Hold, Collide, Come Apart

This is the second of three papers about trenches. In the first, Trenches, Not Silos, I argued that a division of a company is not a silo. A silo stands still. A division is a group of people working with their heads down, digging forward, the way soldiers dig a trench. I argued that trenches do not stay parallel. They bend toward each other or away from each other, nobody measures the bend, and the company pays for it later.¹

Another paper, Nothing Stays Tuned, made a wider argument. Anything that is set to a purpose drifts. Change is not by itself a failure. The craft is in seeing which kind of drift it is, and then deciding on purpose what to do about it.¹

This paper puts the two together. The first paper said that nobody chooses the bend. It comes with the structure. It also said that two divisions serving different markets may be allowed to drift apart, and that drifting apart may be the point. If the bend comes with the structure, then much of it is normal. A company that grows will find two of its divisions working the same ground, because that is where the customers went. It will find another division heading off on its own, because it has become a different business. If nothing in a company bent for ten years, I would guess that nothing in it had grown.

So the useful question about a bend is which kind it is, and answering that takes knowing what the kinds are. The first paper drew them in one figure: lines as drawn, lines that converge, and lines that diverge. Here each one gets a full account.

It also matters when the question gets asked. Take a trench that is off its line by one degree. After ten feet it is about two inches off, and one person with a shovel can fix that before lunch. After a mile it is about ninety feet off.² By then it is a different trench, with a year of work in it and a crew that has never dug anywhere else. The first paper said to act while the cost is small. The trouble is that at two inches a bend is cheap to fix and very hard to see. So for each form I will separate the early signs, which are small, from the late ones, which everyone knows. What companies do when they see a bend late, and what that costs them, is the subject of the next paper.

Three forms, then. Two of them are drift, which the first paper called the bend. In one, two trenches drift toward each other until they collide. In the other, they drift apart. The third is what trenches do when something holds them against drift. None of the three is the healthy one, and none is the sick one. Each has a version a company can live with and a version that costs it dearly.

One caution before I begin. What follows is a way of reading a divided company. It is built from cases and from argument, and it has not been tested the way a rule is tested. Where I say what usually happens, I am describing a pattern I believe is common, and not a measured rate.

The first form is parallel. The divisions move in the same direction and stay out of each other’s way. This is the picture on the organization chart, and in the first paper it was the picture labeled year one. That paper also said that trenches do not stay parallel on their own. They bend. So parallel is not a resting state. When trenches have run side by side for years, something is holding them there, and the useful question about a parallel company is what.

The question matters because parallel is not the same as agreed. Parallel describes where the work is going. It says nothing about whether the people doing the work share an idea of where it should go, or why. A company can have the first without the second for a long time. I can think of three things that will hold trenches parallel with no shared understanding at all.

The first is a pull from outside. A market that is growing fast pulls every division the same way. There is more demand than anyone can meet, so nobody needs anyone else’s ground, and nobody has time to wander. A crisis does the same thing, and so does a single large customer or a hard deadline. The pull works whether or not anyone in the company agrees about anything, and it lasts exactly as long as the pull lasts. When growth slows, companies often say that politics and walls have suddenly appeared. They were there all along, and the growth was covering them.

The second is a person. A founder, or a strong chief, settles every question that crosses a line between divisions. The whole picture lives in one head. The divisions do not line up with each other. Each one lines up with that person, and the result looks the same. This lasts until the company outgrows what one head can hold, or until the person leaves. The first paper gave the arithmetic: the number of relationships to keep track of grows much faster than the number of people.¹ And when the person goes, the divisions find that they never learned to deal with each other, because they never had to.

The third is a rule. Someone divides the ground in writing and says who stays where. General Motors is the best case I know. In the first paper GM appeared around 1920 as a loose group of car companies getting its first real structure. One of its problems was that its own cars overlapped. Each division set its own prices, and several cars landed at nearly the same price and took sales from each other. In 1921 a committee led by Alfred Sloan wrote a policy to deal with that. GM would offer a car in each price class, from the cheapest up, with no duplication between the classes. The slogan GM used for it was “a car for every purse and purpose.” Sloan wrote later that the policy was never carried out exactly, and that the divisions always duplicated and competed with each other to some degree. But it gave each division ground of its own, and for decades it worked.³ A rule like that needs no shared understanding either. It only needs to be obeyed. It lasts as long as someone enforces it and keeps it up to date, and it fails quietly, one exception at a time, or because the ground changes and the rule does not. Nothing Stays Tuned made this point about any written reference: it has to be kept current by someone with the authority to change it, or it turns into a relic.¹

There is a fourth thing that holds trenches parallel, and it depends least on luck. It is the work of keeping a shared understanding alive. That means people in each division who know what the others are doing and why, goals that are written so that they add up, and the regular realignment that the first paper said belongs in the budget. A rule that someone enforces and keeps up to date belongs here too. This work costs money and time every year, and the cost is plain to see. That cost is the bill the first paper described. The other three are not free either. A person’s attention costs time, and a rule costs effort to enforce. But those costs sit inside other budgets where nobody counts them, so the three look free. That is why companies rely on them, and why they are surprised when one of them stops working.

