A master craftsman who had been with the company for twenty years was retiring. The company knew this would have an impact, so they started preparing half a year in advance: asking him to write SOPs, record operations on video, and assigning two younger employees to learn from him.
Six months later, he left. The SOPs were on the file server, the videos were on the cloud drive, and the two younger employees had indeed learned the operational steps. However, over the next three months, the time taken for abnormal situation handling significantly increased, and there were several instances of inaccurate judgments leading to scrap.
Everyone would say, 'The experience wasn't passed on effectively.' But the more precise question is: Which part of experience is inherently uncapturable by SOPs and videos?
Experience Actually Has Three Components
If we break down 'the master craftsman's experience,' it comprises at least three distinct types of elements, and their recordability varies significantly.
1. Operational Steps
'First, close this valve, then open that one when the pressure drops to a certain level.' This layer is clear and repeatable; SOPs and videos can capture it quite completely. Most succession plans focus on this layer, which is why people often feel 'everything that needed to be done has been done.'
2. Judgment Criteria
'This sound is different from usual,' 'The slope of this temperature increase is incorrect.' This layer is usually not written down, because the master craftsman himself may not be able to clearly describe it—he knows something is wrong, but cannot articulate which combination of signals it is based on.
Videos can capture what he did, but not why he did it. SOPs can state 'If temperature exceeds X, shut down the machine,' but they cannot state 'Under these seasonal conditions, with this batch of material, and this machine's status, attention should be paid even before X is reached.'
3. Outcomes After Action
This layer is lost most thoroughly. The master craftsman made a judgment and took action; what was the subsequent outcome? In most cases, this correlation only exists in his own memory. He thus accumulated knowledge of 'which judgments were later proven correct,' and this is precisely the source of his judgment capability.
When he left, the judgment criteria and outcome feedback disappeared with him. What remained were the steps—the layer that is easiest to relearn and has the lowest value.
Why Traditional Methods Fall Short
The three common approaches each have their limitations:
- Writing SOPs: Only parts that can be clearly described can be recorded. Judgment criteria are often lost at the time of writing.
- Recording Videos: Records actions, but not the reasoning behind them. Furthermore, videos are difficult to retrieve—six months later, no one can find 'that similar situation from last time' in which video or at which minute mark.
- Apprenticeship: This is the only way to transmit judgment criteria, but it requires time, and the scope of transmission is limited to situations that happened to occur during that period. Situations not encountered are not learned by the apprentice.
The common limitation of all three is: They are all offline and one-time. Experience is moved and stored somewhere, but it is not continuously accumulated, nor is it linked to subsequent actual outcomes.
Another Approach
If the problem lies in the 'judgment criteria' and 'outcome feedback' not being retained, then what needs to be addressed is not 'how to extract what's in the master craftsman's brain,' but rather to record judgments as they happen, and link them to subsequent outcomes.
Specifically, a single on-site action involves at least these elements:
- What was observed at the time (which values, which phenomena)
- Under what circumstances (which batch of material, which shift, what was the equipment status)
- What judgment was made, which options were considered
- What action was taken, who made the decision
- What was the subsequent outcome
If these five items are recorded when the event occurs, then three years later, another person encountering a similar situation will be able to look up not just 'what to do,' but also 'why that judgment was made last time, and whether it was correct.'
We call this type of traceable, challengeable, and reconstructible record decision memory. The difference between it and a document library is: a document library stores conclusions, while decision memory stores the process and consequences of forming those conclusions.
Where the Difficulty Lies
To be honest: The cost of this approach is not in technology, but in whether the on-site personnel are willing to record.
If recording requires opening an additional system and filling five fields, it will not be implemented. The prerequisite for feasibility is that most of these records can be automatically composed from existing signals—equipment values, event times, and work order statuses are already in the system; people only need to supplement the 'judgment' part.
Furthermore, it must be made clear: This will not make experience transfer automatic. It ensures that judgment criteria no longer reside solely in one person's mind, but to form organizational capability, people still need to read, replicate, and revise practices. What the system can do is ensure the data remains and is traceable, not that someone will necessarily look at it.
The Position of X·Neurons
Connecting on-site events, judgments, actions, and outcomes into traceable records is the core problem that the X·Neurons product direction aims to address. Currently, all capabilities are in a candidate state, and what we describe is the expected behavior during co-validation, not released features.
Enterprise Intelligence is currently our working theory, not an established science. Its boundaries are described on our methodology page, including which claims are currently only hypotheses and lack sufficient evidence.
One Thing You Can Do First
Without needing to implement any system, you can first assess the scale of the problem: Identify three important on-site actions from the past year, and try to answer—
- What was the judgment based on at the time?
- Was that judgment later proven correct?
- Are these two answers currently available in records, or can they only be obtained by asking someone?
If two out of three times you can only ask a person, and that person will retire in a few years, then you are not facing a document management problem.