What do we actually lose when an expert leaves?
The judgement, not the documentation. The files stay. What leaves is the person who knew which file mattered.
Knowledge management research consistently puts the undocumented share of what an organisation knows at the large majority: analysis published by Atlan describes roughly 90% of institutional knowledge as tacit, held in people's heads rather than in any system. Deloitte has framed the retirement wave as a multi-trillion-dollar knowledge exodus, and put the annual cost of institutional knowledge loss to United States companies at $1.3 trillion.
Those numbers are easy to wave at and hard to feel. The version you actually recognise is smaller and more specific. It is the order the steps have to happen in. The parameter that has to converge first. The script that looks deprecated but is not. The reason the obvious approach was tried in 2019 and abandoned. None of that is in the handover document, because none of it felt like knowledge to the person who had it. It felt like common sense.
Why do handover documents never work?
Three reasons, and they compound.
- Experts cannot enumerate what they know. Asked to write down how they do their job, they write the visible procedure and omit the judgement, because the judgement is invisible from the inside.
- The document is written for nobody. It answers questions in the order the author thought of them, not the order the successor hits them, which is why it is opened once and abandoned.
- It is written last. Handover is scheduled into the final fortnight, competing with everything else a departing person has to finish.
The pattern is common enough to be measured. Deloitte's work on the retirement wave reports that the large majority of organisations do not consistently capture knowledge from employees who are about to retire, despite knowing the departure date months or years ahead.
What works better?
Asking, rather than requesting a document.
A structured interview gets at judgement in a way a blank page does not, because a question can be specific in a way an instruction cannot. "Document your process" produces the procedure. "What goes wrong most often for someone new, and how do you spot it early?" produces the thing you actually needed.
These five draw out most of what is missing. Use them with the person leaving, and record the answers.
-
What breaks most often, and how do you notice before anyone else? -
What would a competent newcomer get wrong in their first month? -
Which of these files still matters, and which are dead ends? -
What did we try before that did not work, and why? -
Who do you call when you are stuck, and what for?
Then make the answers askable. The reason handover documents fail is that the successor has a question at 4pm on a Tuesday and no way to ask it. If what you captured can be questioned in plain language, and answers from the actual material with its sources named, it gets used. If it is a folder, it does not.
How does Eternal Gardens handle this?
Two parts, deliberately in this order: sources first, then the interview that draws out what the sources cannot hold.
The person uploads the material that already exists, and a structured interview captures the judgement around it. Both feed a Mentor, a grounded presence that current and incoming members can ask in plain language. It answers from the approved material and names the sources behind each answer, and it declines when it has no support rather than improvising, which matters more here than almost anywhere else. A confident wrong answer about a procedure is worse than no answer.
Around it sit the things that make a handover land rather than sit: handoff guidance for a specific successor, and short quizzes so a new member can check they have actually understood rather than assuming they have.
When should we start?
While the person is still doing the job, not while they are leaving it.
The best material comes out of someone in the middle of the work, because the judgement is live and they are hitting the edge cases in front of you. In the final fortnight you get a summary, delivered by somebody who has already mentally left.
A practical trigger: start when a departure becomes foreseeable rather than scheduled. A retirement two years out, a student entering their final year, a single person who is the only one who can do something. That last case is worth acting on even with no departure in sight, because a single point of failure does not need to resign to become a problem.
How big should the first one be?
Small enough that you can judge it. One person, one body of work, or one team.
Organisation-wide knowledge programmes fail in a predictable way: by the time there is anything to evaluate, the budget is spent and nobody can tell whether it worked. One well-chosen expert produces something a successor either uses or does not, within weeks, and that answer is worth more than a strategy.
Pick the person whose departure would hurt most, and start there.
Common questions
Is this just a wiki with a chat interface?
The difference is where the content comes from and what happens at the edges. A wiki holds what somebody chose to write down, which is the documented minority of what they know. A structured interview targets the judgement that never gets written, and a grounded Mentor declines when it has no support rather than filling gaps, so a successor can tell the difference between an answer and a guess.
Will experts cooperate, or will they feel replaced?
In our experience it depends almost entirely on framing and on who owns the result. People who are asked to be interviewed about their judgement generally engage; people who are told to document themselves before being made redundant do not. It also helps that they review what was captured before anyone else can use it, so nothing is published in their name that they have not seen.
What about confidential or regulated material?
The approved source set is chosen deliberately, so material that should not be in it is not uploaded in the first place. Access is scoped to the workspace rather than public, and every answer names the sources it used, which is what makes an audit possible. If you have a specific regulatory constraint, raise it at scoping rather than after.
How long does a capture take?
For one person or one defined project, the working part is weeks rather than months, and it is limited mainly by how much of the expert's time you can get. Preparing existing sources can start before they are available. The published Research Knowledge Capsule engagement covers one group or defined project and includes 90 days of support afterwards, because the first real questions arrive once the successor starts.
What does it cost?
The Research Knowledge Capsule is published from CAD $7,500 for one group or defined project, covering source preparation, structured knowledge capture, the Mentor itself, a knowledge map, handoff deliverables, evaluation, and 90 days of support. Larger programmes are scoped from there. Teams who want to try the workflow themselves first can do that on the platform.