What parallel buys is everything the divisional form was designed to give. Each division gets on with its work. Decisions are made near the knowledge. It is clear who answers for what. Nobody waits for anybody, and very little time goes into meetings between divisions. When its trenches are parallel, a company gets the benefits of dividing at a low coordination cost.

The costs are harder to see, because they are mostly things that do not happen. When trenches are held parallel by a pull, a person, or a rule, the divisions have no need to deal with each other, and little passes between them. What one division learns tends to stay in that division. The same kind of problem gets solved separately in each one. A customer who buys from two divisions may get nothing extra for it. Parallel of that kind gives a company the sum of its divisions and little more. The fourth kind, where the shared understanding is kept alive on purpose, makes that exchange a regular part of the work.

The second cost is that the company gets no practice. A company whose trenches have been held straight by growth or by one person has never had to see a bend or fix one. It has no habit of looking and no channel for saying what it sees. When the thing holding the lines goes, it has to learn all of that at once, under pressure.

The third cost is the worst, and it is confidence. The quiet gets read as alignment, and the leaders take credit for it. Then, reasonably enough, they stop paying for upkeep that seems to do nothing. The meetings between divisions are cut as waste. The people whose job was to work across the lines are among the first to be moved or let go. So a company that is parallel by luck tends to remove the one thing that would have kept it parallel later.

Parallel can also be held too well. Sometimes the ground moves, and the right thing for a trench to do is bend with it. A rule that keeps every division exactly where it was drawn can keep a company lined up on a plan the market has left behind. This is one reason the bend is often normal, and it is why the aim cannot be to keep everything straight forever.

What separates parallel that is held from parallel that is lucky is whether anyone can answer three questions. What is holding these trenches straight? What happens when that thing goes, when the growth slows, or the founder leaves, or the rule goes out of date? And is anyone looking for a bend now, while none can be seen?

There is also a simple question worth asking. Ask the heads of two divisions, separately, what the company is trying to do this year and what each of them needs from the other. The answers prove nothing by themselves. Matching answers can be a shared slogan, and different answers can be two honest views of two different jobs. But answers that do not fit together are a reason to look closer. The company may be parallel by luck, or it may be two inches into a bend that nobody has seen. Finding out costs an hour.

The usual mistake with parallel is that nobody thinks there is anything to fix. A second mistake follows it. When the holder goes and the trenches start to bend, the company often blames the newest thing it can find: the new chief, the new hires, a change in the culture. The first paper said that culture is downstream of systems, and this is a case of it. The people did not change. The thing that was holding the trenches straight stopped holding them.

The second form is converging. Two trenches bend toward each other until they are working the same ground. The ground can be a customer, a product, a market, a budget, or the credit for a result. In the first paper I called what happens when they meet organizational fratricide. Here I want to slow down and look at how two groups of reasonable people get there. It does not happen in one step, and each step looks different from inside.

It usually starts with the reward. Two divisions are measured on the same thing, most often growth, and the growth is in one place. So both reach for it. Nobody told either of them to stay out, and the company is paying both of them to go. Two other causes often travel with that one. Sometimes the ground itself moves: two things that customers used to buy separately become one thing, and the two divisions that sold them follow their customers to the same spot. And sometimes the line between two groups was never clear. Product and services, sales and marketing, a region and a product line: the border between groups like these is often where the work is, so both claim it.

The first stage is overlap without contact. Each group does a little of the same work, and neither knows about the other. Nobody feels that anything is wrong. Each team believes it is doing its job, and it is. The only cost so far is that the company is paying twice, and no report shows it.

The second stage is discovery. The two groups meet. It may be at a customer, where two people from the same company arrive to sell two versions of the same thing. It may be in a budget review, or when both try to hire the same person. The first reaction is usually decent. People are surprised, and somebody suggests a meeting. When the meeting comes early, this is usually the cheapest moment the company will get. Both efforts are still small. Nobody has tied a career to either one. And the question in the room is still the right question: which of us should be doing this? Sometimes the meeting comes late, after both groups have spent heavily, and then even this moment is expensive.

