Trenches, Not Silos

There is a phrase that turns up in companies once they reach a certain size: "silos of excellence." I have never been able to hear it without wincing. It confesses the disease and claims good health in the same breath. Yes, we have silos. But ours are excellent. Anyone who has worked inside a company of a few hundred people or more knows what sits behind the phrase: divisions that are sealed from each other in every way that matters. Different tools, different customers, different vocabularies, different ideas of what the company is for. They do not talk to each other, and nobody forbade it. There is simply no occasion for it, because the structure was built so that there would be none.

None of that is unusual. Many people reading this work inside a structure like it, and most of them think of it as the way companies are, the way weather is. It is not. It is a choice, and this paper is about what the choice costs.

Start by taking the choice seriously, because the people who make it are not fools. In 1933 a management consultant named V. A. Graicunas did the arithmetic on how many possible relationships a manager has to reckon with as the people under him multiply. With five direct reports, counting the pairs and groups among them, the number is about a hundred. With eight it passes a thousand. His point was that the number grows far faster than the headcount, and the practical conclusion is the obvious one: a company that grows gets cut into pieces small enough for someone to run. In 1945 the economist Friedrich Hayek made the second argument, about economies rather than companies, but it carries over: much of the knowledge needed to make a decision well is knowledge of particular circumstances, and it sits with the people on the spot, in forms that do not travel up a chain of command intact. Push those decisions to where the knowledge is, and give the people there what they need to fit the whole, and you get better decisions. Pull them all to the center, and you get slower and worse ones.

Put the two together and you have the divisional company. Alfred Chandler's history of it is the standard one. Around 1920 DuPont and General Motors were each failing in opposite ways. DuPont was a centralized company built around functions, one sales department, one manufacturing department, one finance department, and that structure could not handle several product lines at once, because decisions about any product had to pass through departments built for one. General Motors was the reverse, a loose federation of car companies with almost no central control over what they spent or built. Between 1920 and 1921 both arrived at the same answer, the multidivisional form: separate divisions, each with its own functions, each run as a business, coordinated by a head office that set goals, appraised results, and allocated money. Chandler's phrase for the lesson was that structure follows strategy. The point I want from it is simpler. The divisional form was designed. Somebody sat down and drew it, because the alternative was worse.

So the case for dividing is real. It buys management something it cannot do without: pieces small enough to run, decisions made near the knowledge, and a way to tell who is responsible for what. What it charges the company for that is the subject of the rest of this paper, and the first thing to say about the charge is that "silo" is the wrong word for it.

A silo is a container. It stands still and holds what was put in it. That is not what a division does. A division is a group of people working with their heads down, digging forward and down, the way soldiers dig a trench. They are not idle in there. They are working hard, and the digging is the work. That is what the structure asked of them. Head down, keep digging, this is your line, that is theirs.

The trouble is that trenches do not stay parallel. Nobody intends them to bend, and they bend anyway, because each trench is dug by people who can see only their own stretch of ground. They bend in one of two directions. Sometimes two trenches converge. Two divisions find themselves doing the same work for the same customer, or building the same tool, or hiring the same skills, and when they meet they fight over the ground. Companies have a polite name for this, duplication, and an impolite one, turf. I call it organizational fratricide, because the people on both sides are on the same side and the casualties are real. And sometimes the trenches diverge. Each division drifts a few degrees from the company's line, toward its own customers, its own measures, its own idea of what matters, and the drift compounds. After a few years the divisions are still digging as hard as ever, and the company as a whole is no longer moving in one direction. It is moving in five.

Heads of divisions know this, and they meet about it. They meet constantly. Offsites, operating reviews, alignment sessions, the quarterly exercise of putting everyone's priorities on one slide. And the meetings work, in the sense that the trenches are a little more parallel on the way out than they were on the way in. Then everyone goes back into the ground, and the digging resumes, and the bend resumes with it. Realignment is not a thing you do to a divisional company once. It is a thing you do to it forever, and it never quite holds, because the force that bends the trenches is the same force the structure was built to harness: people attending closely to their own piece of the work.

