Every organization has an official truth. Every organization also has an actual truth. They are rarely the same thing.
The official truth lives in the ERP, the CRM, the master data, in Confluence, in SharePoint. It is what leadership points to when someone asks: where does that live? It is the place that looks good in audits, that draws cleanly in diagrams, that fits into onboarding decks.
The actual truth lives in heads. The buyer knows which farm reliably delivers in calendar week 27 and which doesn’t; the system says all twelve suppliers are “active”. The planner knows the three workarounds that have kept the officially forbidden schema running for three years. The range buyer knows which farms got dropped just before listing — and why nobody puts them forward again. The route drivers know which box subscriber will stop over the summer, weeks before the email arrives; the cancellation report only catches up when it does.
This is treated as a defect. Mature organizations, runs the unspoken assumption, replace these person-shaped knowledge stores with systems. Single source of truth. Knowledge management. Process documentation. The sentence is always the same: this cannot depend on one person.
But the replacement always fails in the same way. The system captures the explicit decisions. It does not capture the heuristics that produce those decisions. When the person leaves, the heuristics leave with them. The system stays behind — fully documented, and suddenly functionally unreliable.
The structural finding: an organization is a knowledge graph in which most of the edges live inside people, not inside tables. The ERP is a snapshot. The wiki is a snapshot. The person is the running process.
That changes the question. Not: how do we replace these people with systems? But: have we identified who they are, and have we built the structures around them so that being load-bearing is allowed to be the job?
A sharper version of this is taking shape around AI. Retrieval-augmented systems, AI knowledge bases, Copilots in every tab are the latest attempt to consolidate the single source of truth. They inherit the same gap. A RAG architecture answers what has been written down. The unwritten heuristics that give an answer its context still live in the person. The AI tool becomes more articulate; the organizational knowledge problem stays exactly where it was.
The practical question is uncomfortably direct: who in your organization is load-bearing without showing up in any org chart? If the answer is nobody, you don’t have a documented system. You have an undocumented system that hasn’t dissolved yet.