If nobody with authority answers that question, the third stage begins, and the question changes. Each group keeps going, because stopping now would mean giving up people, budget, and a story about growth. Each group now has a reason to exist that depends on the disputed ground. So the question stops being who does the work better and becomes who owns it. That change is easy to miss, and it is the turning point. A question about the work can be settled by looking at the work. A question about ownership gets settled by power.

The fourth stage is politics, and this is where the scheming begins. I want to be exact about why, because it is usually described as a failure of character, and I think that most of the time it is not. Once the contest is about ownership, the rival is a colleague, and the sensible moves all point inward. Keep the plan to yourself, because the plan helps the other side. Reach the boss first. Describe your own charter so that it includes the disputed ground. Hire the other group’s best people. Build your own copy of anything you would otherwise need from them, so they cannot use it against you. Nobody doing these things is being irrational. Each is doing what the structure now pays for. That is why it is so hard to stop, and why telling people to collaborate does so little. They are not confused about collaboration. They have understood their situation correctly.

The last stage is a settlement, and it comes in two kinds. In the first, the top finally rules, usually through a reorganization. One side wins, and many of the people on the losing side leave and take what they know with them. In the second, nobody rules, and the company keeps two of everything for years.

The cost of all this is much larger than the double spending. Some of the rest reaches the accounts too, as hiring, legal bills, and severance, but nothing in the accounts says what caused it. The two groups that know the most about this piece of ground are the two groups that have stopped talking to each other. Every decision that touches the ground has to go to the top, so the top becomes a court. The customer hears two answers and concludes that the company is really two companies. And over time the fight can change who gets ahead. If the people who dislike politics leave, and the people who are good at it are promoted, the company has taught itself that the way to win is to fight inward. That lesson outlasts the fight that taught it.

None of this makes converging bad in itself. The first paper said that when two trenches are doing the same work, a company can merge them while the cost is small, and a merger is often the right end for two groups that the market has pushed together. When two parts of a company reach the same ground without talking to each other, that is good evidence the ground matters. Early on, when nobody can know which approach is right, two attempts can be better than one. And a company can choose overlap on purpose.

In 1931 a young advertising man at Procter and Gamble named Neil McElroy was working on Camay soap. He found himself competing with Ivory, which his own company also made. He wrote a three-page memo about it. He did not propose ending the overlap. As the historian Thomas McCraw tells it, the plan was to give each brand its own team and to aim the two soaps at different buyers, so that they would compete with each other less. That became brand management, and consumer companies copied it widely.⁴ The overlap was seen, it was named, and it was given rules.

What separates chosen overlap from fratricide is not the overlap. It is four things around the overlap. Someone above both groups knew about it. Someone decided what it was for. There was a rule for how the ground would be divided, or for how the contest would end. And neither side was paid for the other side’s failure. With those four in place, converging is competition with a referee. With none of them, it is a fight that nobody in authority is watching. Most cases sit somewhere between. And the four do not make the overlap a good idea. A leader can approve a contest that turns out to be a mistake. What they do is make sure that someone can see the mistake and has the standing to end it.

General Motors shows what happens when a rule like that is left to wear away. The overlap that the policy of 1921 had limited grew over the decades, and the differences between the divisions’ cars shrank.³ I do not think anyone decided that. Each division was judged on its own results, and each step toward a wider range of cars made sense where it was taken. By August 1983 the result was on the cover of Fortune: four mid-size cars from Chevrolet, Pontiac, Oldsmobile, and Buick, parked side by side in the same dark red and very hard to tell apart. GM was still the largest seller of cars in America by a wide margin.⁵ The numbers were fine.

Seventeen years after that cover, in December 2000, GM announced the end of Oldsmobile. It recorded an after-tax charge of 939 million dollars for the phase-out in its results for the last quarter of that year. That figure is the charge recorded at the time, and not a final total. Oldsmobile had about 2,800 dealers, and settling with them took years and went to court.⁶ During its bankruptcy in 2009 the company set out to trim duplicate products again.³ Oldsmobile’s decline had more than one cause, but a brand that sells much the same car as its sister brands has little ground of its own. I cannot name the year when this would have been cheap to fix, and I do not think anyone can. That is the trouble with a bend at two inches. What can be said is that each year added more that would have to be undone: factories, dealers, and customers built around brands selling much the same car.