What happens when nobody realigns for long enough? Carl Icahn told the end of that story at a New York Times conference in 2015. In 1984 he had taken over ACF Industries, a railcar maker with a headquarters in New York that filled twelve floors of a building. He spent days walking the floors with a yellow pad, asking people what they did. He went home and looked at the pad, and in his words he could not figure out what the hell they did. He went back for another day and it was no clearer. He asked the executive who ran the operating business, and he had consultants spend three weeks on it, and then he got rid of all twelve floors. The part he tells for the laugh is what came next: nothing. Nobody called him, nobody wrote, and the company went on making railcars as before.

It is a good story and a bad lesson if you take it the way it is usually told, as a tale about an outsider's nerve. Icahn did not create the twelve floors; he found them. Somebody had let twelve floors of work grow that fed the structure and not the business. The story does not say how, and my guess is the usual way, one reasonable hire at a time, with nobody inside positioned to see the whole of it because everybody inside was in a trench. By the time a buyer with a yellow pad could remove twelve floors without anyone calling to complain, those floors were no longer serving the company; how long that had been true, the story does not say. The emergency was not the day he fired them. It had been running for some time, and the question worth asking is why nobody inside acted. In ACF's case one plausible answer is that the company's earnings were already poor and some of its acquired businesses were losing money, so the trouble had other names. That is one version of the problem. The other, and the one this paper is about, is the company whose numbers say it is fine.

Companies measure what is easy to count and what the market rewards: revenue, margin, customer counts, products shipped, headcount, growth. Those are important measures, among the most important a company has, and a company that stops watching them will not last long. That is exactly why they can betray it. They are proxies for the health of the whole, and a proxy can read normal while the thing it stands for is failing. Two trenches converging into a fight over the same customer may cost revenue this quarter, but nothing in the revenue line says why. Five divisions drifting in five directions may show up in margin eventually, and margin will not say where to look. The fragmentation is real, and it is accumulating cost, and it does not show in the numbers most boards watch, right up until it does.

The record is full of companies whose numbers were excellent while the trenches were at war. Gillian Tett tells the story of Sony, which in November 1999 brought two different digital music players to the same trade show, built by two parts of the company that had not coordinated, and then spent the following decade losing the portable music market it had built with the Walkman to Apple, while its hardware, software, and music divisions fought over how digital music should work. Sony stayed famous throughout, and the group as a whole reported a profit in eight of the ten fiscal years ending March 2001 through March 2010. Microsoft's revenue more than tripled between 2000 and 2012, from about 23 billion dollars to about 74 billion, across the same years in which, by the account of a long Vanity Fair investigation in 2012, its divisions fought each other harder than they fought competitors, its ranking system set colleagues against colleagues, and the company fell behind in search, mobile, and social. The joke of that era was a cartoon of the Microsoft org chart with each division pointing a gun at the others. The numbers did not lie. They just measured something else.

I wrote about this problem more generally in an earlier paper, The Unmeasured Layer, and argued that the administrative reality of an organization sits below the layer that gets measured, and that this is where drift begins. The bend in the trenches lives in that unmeasured layer. It can be seen, by anyone who walks the floors with a yellow pad, or by an audit that goes looking for it, but it is not what the instruments a company watches every week are built to show.

One caution, because there is a fashionable answer to this and it is the wrong one. Measuring culture has become a popular answer to whatever the numbers miss, the thing a board adds when it senses something is wrong. Engagement scores, culture surveys, dashboards of values. The trouble is twofold. First, many people who use the word do not mean anything definite by it, and they tend to assume a company has one culture, the one on the poster, when a divided company has as many cultures as it has trenches. Second, and more important, culture is downstream of systems. People behave the way the structure, the incentives, and the measures make it sensible to behave. By the time the culture survey flashes red, the structure has been producing the behavior for a long time. Culture is a lagging indicator of a design decision, and measuring it tells you that the trenches bent a while ago. Most such measures do not tell you where, or how far, or what to do.

