Capturing knowledge on the shop floor: why typing fails and talking works

An experienced operator hears a machine starting to jam before the fault message appears. He turns one setting back by a fraction, waits a few seconds, and the line starts running again. Thirty seconds of work. The next morning it happens again, this time with a colleague who does not know the trick. The knowledge was there, but it was not captured anywhere. Not because nobody wanted to write it down, but because there is a threshold between solving the problem and documenting it that almost nobody crosses on the shop floor.

This is where every attempt at knowledge capture breaks down. The knowledge is not missing. Capturing it is the bottleneck.

Capture fails because of friction, not unwillingness

The usual explanation is that operators do not want to share knowledge. That is rarely true. The real reason is that the threshold for capturing something is higher than the value it creates for the operator at that moment.

Text-based capture asks for three things at once: stop working, sit behind a screen, and turn practical judgement into written language. Each one costs time and effort at a moment when the operator is still in the middle of the work. The knowledge lives in their hands and judgement, not in complete sentences. That is not a motivation problem. It is the nature of the knowledge itself.

What an experienced craftsperson knows largely falls under what Michael Polanyi called tacit knowledge. On the shop floor this is often called tribal knowledge. Polanyi summed it up in the phrase: “we know more than we can tell” (Polanyi, The Tacit Dimension, 1966). It is practical, action-oriented knowledge built by doing something a hundred times, not by reading about it. Ikujiro Nonaka built on this and showed that implicit knowledge is made explicit through telling and showing, far more readily than through forms and text. Research into tacit knowledge elicitation for shop-floor workers, deliberately drawing out knowledge that would otherwise remain in someone’s head, confirms the same pattern: tacit knowledge is hard to put into words or capture in procedures and documents, and is mainly built through experience and practice (Kernan Freire et al., TU Delft, CHI 2023).

The friction is not in the person. It is in the mismatch between how the knowledge exists and how we ask people to write it down.

Why typing, forms, wikis, and video break down on the shop floor

Four methods keep coming back, and all four collide with the same shop-floor reality.

• Typing interrupts the work. Hands are dirty or occupied, and there is no keyboard workstation next to the machine. Typing is also slow and unnatural for someone whose craft is hands-on.
• Forms force a structure nobody has ready in their head. Shop-floor knowledge is contextual: it is a response to circumstances, not a row of fields. Pressing it into predefined boxes loses the exceptions and judgement that make the knowledge valuable.
• Wikis and knowledge bases require maintenance nobody owns. Writing and updating become separate tasks on top of the real work. Without an owner, content becomes outdated. Once content becomes outdated, nobody trusts the system anymore, and usage fades.
• Video feels easy, but is hard to find and hard to update. Nobody watches eight minutes of footage to find one detail, and a recording that is no longer correct cannot be updated as quickly as a sentence.

Why spoken capture during work fits the shop floor

Speaking is the only capture method that does not require an extra action. The operator does what they already do: explain what is happening in words. The difference is that this time the explanation is captured instead of disappearing.

Three things fall into place. It happens in the moment, while the problem is fresh and the context is complete. It does not interrupt the work, because speaking runs alongside the work instead of stopping it. And the operator describes it in their own words, without a form imposing a structure that distorts the knowledge.

There is also a measurable speed difference. Researchers from Stanford University and the University of Washington showed that speech input is about three times faster than typing, with a lower error rate than keyboard input (Ruan et al., 2016). And where typing, according to the largest typing study ever, based on 136 million keystrokes (Dhakal et al., Aalto University and University of Cambridge, CHI 2018), averages around 52 words per minute, speaking is well above 150. The difference between capturing knowledge and not capturing it is often exactly that difference in effort.

This also matches how implicit knowledge becomes explicit in Nonaka’s work: through telling and showing, not through typing. Technology does not need to change the operator. It only needs to fit what the operator already does. That is the first step; after that comes knowledge preservation: keeping captured knowledge accurate and reliable over time.

How to make capture part of the work

The practical translation is simpler than most documentation projects companies have started.

• Capture knowledge in the moment, not at the end of the shift. Knowledge reconstructed an hour later has already lost half its detail.
• Let people capture in spoken language, without mandatory structure. No fields, no required order, no writing. The operator simply explains what they know.
• Let technology extract the structure afterwards. Transcription and processing turn the spoken fragment into a searchable, structured knowledge item. Ordering the knowledge is a task for the system, not for the craftsperson.
• Add a control step before knowledge becomes truth. Someone who understands the work confirms or corrects what was captured. This keeps quality high without raising the threshold for the operator.

This matters beyond preference. The European agency for vocational education Cedefop points out that workforce ageing can lead to the loss of critical organisational knowledge and experience as people leave. Knowledge that only lives in people’s heads leaves with them. The question is not whether that knowledge is valuable, but whether you capture it before it walks out the door.