This is also why the usual fix fails. The usual fix is to name an owner and tell the other group to stop, without changing what either group is measured on. The group that lost still needs growth and still knows where the growth is. So the work goes on under another name, and the fight comes back a year later with more bad feeling behind it. Nothing Stays Tuned made the same point about restructuring in general: moving the boxes on the chart leaves the problem in place when the rewards and the authority stay where they were.¹

The signs follow from the stages. The early ones belong to the first two stages, and they are small. Two teams build or buy the same tool, which means two teams have the same problem. Two divisions post the same job. Two divisions count the same sale, so the overlap shows in the sales numbers before it shows on the chart. One customer gets two calls from the same company. The customer is usually the first to see two trenches converge, and customers say so, if anyone is collecting what they say. The middle signs belong to the third stage. Questions about who owns something begin to travel upward, and two groups that used to share their plans stop sharing them. By the time there is open war over turf, the company is in the fourth stage, and the cheap moments are behind it.

The third form is diverging. Two trenches bend away from each other. The people in each are still working hard, but every year they share a little less: fewer customers, fewer goals, fewer people who have worked on both sides. The first paper described where this ends, with a company that is no longer moving in one direction but in five. Here I want to look at how it gets there, because diverging differs from converging in one way that matters more than any other. It is quiet.

Converging trenches announce themselves. Sooner or later they collide, and a collision gets noticed. Diverging trenches never collide. Each step apart makes each division’s own work a little easier, because it needs the others a little less. There are fewer meetings, fewer arguments, and fewer things to wait for. From inside, that feels like progress. So nobody complains, and nothing in the company raises a warning. When the complaint finally comes, it often comes from outside.

It starts with a pull that is entirely proper. Each division is pulled by its own customers. Serving them well means taking on their timing, their prices, and their way of talking. That is the division doing its job, and it is the reason the first paper gave for dividing at all: decisions get made near the knowledge. Three other things add to the pull. The measures are set division by division, so a division can meet every target it has while moving away from everyone else. The businesses run on different clocks, and one set of company rules about spending, hiring, and risk suits the business with a three-month cycle and gets in the way of the one with a ten-year cycle. And the company’s own direction is often stated so broadly that every division can claim to be following it. A direction that everyone can claim to follow is not holding anyone.

The first stage is a few degrees. Each division tunes itself to its own customers. The differences are small: a tool here, a process there, a word that means one thing in this building and something else in that one. Nothing is wrong. Each division is getting better at its own work.

The second stage is separate ways of working. Each division builds its own version of things the company used to have one of: its own planning calendar, its own systems, its own idea of what counts as a customer or a sale. People stop moving between divisions, because what they know in one no longer applies in the other. The services the company shares begin to fit badly. They were built for the largest division and are awkward for the rest, so the rest ask for exceptions. Each exception is reasonable.

The third stage is separate goals. The divisions’ goals no longer add up to a goal for the company. The company’s plan becomes the divisions’ plans placed in one binder. The work of the people at the top changes with it. They stop directing and start allocating. They hand out money and settle claims. Ask in each building what the company is for, and the answers differ.

The fourth stage is separate identities. People say they work for the division, not for the company. Shared costs are spoken of as a tax. The company’s name is treated as a limit and not a help. Divisions ask to buy outside what the company provides inside, to hire under their own name, and to keep what they earn. Customers who need something from two divisions have to join the pieces together themselves.

The last stage is separate businesses under one roof, and sometimes they compete. One division sells against another, or one division’s plan can only succeed if another’s fails. From here there are three ends. The company breaks up. Or it forces the parts back together at great cost. Or it carries on as a holding company that owns several businesses and directs none of them.

The largest cost is not the duplication, though there is plenty of it. The largest cost is that the company struggles to do work that needs more than one division. The reason to be one company is that the parts can do something together that they could not do apart. Divergence removes that and keeps the cost of staying together. The headquarters is still paid for. The shared services are still paid for. What the company gets for the money is several businesses that happen to share a name. This cost stays hidden until something demands a joint answer. A rival offers in one product what three of the company’s divisions do separately, and the company finds that it cannot answer well, because a good answer would take all three working as one. Sony, in the first paper, had the hardware, the software, and the music under one roof.¹ Its joint answer, an online music store tied to its own players, opened in 2004, a year after its rival’s store, and closed in 2008.⁹ Losing that market had more than one cause. But the parts were all there, and they did not come together in time.

None of this makes diverging bad in itself. The first paper said that two divisions serving different markets may be allowed to drift apart, and that drifting apart may be the point. A division held close to its sister divisions is often worse at its own work. And a new business often has to get away from the old one. Measured by the old business’s numbers, the new one looks small and unprofitable, and it tends to lose the arguments about money. In 1980 IBM’s chief executive gave one of his managers a year to bring a personal computer to market. IBM’s normal process for a new product took four or five years. So the team, based in Florida and far from headquarters, was allowed to work entirely outside the company’s standard procedures. It bought its chip and its operating system from outside companies and sold through outside stores as well as IBM’s own. The IBM PC was unveiled in August 1981.⁷ That was divergence chosen on purpose, by someone with the authority to choose it.