The usual answer to the trench problem is a different drawing of the org chart, and there have been two famous ones. The first is the matrix, where a person reports to a product line and to a function at the same time. In 1990 Christopher Bartlett of Harvard Business School and Sumantra Ghoshal of INSEAD looked back at how that had gone, and their verdict was in their title: matrix management is not a structure, it is a frame of mind. Dual reporting had produced conflict, confusion, and a loss of accountability, and by 1995 the same authors could name companies that had abandoned it, Digital Equipment, Citibank, and Westinghouse among them. The matrix did not remove the trenches. It made every employee stand in two of them.

The second is the flat organization, which promises no trenches at all. The best-documented attempt is Zappos, which moved to holacracy from 2013, told every employee in March 2015 to commit or take a buyout, lost 18 percent of its people within a year, and by 2020 had quietly brought managers back. Spotify's squads and tribes, sold to companies everywhere as "the Spotify model," met a similar judgment from inside: in 2020 a former Spotify employee wrote that the company had never fully used the model itself and that nobody else should either. In these cases the flat company grew its walls back, and the matrix company added walls in the name of removing them. Both answers ask how to arrange a large workforce on paper. Neither asks whether the arranging is what makes the trenches.

And the trenches do not stop at the division. Inside every division of any size, sub-divisions form, with their own heads-down work and their own bends, and inside those, teams, and inside the teams, the two people who handle the thing nobody else understands. Draw the chart at any scale and the same shape appears.

So is it fate? The honest answer has three parts. The lines are chosen, though not always on purpose. Sometimes somebody draws the divisions, for the good reasons given above, and that is a decision with a name on it. Often the lines arrive on their own, with growth, with an acquisition, with a new function that had to sit somewhere, and nobody decides anything. But not choosing is a choice too. A line that nobody drew and nobody removes has been chosen by everyone who could have removed it, every year they left it there. The bend is not chosen. Nobody decides that the trenches will drift; they drift because of what the lines ask of the people in them, and that part is as close to fate as organizations get. The division of labor is as old as work. Adam Smith opened The Wealth of Nations with a pin factory in which ten men, dividing about eighteen operations among them, made something like 4,800 pins apiece in a day, where a man working alone and untrained in the trade might not have managed twenty, and nothing since has repealed that arithmetic. Grouping people is how human beings handle work beyond what one person can hold. So the tendency to dig, and the tendency of what is dug to bend, are given. What is chosen is everything else: where the lines go, how deep they run, what crosses them, whether anyone is watching for the bend, and, once the bend shows, whether anyone does anything about it.

That last choice is the one this paper is about, and the plainest way to see it is to think of a building. Any building, even a well-designed and well-built one, costs something to keep standing. The roof needs work on a schedule, the plumbing on another, the foundation on a third, and some of the trouble stays hidden unless someone goes looking for it, which is why buildings get inspected and not just admired. An organization's structure is the same. Divide a company and you have built something that needs maintenance in known places: the seams between divisions, the shared customer, the duplicated function, the decision that crosses a wall. Nobody thinks a building maintains itself. Companies routinely assume their structure does. Sometimes a company looks at the bill and chooses to defer it, and sometimes that is the right call for a year, the way an owner might patch a roof rather than replace it while the money is needed elsewhere. But deferring is not the same as saving, and it is never wisdom on its own. It is a decision that the bill will come due later, with interest attached, whether or not anyone is ready for it. Sometimes the interest is manageable. Sometimes it is twelve floors. And sometimes it is terminal, and the company that finally reads the bill is no longer in a position to pay it. A company does not decide to have trenches. It decides where to put them, forgets that it decided, and then decides again every year, by deferral, to leave the maintenance for later.

