What Only Paulo Knows: The Knowledge Nobody Wrote Down

A packaging manufacturer, 150 people. Paulo has been the planning coordinator for eleven years. He knows the client in the south accepts partial delivery if you warn them by Wednesday, that extruder two loses stability once humidity passes a certain point, and that one distributor's order always arrives with the wrong code — and which code is right.

None of that is written down. When Paulo took twenty days off in January, the company had three rejected deliveries, a two-hour line stoppage, and an argument with sales. When he came back, the director asked him to 'document everything you do.' Paulo delivered fourteen pages in three weeks. Those fourteen pages contained none of the three things in the paragraph above — and not for lack of goodwill.

01The idea: some knowledge doesn't fit in text #

Michael Polanyi, a Hungarian chemist and philosopher, wrote the sentence that frames the whole subject in The Tacit Dimension, in 1966: we know more than we can tell. He used face recognition as his example — we pick one person out of thousands without being able to list the criteria we used. Riding a bike is the same. The knowledge exists, it works, and it resists explicit formulation.

Ikujiro Nonaka and Hirotaka Takeuchi took that distinction inside the company in The Knowledge-Creating Company, published in 1995 by Oxford University Press. Their thesis: the competitive advantage of the Japanese firms they studied didn't come from hoarding knowledge, but from converting tacit knowledge into explicit and back again, continuously.

They described that movement in four modes, the SECI cycle: socialization (tacit to tacit, learning together by watching); externalization (tacit to explicit, putting it into words, analogy, diagram); combination (explicit to explicit, assembling and systematizing what's already written); and internalization (explicit to tacit, practicing until it becomes automatic).

What Paulo was asked to do skipped three of the four modes. It asked for pure externalization, alone, in a room, with no counterpart and no concrete case. Nonaka and Takeuchi insist externalization happens in dialogue — someone asks, the other tries to explain, the explanation fails, and it's on the second attempt that the implicit rule shows up.

02Why 'document everything' fails #

Three reasons, all observable in any mid-size company.

The first is Polanyi's: whoever does the work has no conscious access to most of what they do. Ask an experienced person how they decide and you'll hear the official version of the procedure, not the real one. The actual rule only surfaces against a concrete case — 'and what if it comes in like this?'

The second is economic. Documentation written outside the flow of work goes stale the next day and nobody has an incentive to update it. Within six months it's wrong, and a wrong document is worse than no document, because whoever trusts it makes mistakes confidently.

The third is about form. Much of what Paulo knows is conditional: if A and B, then C, except when D. That isn't prose — it's a process rule. Writing as prose what should be a step with a condition is the surest way to lose the information.

03The cost of dependency #

In that 150-person mid-size Brazilian company (figures in Brazilian reais), with the assumptions on the table and openly estimated:

Want to see how this looks inside a real operation? Explore the platform.

The R$ 42 mil estimate adds up three rejected deliveries with return freight, two hours of machine downtime, and the management time spent working around each episode. It's napkin math. Even if it's off by half, it's bigger than the cost of two hours a week of structured conversation with Paulo over one quarter.

04Where this gets stuck in practice #

1. Knowledge lives where the work doesn't. The procedure sits in a file in a shared folder; execution happens in another system. Nobody opens the file in the middle of a task.
2. Exceptions never get recorded. The happy path is documented; what eats the specialist's day are the exceptions, and those live in his inbox.
3. Onboarding is shadowing and nothing else. It works — it's legitimate socialization — but it depends on the veteran having time, and it doesn't survive his departure.

05What data-driven management answers #

Knowledge holds up when it sits in the same place the work happens. A process step with the condition written into it is documentation nobody has to remember to open: it shows up at the right moment, for the person doing the work, and gets updated by whoever uses it.

That's the design behind the documents, processes, and apps modules inside the platform — and it's where AI agents change level, because an agent trained on the operation's real history answers 'how do we do this here' from what the company recorded, not from what someone wrote once. It's also one of the capabilities that define the Collaborative Work Management category.

Two honest caveats. Externalizing takes the specialist's time, and the specialist is exactly the busiest person in the area — if that time isn't protected on the calendar, nothing happens. And over-recording has a cost of its own: recording everything is as useless as recording nothing. Pick the critical processes, the same way you pick which variation deserves investigation.

Paulo isn't a risk. Paulo is an asset the company never bothered to record — and that part is a management decision, not a problem with the person doing the work.