Organisations rarely lose the file
When people talk about knowledge leaving a business, they often describe it as though the information itself has disappeared. Usually it has not. The documents remain. The strategy is on the server, the board papers are archived, the policy still exists and somebody can probably find the presentation if they search long enough.
What walks out of the building is the explanation.
The person who remembers why the board rejected one option and chose another. The executive who knows why an apparently strange process exists. The company secretary who remembers that the same issue has already been debated three times. The person who knows that a paragraph buried in a paper from four years ago was actually the beginning of an important change in direction.
That kind of knowledge is difficult to store because it rarely lives neatly in one place. It exists across formal documents, decisions, conversations and the accumulated understanding of people who have been around long enough to connect them.
AI creates an interesting possibility here, because for the first time the archive itself can begin to help make those connections.
From archive to memory
A conventional board portal is very good at telling you where the March 2024 papers are stored. An AI working across the same information could potentially tell you that the issue you are looking at first appeared in March 2024, returned in September, changed materially in February the following year and was eventually resolved after management altered one of the assumptions underneath it.
That is a meaningful change.
The value is not that the AI has created new knowledge. It has made existing knowledge navigable.
This is particularly powerful for boards because corporate decision-making is cumulative. Today's strategy is built partly on yesterday's decisions. Risk appetite develops through experience. Exceptions become precedent. Commitments are made and occasionally forgotten. Directors rotate, executives leave and new people inherit choices whose origins may no longer be obvious.
The AICD's broader guidance on board information reflects this basic reality. Directors need timely access to information that allows them to understand performance, risk and strategic direction, and good governance depends partly on the quality and continuity of that information. AI does not change the need. It changes what access can look like. (aicd.com.au)
The archive no longer has to sit there waiting for someone to know which document to open.
The problem with remembering everything
There is an obvious catch. Organisations do not contain one clean, agreed version of everything they have ever thought. They contain layers.
The board-approved strategy sits beside the draft that came before it. The current risk appetite statement sits beside older versions. A formal minute may coexist with someone's meeting notes. A proposal that was rejected may look almost identical to the one eventually approved. Legal advice might sit near general background material while carrying completely different access requirements.
A human who has worked inside the organisation for years often understands those differences without consciously thinking about them. They know which document matters, which one is obsolete and which one was simply somebody's working view at a moment in time.
An AI will not necessarily know that unless the organisation gives it a way to know.
That is why throwing every document into one enormous knowledge repository does not automatically create institutional memory. In some ways it creates the opposite: perfect access to every version of what the organisation has ever believed without a reliable way of knowing which version it still believes.
The useful architecture has to preserve more than content. It has to preserve authority, context and provenance. An answer about the organisation's risk appetite means very little if the AI cannot distinguish the current board-approved position from a discussion paper written three years earlier. An answer about strategy is less valuable if it quietly combines material from documents that were never intended to be read together.
This is where the librarian analogy becomes useful.
The librarian does not write the books
A good librarian does not simply know where all the books are. They understand the collection. They know which edition is current, how material relates, what belongs together and where to look when the obvious source is not enough.
That is a much more interesting role for organisational AI than simply generating more material.
Instead of being the machine that writes the next board paper, AI can become the layer that helps the organisation understand the papers it already has. It can help a new director trace the history of a decision rather than starting with whatever happens to be in the latest pack. It can help management find the origin of an assumption before repeating it as fact. It can surface a commitment that disappeared between meetings or show when the explanation for a strategy has quietly changed.
But the librarian still needs a library with some order.
Someone inside the organisation has to care about which information carries authority, which material has been superseded and how access should work. That is not a technology problem that disappears once a sufficiently clever model arrives. It is a knowledge-management problem that becomes more important precisely because the model is capable of searching everything so effectively.
The AICD's broader work on AI governance makes a similar point at a higher level. AI governance is not simply about choosing a safe tool. It depends on organisational structures, quality information, human accountability and clarity about how AI-supported conclusions can be understood and challenged. (aicd.com.au)
The asset was never just the document
Companies have talked for years about intellectual property walking out of the building when experienced people leave. Often what they really mean is not formal intellectual property at all. They mean memory, context and accumulated judgement.
The interesting opportunity with AI is not to capture every conversation in the hope that nothing can ever be forgotten. That would create as many problems as it solves. The opportunity is to get much better at preserving the connections between the things the organisation has already decided are worth keeping.
That has implications well beyond the boardroom. Every organisation has people who act as unofficial maps of the place, the ones who know why things are the way they are and where the useful history is buried. AI cannot replace that accumulated understanding simply by ingesting documents, but it can help make more of it accessible and durable.
Much of the current conversation about AI still centres on production: write this, summarise that, make this faster. Those gains are real, but they may not be the most valuable part.
The deeper opportunity may be giving the organisation a better memory.
Not an AI that makes the decisions, and not an AI that remembers everything indiscriminately, but one that knows where the thinking lives and can help people find their way back to it.
That is less like an assistant.
It is much closer to a librarian.