What separates chosen divergence from fragmentation is, again, a short list of things around it. Someone with authority decided that this division may go its own way. Someone said how far, which means saying what stays shared no matter what. The division’s goals still include something it owes to the whole. And someone keeps asking the question that divergence always raises: is there still something these parts do together that they could not do apart? If the answer is yes, that thing has to be protected with real rules and real rewards. If the answer is no, the first paper named the honest end: separate them, or spin one out. A separation chosen early can be made on good terms. A separation arrived at late is a breakup.

This is why the usual fix fails. The usual fix is a campaign. The company announces that it is one company. There is a new statement of values, a council that crosses the divisions, and a target for savings from working together. Nothing changes in what each division is measured on, and nobody has decided what the divisions are for together. So people are asked to feel like one company while everything they are paid for tells them otherwise. The other usual fix is the opposite one, which is to pull the decisions back to the center. That stops the divisions from drifting apart, and it also ends the fit with the market that the drifting had bought.

The signs follow from the stages, and because diverging is quiet, they have to be looked for. Early on, people stop moving between divisions. Two divisions use the same word for different things. A division asks for an exception to a company rule, and then for another. The head of one division cannot describe what another division does, and has no need to. In the middle stage the company’s plan is the divisions’ plans in one binder, and the meetings at the top are about dividing money, not about direction. It takes two sentences to say what the company does, where it used to take one. Late on, many of the signs come from outside. A customer asks why there are two contracts and two people to call. An investor adds up the value of the parts and gets a larger number than the value of the whole.

The two bends are not opposites. One feeds the other. Divisions that stop sharing a purpose still share a budget, so they diverge in direction and converge on the money. Sears is the clearest case, because there the separation was created on purpose. In 2008 its chairman split the company into more than thirty units. Each had its own management and its own profit and loss, and executives’ bonuses were tied to the profit of their own unit. The idea was that units acting in their own interest would make the whole company stronger. Five years later a reporter for Bloomberg Businessweek interviewed more than forty former executives. They described units fighting each other for space in the company’s advertising circular and for a share of less and less money. One of them said the units behaved like tribes at war.⁸ Sears had other troubles, and I do not claim the structure caused all of them. But this is what separate and competing businesses under one roof look like. It shows both bends in one company: more than thirty units set apart on purpose, and all of them fighting over the same money.

These are forms, not stages. A company does not pass through them in order. It has many pairs of trenches, and at any moment some pairs are parallel, some are converging, and some are diverging. The answer also depends on where you stand. Two divisions can look parallel to the people at the top and be fighting over a customer at the working level. And as the first paper said, the trenches do not stop at the division. Teams inside one division take the same three forms.

Set side by side, the three forms have three things in common.

The first is the main claim of this paper: the cause is more often the arrangement than the people. Parallel trenches are held by a pull, a person, or a rule. Converging trenches are bent by rewards that overlap. Diverging trenches are bent by customers, measures, and clocks that differ. In each case people are mostly doing what their situation makes sensible. If that is right, it explains why replacing the people so often changes less than expected, and why the first paper called culture a lagging sign of a design decision.¹

The second is that the version a company can live with and the version that damages it look the same at the start. Chosen overlap and fratricide both begin with two groups on the same ground. A new business given room and a company coming apart both begin with a division going its own way. What separates them at the start is not the shape. A large part of it is whether someone with authority knew about it, decided what it was for, and said how it would end. That does not make the decision right. A chosen overlap or a chosen separation can still be a mistake. But a choice that someone made can be reviewed by someone, and a bend that nobody chose may have no clear owner for its review.

The third is that each form gives small signs early and loud ones late. The small ones are a tool built twice, a customer called twice, a word that means two things, a transfer that no longer happens. Somebody in the company almost always sees them. They are seldom collected, and they seldom reach anyone who can act.

So this paper ends with a smaller request than the first one did. The first paper asked companies to carry the cost of their structure on the books. This one asks them to know, for each pair of divisions that matters, which form the pair is in and whether anyone chose it. That will not straighten anything. Trenches bend because of what the lines ask of the people in them. But a company that can name the form has a choice about what to do next, and seeing it early may leave cheaper choices open. What most companies do instead, and what it costs them, is the subject of the next paper.