How Tacit Is Tacit Once the Work Leaves a Digital Trail?

Mike says most of what he knows is in his head. The company believes him. Then someone looks more carefully. Years of customer emails show how he responds when confidence begins to thin. Pricing records show which exceptions he allowed and which ones he refused. Maintenance notes show the sequence of observations before a shutdown. Calendar history shows who enters the room when a certain kind of decision appears.

None of these records is a manual. Together they may show regularities Mike would struggle to describe from memory. The company is no longer limited to asking what he can explain in an interview. It can examine what the work itself has recorded.

That changes the problem. It does not finish it. A trail proves that something happened in the system. It may still be silent about what Mike noticed, what he feared, which option he dismissed, or why the same visible action meant something different in another case.

Nonaka’s 1994 theory describes organizational knowledge creation as a continuing dialogue between tacit and explicit knowledge, with organizations helping articulate and amplify knowledge developed by individuals. It is a theory of knowledge creation, not a promise that everything a person knows can be written down.[A Dynamic Theory of Organizational Knowledge Creation]

Nonaka and von Krogh later addressed controversy around knowledge conversion and described tacit and explicit knowledge as conceptually distinguishable along a continuum. Their review is a useful correction to the idea that knowledge must be either perfectly codified or permanently inaccessible.[Tacit Knowledge and Knowledge Conversion: Controversy and Advancement in Organizational Knowledge Creation Theory]

For succession, the practical question becomes more precise: which parts of the judgment have left observable evidence, which can be reconstructed through a person’s account and surrounding records, which can be practiced by someone else, and which remain uncertain or inseparable from personal experience?

Brynjolfsson, Li, and Raymond studied 5,179 customer-support agents after access to a generative-AI assistant. They found a 14 percent average productivity increase, with a 34 percent increase among novice and lower-skilled workers, and reported suggestive evidence that the tool disseminated practices associated with stronger performers.[Generative AI at Work]

Customer support leaves unusually rich tracks: repeated text conversations, recurring problem classes, knowledge articles, resolution signals, and rapid feedback. A system can observe enough examples to offer useful assistance at the moment of work. In that bounded setting, part of the experience curve became more available to people who had not yet lived through as many cases.

The result should unsettle any easy claim that experience cannot travel through technology. It should not be stretched into a claim that every form of executive or technical judgment has the same data, frequency, feedback, or measurable outcome. The trail differs by work.

A CRM may show that Mike called the customer before approving an exception. It may not show that the customer’s silence, rather than the request itself, prompted the call. An operating system may record a shutdown. It may not contain the sound, smell, or pattern of small readings that changed the engineer’s interpretation. An email archive may preserve the final recommendation while omitting the rejected draft and the private obligation that constrained it.

The data also inherit the boundaries of the system that collected them. If no field records uncertainty, every decision can appear more settled than it was. If success is measured by speed, the quiet avoidance of a future failure may have no label. If only accepted actions remain, the organization loses the alternatives the expert considered and declined.

Digital records are therefore evidence about work, not the work in full. They become more useful when someone can connect the trace to the incident, recover the changing interpretation, and say what the system did not observe.

Start with a decision that mattered and build a timeline from the available record: messages, system events, notes, drafts, meetings, changes in access, and outcomes. Ask the expert to correct it. Where did the record begin too late? Which visible event was irrelevant? What did they know from a relationship or prior case that never entered the system?

Critical Decision Method adds structured questions around cues, expectations, goals, alternatives, and turning points. Cognitive Task Analysis helps identify the mental demands beneath the visible sequence. Naturalistic Decision Making keeps the inquiry anchored in the uncertainty and organizational constraints of the real situation.

Skagway’s approach is a founder-led practitioner synthesis of those traditions, incident reconstruction, documentary evidence, contrasting cases, and successor rehearsal. How We Work explains the lineage and its limits. The method does not certify that a record contains a mind; it creates a more useful and challengeable account of selected judgment.

A model may reproduce the historical action. A successor may repeat the lesson. Neither result shows that the underlying distinction will survive when the case changes. Alter the conditions that should matter and preserve the ones that should not. Introduce a familiar cue in the wrong context. Remove the person or data source that normally resolves uncertainty.

Then observe the response. Does the successor ask for the missing evidence? Does the system escalate? Does either recognize that the historical pattern has crossed its boundary? The test should record support, errors, corrections, and uncertainty rather than produce a theatrical declaration that judgment has been transferred.

Before You Automate Expert Work, Find the Exceptions develops this boundary problem for AI-supported workflows. The same discipline belongs in succession: the recurring case may be the easiest part to preserve and the least revealing test of independent judgment.

A credible record identifies its sources, dates, owners, and blind spots. It distinguishes an observed behavior from the expert’s later explanation, and both from the organization’s interpretation. It gives assumptions a review trigger. It leaves room for the successor to disagree when the terrain changes.

Organizational Memory Is Not a Database explains why storage alone cannot provide retrieval, interpretation, or challenge. The digital trail can deepen the memory, but only if someone knows when to bring it forward and when to question it.

The first question was whether the work left enough evidence to recover. The next is what a different person can do with that evidence when the familiar pattern breaks. The Map begins with the traces. The Passage tests whether someone else can carry the decision.