double-check, when to escalate.
- **Context:** the history with a client, a regulator or a supplier, and why things are done a certain way.
- **Templates and examples:** the best past outputs, which experts reuse and newcomers cannot find.

## Why do wikis and handover documents fall short?

Most companies try a wiki or an exit handover. Both tend to fail for the same reasons.

Handover notes are written in a rush during the last two weeks, by someone who is already leaving. Wikis are written once and rarely updated, because updating them is extra work with no immediate reward. Neither is used at the moment the work happens, so people keep asking colleagues.

There is a parallel with AI tools. According to [MIT NANDA (2025)](https://virtualizationreview.com/articles/2025/08/19/mit-report-finds-most-ai-business-investments-fail-reveals-genai-divide.aspx), most generative AI systems fail to deliver returns partly because they do not retain feedback or improve over time. Static documents have the same weakness.

## How do you turn experience into versioned skills?

A skill is a short, structured description of how your company does one recurring task. It works when it is written as part of the job, reviewed by someone senior, and used directly in daily work.

1. **Pick recurring, high-value tasks.** Start with work that repeats weekly or monthly and depends on a few experienced people, such as tender responses, credit checks or month-end reports.
2. **Capture the method while the work happens.** Ask the expert to do the task with an AI assistant and explain each step. The assistant drafts the skill from that session.
3. **Write it in a standard shape.** Each skill should state its purpose, inputs, steps, quality checks, common mistakes and one or two good examples, with links to source documents.
4. **Have a human approve it.** A senior person reviews and signs off before the skill enters the shared library.
5. **Version every change.** When someone finds a better way, or the agent notices repeated corrections, propose an update. Keep the history so anyone can see what changed and why.
6. **Use the skill in daily work.** People and agents run tasks through the skill itself. A skill that is used every week stays current.

| Approach | When it is written | Kept up to date | Used during work |
|---|---|---|---|
| Exit handover note | Last weeks of employment | Rarely | Rarely |
| Wiki page | Once, by a volunteer | Rarely | Sometimes |
| Versioned skill | During normal work | With every approved change | Every time the task runs |

## How do you onboard new hires from the playbook?

When skills and source documents live in one place, onboarding changes. A new hire can ask the assistant how the company prepares a quarterly client review and get the current approved method, with links to the documents behind it.

They can run the same skill a veteran uses. A reviewer checks their first outputs, and the corrections improve both the newcomer and, where relevant, the skill. The new hire learns on real work from day one, and senior staff spend their time reviewing instead of repeating explanations.

## Who should own the skill library?

A skill library needs owners, in the same way a policy manual does:

- **Function heads** decide which tasks need a skill and set priorities.
- **Senior practitioners** write and approve the skills in their area. Their name and the approval date appear on each version.
- **A knowledge lead** checks that skills follow the standard shape, link to current sources and are retired when a process changes.

Make reviewing and contributing skills part of senior staff's objectives. When sharing know-how counts toward performance, experienced people have a reason to write it down long before they move on.

## How Redstone does this

[Redstone](https://chatredstone.com) keeps know-how in two layers. Personal memory holds each user's preferences and context. A shared, versioned skill library holds company methods. When the agent finds a better way to do a task, it proposes an improvement, and a human approves it before it reaches everyone.

The knowledge base stores company documents with hybrid search and answers with citations, so new hires see where each answer comes from. Access follows four data levels (Public, Internal, Confidential and Secret), level-3 data runs only on local models hosted in Vietnam, and company data is not used for training.

In the AI Factory module, workflows are written in plain Vietnamese and each value-chain stage has an agent that acts as the assigned person, with humans reviewing the output. When a person changes roles, the workflow, the skills and the review history stay with the company. Redstone runs on-premise, works inside Zalo, Slack and Mattermost, and keeps a full audit log of every change.

[Book a briefing](/#contact) to see how Redstone can capture your team's know-how.

## FAQ

### What does a company lose when an experienced employee leaves?

Beyond hiring costs, the company loses methods, judgment and context that were never written down: how a key client prefers to work, which supplier is reliable, why a process has an extra check. Gallup estimates replacing an employee costs half to twice their annual salary, before counting this lost know-how.

### How do you capture an employee's know-how before they leave?

Capture it continuously, well before anyone resigns. Record how experienced people handle recurring tasks, write each method as a short skill with steps, checks and examples, have a senior reviewer approve it, and keep version history. When work is done through an AI assistant, corrections can become proposed skill updates.

### How can AI help onboard new hires?

An AI assistant connected to approved skills and a cited knowledge base lets a new hire ask how the company does something and get the current method with its sources. They can run the same skills veterans use, while a reviewer checks their first outputs, so learning happens on real work.

## Sources

- [Gallup, This Fixable Problem Costs U.S. Businesses $1 Trillion, 2019](https://www.gallup.com/workplace/247391/fixable-problem-costs-businesses-trillion.aspx)
- [Panopto and YouGov, Workplace Knowledge and Productivity Report (vendor survey), 2018](https://www.prnewswire.com/news-releases/inefficient-knowledge-sharing-costs-large-businesses-47-million-per-year-300681971.html)
- [MIT NANDA, The GenAI Divide (via Virtualization Review), 2025](https://virtualizationreview.com/articles/2025/08/19/mit-report-finds-most-ai-business-investments-fail-reveals-genai-divide.aspx)