Tacit Knowledge: What AI Cannot Learn
Tacit knowledge is the experiential know-how employees carry in their heads but have never written down. It determines how well an organisation actually works. And it is exactly the knowledge no AI can reach.
Every organisation that introduces AI runs into the same paradox sooner or later. The models are capable, the integration is live, the licences are paid for. And still the system fails to deliver useful answers to the questions that actually matter day to day.
The reason is rarely the technology. The reason is that an AI can only work with what sits in documents, tickets, wikis and databases. The knowledge that keeps a company running is usually not in any of them.
What is tacit knowledge?
Tacit knowledge is knowledge a person has acquired through experience and can only put into words with difficulty. The term goes back to the chemist and philosopher Michael Polanyi, who coined it in his 1966 book “The Tacit Dimension”.
We know more than we can tell.
Michael Polanyi, The Tacit Dimension (1966)Polanyi’s example is riding a bicycle. Everyone who can do it can do it reliably. But hardly anyone can explain how exactly they keep their balance. The knowledge sits in the body, not in the head, and it cannot be transferred by reading.
At work, tacit knowledge looks less spectacular but is just as hard to grasp. It is the judgement of which client gets nervous at which phrasing. The feeling for when a machine is about to fail even though no warning light is on. The knowledge that on project X you do not use the official approval chain, you simply call the colleague on the third floor.
Explicit and tacit knowledge compared
Explicit knowledge
Documented and transferable
- Process descriptions and manuals
- Product data and specifications
- Contracts, policies, checklists
- Training material and wikis
- Can be copied, stored and indexed
Tacit knowledge
Experience-based and tied to individuals
- Judgement calls and gut feeling
- Knowing who to ask and when
- Unwritten rules and shortcuts
- Relationships with clients and colleagues
- Grows through experience, not through reading
In knowledge management research, tacit knowledge has been considered the far larger share for decades. Depending on the study and the definition, figures between 42 and well over 90 percent of all organisational knowledge are quoted. The spread mainly shows one thing: the share cannot be measured precisely, because you cannot count what is written down nowhere.
Why AI fails at exactly this point
An AI works with what it can read. Inside a company that means SharePoint, Confluence, ticketing systems, email archives, CRM records. All sources that hold explicit knowledge. The technology behind it, usually a combination of vector search and a language model, is very good at searching, summarising and repackaging that material.
What it cannot do is fill a gap that was never closed in the source material. If nowhere it is documented why a particular supplier has not been approached since 2019, that information is not in the index either. The AI will politely reply that it cannot find anything on the subject. Or, worse, it will invent a plausible explanation.
In 2025, MIT analysed 300 public AI deployments for its report “The GenAI Divide” and surveyed executives as well as employees. The result: 95 percent of the pilots had no measurable effect on business figures. The authors name as the main cause not model quality but a learning gap between tool and organisation. Systems that do not know a company’s context and do not learn from it stay stuck at pilot stage.
For HR and knowledge management this leads to an uncomfortable conclusion. The more a company bets on AI, the more visible it becomes how little of its actual knowledge was ever written down. The technology exposes a gap nobody noticed before, because in daily work it was bridged by people.
The double time pressure
That gap would be a manageable problem if companies had unlimited time to close it. They do not. Two developments are running at the same time and reinforcing each other.
Requirements are shifting
For its Future of Jobs Report, the World Economic Forum regularly surveys more than 1,000 employers worldwide. For the period from 2025 to 2030, these employers expect 39 percent of today’s core skills to change or become outdated. The figure has dropped slightly compared with 2023, when it stood at 44 percent. It remains high nonetheless: four out of ten skills your organisation runs on today will look different in 2030.
Closing that gap through hiring alone will not work. The labour market simply does not supply that number of specialists.
The experience carriers are leaving
According to calculations by the German Federal Statistical Office based on the 2024 microcensus, around 13.4 million people in the German workforce will pass the retirement age of 67 by 2039. That is close to a third of everyone currently available to the labour market. The German Economic Institute expects a shortfall of roughly 4.3 million workers by 2036, because far fewer young people are entering than older ones are leaving.
Each of these people takes tacit knowledge with them that has grown over decades. A survey by Atradius among more than 330 companies shows that while the majority does take concrete measures, one in five has no systematic strategy for securing that knowledge.
The arithmetic is uncomfortably simple. Over the next few years you will need new skills while losing the people who carry the most experiential knowledge. The fastest reskilling programme available is already on your payroll: colleagues who can do exactly what others need right now.
Why documentation drives usually fail
The obvious reaction is a large documentation project. Everyone writes down what they know, the wiki gets filled, the AI finally has something to work with. In practice this rarely goes well, for three reasons.
- You cannot write down what you do not know you know. Tacit knowledge is unconscious by definition. Experienced employees treat their own judgement as self-evident and never think of documenting it.
