Everybody was right and nothing got fixed

A problem shows up. It’s odd, it’s not urgent yet, and it lands on someone’s desk in one division. They look at it properly, and they conclude; correctly - that it isn’t theirs. The behaviour points somewhere upstream. They write a careful note and pass it on.

The receiving division looks at it properly too. And they conclude, also correctly, that this isn’t theirs either. Under their definition of their scope, it plainly belongs to a third group. They write their own careful note and pass it along.

Five weeks later the problem is back where it started, having acquired a small archive of correspondence and no answer. Nobody obstructed anything. Every individual judgment in that chain was defensible, and most were right. The organisation still spent five weeks not solving something.

This is the part that took me a while to accept: there is no villain in this story. Every step is locally correct and the whole thing is globally useless.

The org chart is a map of accountability, not of knowledge

Those divisions are connected. There is a line between them. There are shared programmes, escalation paths, a monthly steering meeting, a matrix reporting structure someone spent a quarter designing.

None of that describes who knows what.

An org chart answers who is answerable for this. It is built for accountability, and for accountability it works. The question people actually have is different - who has seen something like this before - and the org chart is silent on it. It cannot help, because knowledge is not distributed the way responsibility is. The two graphs have different shapes and nobody maintains the second one.

So people navigate using the map they have. They route the question by ownership, because ownership is the only thing that’s written down. And ownership routing works fine right up until the problem doesn’t belong to anyone - at which point it circulates indefinitely, because there is no division whose remit is problems that don’t fall inside a division.

Three names for the same thing, at the same time

Here’s what makes it worse, and this is the part I find people underestimate.

Each division describes the problem in its own language.

In a manufacturer, the software group logs it as a start-up sequence fault. The hardware group has it as an intermittent connection issue. Quality has it as a field return with an unconfirmed cause. Three tickets, three systems, three vocabularies, one underlying event - and no one is in a position to notice, because nobody reads all three.

In a hospital it’s a surgical complication, a medication reconciliation issue, and a discharge planning failure. In a bank it’s a fraud pattern, a control gap, and a product defect. Same event every time. Three descriptions that share almost no words.

This is not a communication failure in the soft sense. Everyone communicated. Everyone documented. Every division did exactly what it was supposed to do, in the terminology its own function is required to use. The record is complete and it is fragmented into pieces that cannot be recognised as parts of the same thing.

And the cost of that has no line item anywhere. When a question spends five weeks bouncing, nothing goes wrong. There is no incident, nobody is blamed, and the correspondence trail looks like diligence, because it is diligence. From the inside it is indistinguishable from ordinary competent work. It only becomes visible at the far end, when the thing everybody partially understood reaches a customer or a patient or a regulator, and someone reconstructs the history and finds that three divisions each held a third of the answer for eleven months.

Regulation builds the walls on purpose

Now the uncomfortable part for anyone working in a regulated industry.

The silos are not an accident of culture. They are frequently a requirement.

Segregation of duties. Independent quality assurance. Validation kept structurally separate from development. Distinct audit responsibilities with distinct reporting lines. First line, second line, third line. These exist for good reasons and they are, in many cases, legally mandated. The wall between two functions is a compliance feature.

It is also, at the same time, the reason the two functions cannot see that they’re looking at one problem. The industries with the strongest obligation to know things; pharma, aviation, financial services, medical devices, energy - are the ones whose structure is most deliberately arranged to prevent any single vantage point. You cannot fix that by asking people to collaborate more. The separation is the control, and the control is staying.

Where AI is usually pointed, and where it should be

Almost every AI deployment I have seen inside these organisations is bought by one division, trained on that division’s data, and speaks that division’s vocabulary.

Which means it makes each silo faster at being a silo. Quality gets a better quality assistant. Engineering gets a better engineering assistant. Each of them is now more productive inside its own walls and no closer to the group next door. The walls get taller, and they get taller at the speed of software.

The version worth building goes the other way. It reads across the divisions without asking any of them to change how they work or what they call things, and its whole job is recognising that these three descriptions are one event. It doesn’t need anyone to collaborate more, which is fortunate, because they can’t.

As I wrote last time, the judgment still has to come from people who have spent decades being reliably right. That hasn’t changed and won’t. What can change is whether their judgment stays inside one division.

Next time: the mechanical part: what actually happens when you search a complete, well-kept record and it returns nothing at all.