“A company does not decide to have trenches. It decides where to put them, forgets that it decided, and then decides again every year, by deferral, to leave the maintenance for later.”

One company wrote the connections into the design, and it is worth a look not as a model but because it shows the choice being made. Around 2002, Amazon's founder sent his engineering teams an instruction a former employee later made famous: every team would expose its data and functions through a defined service interface, teams' software would talk to other teams' software only through those interfaces, and anyone who did otherwise would be fired. The mandate did not remove Amazon's divisions. It made them harder-edged than ever, in Amazon's manner, which not everyone would choose. What it did was decide, at the moment of division, how the trenches would connect, instead of leaving that to a committee years later. Melvin Conway had described the logic in 1968: a system's design will copy the communication structure of the organization that designs it. If the structure is going to shape the work, then the connections are part of the design, whatever else one thinks of the company that did it.

Which brings me to the part this paper exists to say, and it is addressed to the people who design and sustain these structures and then spend years trying to reconnect them, though it is worth hearing by anyone who works inside one. Large organizations run programs whose purpose is to carry work across their own internal walls: the cross-functional task force, the integration office, the "one company" campaign, the alignment offsite, the dotted-line role that exists to relay messages between two trenches. Each of those is a design to compensate for a design. The people who commission them did not choose the bend, and many of them never chose the lines either; they inherited them, or watched them form. But they chose to keep them, every year, by not choosing otherwise, and that is the same decision wearing quieter clothes. There is nothing dishonest in that; dividing was probably right. What is dishonest is the accounting. A company that spends heavily on reconnection while describing its divisions as excellent is paying twice and admitting nothing, and the second payment is the one that never ends. If you are one of the people who drew the chart, the frustration you feel when two of your teams cannot work together is not a mystery about people. It is the bill for the structure, arriving on schedule. And if you are one of the people in a trench, it helps to know that the wall you keep hitting was not put there by the person on the other side of it. It was drawn, above both of you, or allowed to form there, and it will stay until someone above both of you decides otherwise.

I do not have a silver bullet, and I distrust anyone who claims one for this problem. What I have is three stances, which is a smaller thing.

First, when you divide, decide which trenches must stay parallel and say how you will know if they bend. Two divisions that serve the same customer have to stay parallel. Two that serve different markets may be allowed to drift, and the drift may be the point. The decision is easy at the moment of division and nearly impossible five years later, so make it at the moment of division, and write it down.

Second, treat realignment as budgeted work with a schedule, not as heroics after the collision. If you know the trenches bend, and you do, then the correction is a maintenance cost of the structure you chose, like the roof, and it belongs in the plan with the rest of the maintenance. The offsite that fixes everything is the sign that nothing was being maintained.

Third, measure the bend, and not with a culture survey. The measures that matter here are the ones that catch two trenches converging or drifting while it is still cheap to act: where the same work is being done twice, where the same customer hears two answers, where a decision that should take one meeting takes four because it crosses a wall, where a division's own goals have quietly stopped adding up to the company's. None of those is hard to count. They are usually not counted, because nobody assigned them to the layer that gets measured. When they show two trenches doing the same work, or one drifting into a business of its own, say so and act while the cost is small: merge them, separate them, or spin one out. The alternative is the yellow pad, an outsider walking the floors years later, unable to tell what anyone does, removing twelve floors at once because nobody inside removed one floor at a time. A company that ends that way had quieter chances first, and its numbers may well have looked fine through all of them.

None of this is a design. It is an accounting. It asks the people who choose the structure, or keep it, to carry its cost on the books instead of in a slogan, and to stop describing the walls as an achievement. The silos are not excellent. They are trenches, the lines were drawn or allowed, and the bend was left alone, and the work of keeping the trenches aligned is the price of having them. A company can pay that price openly and get a great deal for it, or it can pretend the price is zero and pay it anyway, in duplicated work, in fratricide, in a workforce moving in five directions at once, and eventually in a stranger with a yellow pad.