- The context gets lost. Even when someone writes down a rule, the why is missing. “Always follow up with client Y by phone” is an empty instruction without the story behind it, and it will be ignored at the first objection.
- Documentation ages faster than it is created. With 39 percent of skills shifting by 2030, a large share of what you laboriously write down today will be obsolete in five years.
That does not mean documentation is pointless. Processes, specifications and policies belong in writing. But documentation is the wrong tool for the part of knowledge this article is about.
What actually works: socialisation
Knowledge management research has had a model for this problem since the 1990s. In their SECI model, Ikujiro Nonaka and Hirotaka Takeuchi describe four ways in which knowledge emerges and moves within organisations. The first is called socialisation and means the direct transfer from tacit to tacit knowledge.
Socialisation does not work through documents. It works through shared experience, observation, imitation and above all conversation. The master craftsman and the apprentice. The senior developer and the junior in pair programming. Two colleagues from different departments who happen to talk and discover that one has spent months solving a problem the other solved long ago.
The decisive point: tacit knowledge travels from person to person. It does not need a storage location, it needs an encounter. And those encounters have become rarer in recent years, because hybrid work removed the chance conversations in the hallway and the canteen without replacing them.
How that effect plays out in practice and what helps against it is covered in the article on networking in hybrid teams.
How companies secure tacit knowledge in practice
If knowledge travels through encounters, the task is not to document more but to enable more encounters. And not at random, but deliberately between the people for whom the exchange pays off.
Identify knowledge carriers
Who is retiring in the next three to five years? Which roles are staffed by a single person? Where does a critical process hang on one individual? This mapping is the precondition for everything else.
Form tandems
Bring experienced employees together with younger ones on purpose, across departmental boundaries. Not as formal mentoring with target agreements, but as a regular conversation without a fixed agenda item.
Let knowledge flow both ways
Reverse mentoring turns the direction around: younger employees bring digital skills and fresh perspectives to experienced colleagues. Both sides give, both sides receive. That increases the willingness to engage.
Structure peer learning
Learning groups, peer coaching and collegial case consultation turn individual knowledge into shared knowledge. The side effect: whoever has to explain something understands it better afterwards.
Keep the exchange running
A one-off workshop does not create a knowledge network. Exchange only becomes effective once it happens regularly and turns into part of everyday work rather than a special event.
The individual building blocks are described in more detail in the articles on securing knowledge transfer, reverse mentoring and peer learning.
The point where it breaks organisationally
In a team of 20 this organises itself. With 500 or 5,000 employees it does not. Then questions come up that nobody can answer on the side: who has already talked to whom? Which areas are well connected and which are blind spots? Where does expertise sit that others know nothing about? Anyone without those answers keeps working against silo thinking without seeing where the silos actually run.
This is exactly where a networking platform comes in. Workdate brings employees together systematically, across departments, locations, hierarchies and generations. The matching takes skills, interests, location and seniority into account, so conversations are not random but happen where knowledge can actually flow.
- Cross-generational matching for the transfer of experiential knowledge
- Topic- and skill-based pairing for peer learning
- Continuous meetings instead of rigid rounds, integrated into everyday work
- Analysis of networking levels across the organisation, including blind spots
- Voluntary, GDPR-compliant, no performance monitoring
More on the Generation Coffee use case →
More on the Peer Learning use case →
Frequently asked questions
What is the difference between tacit and explicit knowledge?
Explicit knowledge is documented and can be copied, stored and passed on: manuals, process descriptions, specifications. Tacit knowledge sits in experience, judgement and relationships. It is hard to put into words and travels mainly from person to person through direct collaboration and conversation.
Can AI capture tacit knowledge at all?
Only indirectly, and only once it has been externalised. An AI can analyse meeting notes, recorded rationales behind decisions or captured explanations. As long as that material does not exist, the knowledge remains invisible to the system. AI does not replace knowledge transfer between people; it can only make it usable after the fact.
What percentage of corporate knowledge is tacit?
There is no reliable single figure. Knowledge management literature quotes values between 42 and well over 90 percent, depending on definition and method. The reason for that range lies in the nature of the subject: what is documented nowhere cannot be measured precisely. What matters is less the exact number than the insight that it is the larger share.
Is a documentation project still worthwhile?
For explicit knowledge, yes. Processes, specifications and policies belong in clean documentation, and an AI can do a lot with them. For experiential knowledge, documentation is the wrong tool. The sensible approach is a combination: document what can be documented, and organise direct exchange between people for the rest.
How do I start when experienced employees are about to retire?
With a mapping exercise: who is leaving when, and which knowledge depends on that person? Then form targeted tandems, ideally 12 to 24 months before departure. Starting early is essential. Knowledge transfer that only begins with the resignation letter comes too late, because tacit knowledge needs time and repeated conversations.
Related topics
Knowledge travels between people
See how Workdate connects your employees on purpose, so experiential knowledge is passed on before it leaves the company.