Before you start: what "from scratch" means
A knowledge management system is not a folder tree. It is the up-to-date answer to five questions: what do we know how to do, who knows it, at what level, what would we lose if a person left, and what are we doing about it. To answer them, each step feeds the next: without an org chart, a person has nowhere to belong; without people, mastery has nobody to measure; without knowledge, the criticality grid has nothing to score. Skipping a step saves no time: it comes back, more expensive, two screens later.
Second rule: start small. Importing three thousand people on day one produces a vast, useless map where every mastery level is unknown and every criticality meaningless. Pick one process that already worries you, the five or six people who keep it running, and go all the way with them. Your first hour describes that short path.
In KnowledgeCapital the trial offers two spaces: demonstration data, to see the product filled in, or a blank system — your empty space with your administrator account — to build your own. This guide follows the second case. The Administration › System status screen is your compass: it lists what is configured and what is missing, with a green or orange dot for each item.
Step 1 — The org chart, the backbone
Everything attaches to a structure: people, processes, objectives, accounts. So the org chart comes first. The root already exists: rename it after your organisation rather than recreating it. Then review the levels supplied — from "Parent organisation" down to "Team" — and switch off the ones you do not use: three or four are often enough. Create each structure directly under its parent, with the name people use out loud, then place it on the map.
Step 2 — People, the carriers of knowledge
Create the job positions first (five to fifteen to begin with), then the person records, one by one or from a file. Each person belongs to one structure only, the finest one that truly describes them. Date of birth is optional, but it is what makes retirement forecasting possible. Import happens in two stages: the screen first analyses the file line by line, then nothing is written until you click to apply the import. The number-one rejection reason is an unknown structure — hence step 1. Do not open sign-in accounts yet; step 7 explains why.
Step 3 — Objectives, what gives knowledge its weight
Declare the organisation's strategic objectives. They feed one specific criticality criterion, strategic fit: knowledge that serves a declared objective weighs more than knowledge on its way out.
Step 4 — The catalogue, a shared vocabulary
The knowledge catalogue is the organisation's dictionary. Create five to twelve skill families that mean something to the teams ("Maintenance", "Processes", "Quality", "Safety"), then finer skills, then pieces of knowledge. The golden rule: name the knowledge, not the document — "Adjust the flame on oven 3", not "Oven manual". Two sentences of description say where the knowledge starts and ends. The right grain: something one person can learn from another in a few weeks, and that can be tested. Each piece of knowledge gets a permanent code that never changes, even if it is renamed. An existing job and skills catalogue can be imported from an Excel template; each suggested link is validated line by line.
Step 5 — The criticality grid, from worry to number
Each piece of knowledge is scored on six criteria. On the vulnerability side: availability (how many people truly hold it — computed, never typed in), transferability and confidentiality. On the relevance side: usage, strategic fit and historical importance. Each criterion is ranked 1 to 3, rank 3 always being the most critical, and carries an adjustable weight; if the team cannot agree on weights, a pairwise comparison computes them. The method is published. Start with the most sensitive knowledge: an unscored criterion counts as an average value, because the system prefers an honest average to a false zero.
Step 6 — Processes and tasks, the operational skeleton
A process is named with a verb and its object ("Prepare shipments") and attached to the department that actually runs it. Split it into weighted tasks adding up to 100%, then assign each task to the people who hold it today: a task with nobody on it enters no calculation. Finally, state what each task requires: a skill, a required level from 1 to 5, and whether it is essential. Crossing tasks, assignments and mastery is what reveals fragility: a task held by a single person whose knowledge is in the red zone.
Step 7 — Who knows what: the holders
A new space has no mastery data, and without it nothing can be computed. Two moves fill the system in one go. The first: the skills matrix you probably already keep in Excel — people in rows, knowledge in columns, one level per cell, in ILUO format, stars or figures. Download the template, which carries your real knowledge items as columns, fill it in and upload it: the screen shows cell by cell what it would do, and writes nothing until you apply the matrix. It creates neither people nor knowledge, never lowers a level and never overwrites a level measured by a test.
The second: the "Who do you call?" questionnaire, generated for each person at first sign-in from the knowledge attached to their tasks. That is why the order is strict: assign tasks before opening accounts. A person with no task lands on an empty questionnaire.
Step 8 — Validation committees
A validated sheet carries authority: people trust it without re-checking. That authority comes from a committee, not from a lone click. One general committee is enough to begin with; choose an odd number of members, three or five, since decisions are taken by majority. Members are user accounts: open the ones you need. A committee with no members silently blocks everything waiting for it — the system status screen flags it.
Step 9 — Assessment campaigns
Up to now, mastery levels are declared. Campaigns measure them. First build a question bank in the question workshop: by hand (a question, two to six options, a difficulty), or generated by the local AI from your validated sheets — on a new space, prefer writing by hand. Each question starts as a draft and quizzes nobody until it is validated. Then launch a targeted campaign: the system identifies the holders of orange- or red-zone knowledge, and you tick whom to invite, the deadline, the location and the number of questions. The result sets the stars — 90% correct or more earns 5 — and the measured level replaces the declared one; criticality is recomputed at once. The figure that makes the campaign worthwhile: the gap between declared and measured, skill by skill.
Step 10 — Accounts, then the invitation
Set roles and rights, then open accounts in waves. An LDAP or Active Directory directory can be connected so nobody needs a second password; otherwise accounts are local and a password change is required at first sign-in. You do not invite people into a house with no walls: that is why this step comes last.
Connecting what already exists
As soon as the catalogue exists, your document systems can be read, read-only, to link each document to the knowledge it describes: SharePoint, your document management system, Notion, your intranet or Confluence. It is actually a good order: connect, read the titles, name the knowledge, then recompute the links.
Running it day to day: twenty minutes
- The dashboard: an at-a-glance health check of your structures. You do not act there; you spot where to look.
- The knowledge map: red knowledge held by one person is the permanent priority. For each, decide an action — pairing, documentation, a succession plan — assign it and date it.
- Freshness of the base: a red-zone sheet is reviewed every 180 days, an orange one every 365.
- Committee backlogs: chase them, never validate in their place.
- Campaigns: one per skill family per semester is often enough; the declared-versus-measured gap sets the next one.
- Settings: one change at a time, with its reason, and a week to see its effect.
Once a month: the system status screen, account hygiene (leavers deactivated), a backup that has been tested rather than merely present, and the audit log whenever in doubt.
The mistakes that cost the most
- Importing everything at once, before a single process has been assessed.
- Naming documents instead of knowledge.
- Opening accounts before assigning tasks.
- A committee with an even number of members, or none.
- Taking declared mastery for measured mastery.
For the question of the tool itself, see the evaluation grid for choosing a knowledge management system.