A proposal remains untouched because only one person knows the special commercial terms. A complaint is passed from colleague to colleague until the experienced employee returns from leave. A new site manager can access every folder yet still cannot judge important exceptions. These are illustrative situations, not Fassbender client cases.
In such cases, the business does not necessarily lack information. It has an operational dependency. Certain work functions only when a particular person is available. As the business grows across customers, locations or headcount, so can the queue of questions for the same small group of people.
I would not start by procuring a knowledge portal. First, establish which work stalls without those people and what a properly prepared deputy may actually do and decide. Knowledge has only been transferred effectively when someone else can act on it reliably within an agreed remit.
The risk starts before someone resigns
Knowledge loss is often treated as a departure issue. The response is then a handover checklist, interviews and one final folder. That may help create an orderly departure. It comes too late for operational scaling.
The dependency is already visible when a specialist reviews every exception, is repeatedly interrupted by new employees or regularly has to set aside their own work. Naming a deputy does not resolve this if that person has neither the context nor the authority. Two people can interpret the same unclear rule differently.
NASA’s practical guidance on knowledge continuity starts earlier. It advises leaders to identify critical knowledge and sole knowledge holders, make time for transfer as part of the work and pass knowledge on continuously. Mentoring, joint casework and role descriptions are part of that approach. This is practical guidance from a particular organisational setting, not evidence of a specific saving for small and medium-sized businesses. The management question is nevertheless immediately useful: what knowledge does our business need before someone is unavailable? 1
What is not in the folders
A work instruction will often describe the standard case. The experienced employee also knows when the case is not standard. They recognise an unusual combination of circumstances, understand the background to an agreement or notice that an apparently complete request omits something material.
Some of this can be explained and documented. Another part develops through practice, feedback and working with others. Not every experience becomes a reliable decision rule through a recorded interview.
A 2024 systematic review by Igoa-Iraola and Díez analysed 28 research articles on knowledge transfer during generational change and in knowledge management. It distinguishes between knowledge that can be documented and more difficult-to-articulate experiential knowledge, while also emphasising the organisational and social conditions needed for transfer. The literature reviewed concerns predominantly larger businesses, particularly in industrial sectors. It supports neither a general rate of knowledge loss nor a guaranteed method for every small and medium-sized business. 2
In practice, a document can explain a boundary. Whether a colleague recognises that boundary in a complex case is something they need to apply and discuss. That requires time with experienced people. That time cannot be left over only once every other task has been completed.
Do not secure all knowledge. Secure the critical work.
An attempt to capture everything quickly becomes an endless archive project. I would begin with one business-critical process, such as handling demanding service cases. Not with the entire knowledge landscape.
The functional lead should record which cases currently wait for a particular person. The consequences matter: does a customer go without an answer? Is a service promised incorrectly? Is a machine at a standstill? Or is this simply a helpful shortcut whose loss would be inconvenient but manageable?
Priority should go to knowledge whose absence could cause material harm, which would be difficult to rebuild and for which no reliable cover exists. At first, this assessment is a management judgement. It should be tested against real working cases, not replaced by a score that only appears precise.
Each prioritised case needs an accountable role, a deputy, authorised decision boundaries and a route for unresolved questions. A list of expert names is not enough. It merely shows whom people would call again.
A case must be handed over with its reasoning
Take a service goodwill decision as an explicitly illustrative example. The system shows that a replacement service was offered to a customer. Without the reasoning, the deputy might regard this as general practice. In fact, it may have been a one-off exception with explicit approval.
The transfer must therefore preserve more than the outcome: what facts applied? What rule was in force at the time? Why was an exception permissible? Who approved it and where does comparability with the next case end?
The specialist first works through an appropriate case jointly with the deputy and explains their approach. In the next case, the deputy takes the lead; the specialist reviews both the reasoning and the outcome. What is repeatedly missing is added to the working materials. Where a rule is unclear, the functionally accountable role decides rather than allowing an interview recording to become policy by default.
This does not require a new system. The existing working environment will often suffice initially, provided that the source, scope of applicability and accountable maintenance are clear. The key question is whether the material can genuinely be used while handling the work.
People must not feel that they are training themselves out of a role
Someone who has resolved difficult cases for years is now expected to explain their own knowledge alongside everything else. All previous expectations remain in place. In those conditions, reluctance is not surprising.
Leadership needs to state the purpose openly: the organisation should be able to operate reliably and specialists should not have to resolve every repeated question themselves. This must not become an implicit promise that their expertise will be dispensable after an interview. Sound judgement remains necessary, particularly for new and consequential cases.
I would reserve time for joint casework as a firm commitment and reduce other duties accordingly. Passing on knowledge is part of the role’s contribution and should be recognised as such. Employees must also know what records are being created, who will see them and how corrections can be made.
There is a real tension here. Wider access can make collaboration easier, but it can also breach confidentiality. Customer files, personnel information and protected agreements do not belong automatically in a company-wide search space. A person may hold knowledge without the business being entitled to reuse every one of their messages without restriction. Access rights and permitted use must be clarified before any technical connection is made.
Where AI can help and where it merely conceals the dependency
AI can structure approved case notes, flag inconsistencies for review or retrieve relevant documents for a working case. An assistant can also help derive unresolved questions from an expert interview for the next joint casework session. That does not make the underlying specialist content automatically correct or authoritative.
A field study published in the Quarterly Journal of Economics in 2025 examined the staggered, non-randomised introduction of an AI assistant using data from 5,172 support workers at one software company. Access to the assistant increased the number of issues resolved per hour by 15% on average. Less experienced employees benefited more; small declines in quality also appeared among the most experienced staff. The authors find evidence consistent with the diffusion of successful working practices. 3
This is a concrete signal of potential, not evidence that a chatbot can replace the knowledge of any key individual. The study concerns a specific support setting, a historical GPT-3-based application and a rollout that took place primarily in 2020 and 2021. It does not measure a current retrieval-augmented generation solution and it promises neither a 15% saving nor a comparable effect in your business.
An internal knowledge assistant requires reviewed, current sources with a defined scope and access rights. Retrieval-augmented generation, which incorporates retrieved sources into answer generation, can improve the basis for a response. Errors remain possible. NIST AI 600-1, a voluntary US framework, expressly recommends checking sources and citations as well as the provenance of the data used. It is not a certification or legal guarantee. 4
The real test is therefore not: can the assistant produce a persuasive answer? It is: can the deputy handle the working case correctly with the available support and refer an unresolved decision to the accountable function? We address the behaviour required when business rules are missing in greater depth in the executive article on reasoned abstention from answering.