80% of a company’s knowledge is not written anywhere. 600,000 retirements per year prevail. AI will not fill this void: it reveals it.
Astrophysicists call this invisible mass “dark matter” which structures the universe without being able to see or measure it directly. Companies have their own: these unwritten procedures, these adjustments transmitted orally, these tricks of the elders, these files named “FINAL_OK_vDef”. This tacit, dispersed, never formalized knowledge that makes things work. Until one day they no longer work.
600,000 departures per year, zero transmission
In France, this day is approaching. More than 600,000 people retire each year. By 2030, there will be more people over 65 than those under 20 for the first time. However, these departures are not simple administrative transitions. They carry decades of know-how that no one has taken the time — or had the idea — to document. According to Nonaka and Takeuchi’s SECI model, up to 80% of organizational knowledge is tacit, never written, never transmitted other than by word of mouth, from gesture to gesture. When the expert leaves, this dark matter disappears with him.
The nuclear industry offers a textbook case. According to the Global Energy Talent Index, one in four employees in the sector is over 55, one in ten over 65. To compensate for the cascade of departures, companies in the sector are now calling back retirees. An example illustrated by the Experconnect firm, which claims to place several thousand each year in the sector. It’s not nostalgia: it’s operational survival. Nobody knows how certain things work anymore.
The cost of this silent hemorrhage far exceeds recruitment costs. Already in 2017, the International Data Corporation estimated that large companies lose more than $31 billion per year due to poor knowledge sharing. But the problem does not only concern departures. It also concerns those who remain. A 2025 Atlassian study reveals that French employees spend on average nearly nine hours per week searching for information to accomplish their tasks — or more than 450 hours per year. Half of them say the only way to get what they need is to ask someone. When that someone is no longer there, the system seizes up.
France is not alone in facing this challenge, but it is lagging worryingly behind. In Denmark, 28% of companies already use artificial intelligence technologies to structure and exploit their internal knowledge; in Belgium, 25%; in Germany, 20%. In France: 10%. This technological reluctance reflects a deeper problem: our companies are slow to consider their intangible capital as a strategic asset to be protected and transmitted.
AI doesn’t fill the gap, it exposes it
Faced with this observation, many are counting on artificial intelligence to fill the void. The illusion is attractive: assistants capable of answering all questions, summarizing all documents, compensating for all departures. But AI does not create knowledge. It reorganizes what already exists. If the procedures are not written, if the adjustments remain in the heads of the elders — the AI can do nothing. Worse: she “hallucinates”, she invents plausible but false answers, with an assurance that can mislead the most informed users.
The paradox is cruel. Companies that deploy AI tools discover, in trying to power them, that they don’t know what they know. The technology supposed to solve the problem reveals its scale. And more than half of organizations have no structured process for capturing or transmitting tacit knowledge. No mapping of critical knowledge. No transmission procedures before departure. No organized collective memory. Dark matter remains in the shadows until it disappears.
The next crisis for French businesses will perhaps not come from an external shock – neither regulatory, nor competitive, nor technological. She will have the ordinary face of an expert who is leaving, of know-how that is evaporating, of a question that no one knows how to answer anymore.
Dark matter, by definition, cannot be seen. But when she disappears, everything falls apart.
