Go and see

Every organisation produces a representation of itself — reports, dashboards, decks, summaries. That representation is not merely incomplete; it is lossy in a particular direction. It bends toward what the person writing it believes the person reading it wants. Which means a leader working only from the representation is not under-informed. They are confidently misinformed, and the confidence is the dangerous half.

Four traditions I have run into, in four different worlds, have arrived at the same correction. None of them cites the others. That convergence is the reason I trust the idea more than I trust any single version of it.

The factory: gemba

Manufacturing has a word for the place where the work actually happens — gemba, the actual place — and a principle attached to it: genchi genbutsu, go and see for yourself. Taiichi Ohno is said to have drawn a chalk circle on the floor of a Toyota plant and made a manager stand inside it for hours, not to humiliate him, but because it takes that long before you stop seeing what you expected and start seeing what is there.

The practice is unglamorous and specific. You watch the flow. You count the work in progress piling up between two stations. You notice that an operator walks eleven metres for a part that could sit at arm's reach, and that nobody has mentioned it because it has always been that way. None of this arrives in a report, because none of it is anybody's number.

There is a counterfeit version, and it is worse than not going. The executive site visit — the scheduled route, the cleaned line, the prepared deck in the meeting room afterwards — has the form of gemba and none of the content. It is theatre, and its real damage is that it inoculates: the visitor leaves with the sensation of having seen, which is precisely what stops them going back to look properly.

The battlefield: shared consciousness

Some organisations cannot be walked. Stanley McChrystal's account of running special operations in Iraq describes an enemy network that adapted faster than a hierarchy could transmit a decision — the problem was not that anyone was slow, it was that the structure imposed a delay between something being seen and something being done.

The answer had two halves, and they only work together. Shared consciousness: make ground truth ambiently available to everyone at once, so a junior analyst's observation reaches the whole organisation without being staged through three layers of summary. And empowered execution: push the decision to the person who can see the situation, because the value of information decays and the loop has to close where the observation was made.

Where gemba is a leader going to the truth, this is a system bringing the truth to everyone. It is the same principle at a scale where walking the floor is not physically possible.

The routing rule: two wrappers

The third version is about how a message travels. The rule, stated most bluntly by Elon Musk in an internal communications policy, is that information should take the shortest path to the person who can act on it — not up one branch of the org chart, across, and back down. Insisting on chain-of-command communication is, in that formulation, a firing offence rather than good order.

Whatever one makes of the messenger, the mechanism is sound, and it has a much gentler expression closer to home. Nordic organisations, and Swedish ones in particular, are unusually flat by international standards. Anyone speaks to anyone. The chief executive is reachable, is addressed by first name, and is not insulted by being approached directly by someone four levels away. In a more hierarchical organisation the same request travels differently: reaching the person who knows the answer means passing through several who do not, and every one of those is a point at which the question can be reshaped or the answer softened. Hierarchy has its own virtue — it is clear, and clarity is not nothing — but each additional hop is a place where information changes on the way.

Two very different wrappers around one idea: the path a fact takes should be determined by the problem, not by the hierarchy. One version arrives by decree and the other by culture, and the cultural one is more durable, because a rule can be reinstated the moment nobody is watching and a norm cannot.

But flat is not the same as consensus

Here I want to be careful, because Swedish corporate culture is regularly described as one thing when it is really two, and they pull in opposite directions.

Flatness is about access — who may speak to whom, and how quickly a fact can travel. It is an asset for everything above.

Consensus — förankring, the anchoring of a decision with everyone it touches before it is taken — is about agreement. It has real strengths: once a decision is anchored it executes quickly, because nobody is relitigating it in the corridor.

But it carries a specific failure mode, and it is exactly the one this essay is about. When a decision requires broad agreement, raising an inconvenient fact does not merely add information — it threatens the alignment being built. So the fact gets softened. Not suppressed, and rarely dishonestly; simply rounded toward the version that keeps the room together. Conflict avoidance is a filter with better manners than hierarchy, and it is harder to see, because everyone involved is being reasonable and pleasant.

A flat organisation with a consensus reflex can therefore end up with the same distorted picture as a rigid hierarchy, arrived at by a completely different route. Which is why I would not treat Nordic flatness as sufficient on its own. It opens the channel. It does not guarantee that what travels down the channel is undiluted.

The third filter: the room

Hierarchy filters by adding hops. Consensus filters by making an inconvenient fact socially expensive. There is a third, and it keeps working even after the other two have been dealt with: the setting in which the question gets asked.

The person who knows most about a specific thing is very often the most junior person in the room, because they are closest to it. They are also, in a meeting of twelve people spanning three levels of seniority, among the least likely to say it. Not because anyone silences them. Large rooms reward confidence over knowledge, and whoever speaks first sets the frame that everyone afterwards has to argue against. A quiet person with the decisive fact and no appetite for contradicting a senior colleague in public will simply not mention it, and nobody will notice that anything was lost.

Nobody intends this, which is exactly what makes it harder to correct than an intention would be.

It is also the part of gemba I had underrated. Going to where the work happens is not only about seeing the thing; it is about the conversation that becomes possible standing next to it. The person who said nothing in the review will tell you, at their own station and one to one, precisely what is wrong with the process and what they have been working around for two years. The observation and the setting arrive together, and the setting may be doing more of the work.

It is part of what shared consciousness buys, too: a finding that reaches everyone at once does not first have to win a room.

The corollaries are small and unglamorous. Ask the person doing the work, where they do it. If you are the most senior person present, speak last — an early view from you is not a contribution, it is an anchor. And treat the size of the room as something you choose rather than something the calendar decides.

The machine, which cannot walk anywhere

The fourth instance is the one I ran into most recently: the same problem in an unfamiliar costume.

In a system where the work is done by software agents rather than people, there is no floor to visit. What there is instead is a strong temptation to let each agent report what it believes it did — and agents, like people, report optimistically. What worked was to stop trusting those reports: parallel workers re-scan the filesystem for what has actually been built rather than trusting their own record of building it, and no builder's account of its own work is accepted as evidence of anything. Verification is mechanical, and it runs before any opinion is invited.

That is gemba, implemented in code. Go and look at the actual place. Do not take the summary when the thing itself is available.

What it is good for

The practical value of noticing the convergence is that it tells you which tool fits which situation. If the work is physical, walk it, and stay long enough to stop seeing what you expected. If the organisation is too distributed to walk, build the mechanism that makes ground truth ambient rather than requesting it up a chain. If the problem is routing, remove the rule that forces a fact to travel the org chart. And if the workers are machines, make them re-observe reality instead of reporting on it.

It also gives you one useful question. Distance from the truth is not set by the size of a company or by anybody's title — it is set by what the organisation treats as authoritative. So the question worth asking of any business is: when the report and the floor disagree, which one does this company believe?

Everyone has an answer ready. The useful part is the example they reach for next, and whether they have one at all.