• Archive

Making the Decision to Decentralize

Is your business better off with a central command structure or decentralized? Your future depends on the right answer. An excerpt from the new book, The Future of Work.

Author Thomas Malone, a professor at MIT Sloan School of Management, says that the cheap cost of communication—e-mail, instant messaging, the Internet—is making possible a new type of organizational structure. This organization of the future will be decentralized, the term defined as "participation of people in making the decisions that matter to them." Decentralization brings with it increased productivity and quality of life. But decentralization isn't right in every situation. In this chapter from Malone's new book, he asks: When should you decentralize?—Ed.

Soon after Lou Gerstner became CEO of IBM in 1993, he made what he calls probably the biggest decision of his entire career.1 At the time, many people in IBM and the business press were convinced that the best course for the "lumbering dinosaur" was to break itself up into smaller companies. By decentralizing in this way, they said, IBM would obtain the benefits of smallness that it sorely needed—things like flexibility, speed, and entrepreneurial motivation. And the market would be able to coordinate the interactions of the resulting companies better than IBM's corporate executives could.

But Gerstner became convinced that the best choice was to do exactly the opposite: keep IBM as a single large company and use its unique size and capabilities to help customers integrate the diverse components of their information technology (IT) systems. In other words, he wanted to use the hierarchical decision-making structures of an integrated IBM to help coordinate all the IT decisions that customers would otherwise have to make on their own (or hire someone else to make for them).

We now know, of course, what happened. Gerstner loosened up IBM's organization but did not break it apart. And his plan worked. IBM's stock price increased by almost a factor of ten during Gerstner's tenure, and many people credit him with pulling off a stunningly successful turnaround against very steep odds.

Of course, we don't know what would have happened if a different CEO had gone ahead with the breakup plan. The spin-off companies might have been even more successful collectively than the integrated IBM was. But we do know that Gerstner's choice to keep the company centralized was extremely successful.

You might think that Gerstner would be a zealous advocate of centralization since it served him so well in this situation. But even within IBM, he advocated substantial decentralization: "Let's decentralize decision making wherever possible, but . . . we must balance decentralized decision making with central strategy and common customer focus."2 Even more surprising, he believes that the success of centralization at IBM was unusual: "CEOs should not go to [the level of integration IBM did] unless it is absolutely necessary" [Gerstner's italics].3 This level of integration, he believes, is a "bet-the-company proposition." It is often tried, he observes, but almost never succeeds.

You may never face choices about centralizing or decentralizing on the scale Gerstner did, but if you're like most managers, you face such decisions on a smaller scale all the time. How should you make them? How can you tell whether your situation is one for which decentralization makes sense? And if you're going to decentralize, how can you know which kind of decentralization will work best?

In many cases the best solution is to create a custom system that combines elements of more than one basic structure.

As Table 8-1 shows, centralized hierarchies and the three basic types of decentralization—loose hierarchies, democracies, and markets—each have strengths and weaknesses. When you need to economize on communication costs or when resolving difficult conflicts of interest is critical, centralized hierarchies may be best. When you need to maximize employee motivation and creativity or tap into many minds simultaneously, markets are especially attractive. When aspects of all four dimensions are important, the two intermediate structures (loose hierarchies and democracies) may work well.

In many cases, however, the best solution is to create a custom system that combines elements of more than one basic structure. You can, for instance, use different structures for different types of decisions. That's what often happens in internal markets: The basic operational decisions are made through the decentralized market, but hierarchical managers choose the participants, set the ground rules, and intervene when the market would otherwise fail to do what is best for the organization overall.

Assigning different decisions to different structures isn't easy; it requires a detailed understanding of your own specific situation and goals. But…there is a systematic way to think about the problem.4 For each major kind of decision your company makes, you can ask yourself the following three questions: (1) Are the potential benefits of decentralizing important? (2) Can you compensate for the potential costs of decentralizing? (3) Do the benefits of decentralizing outweigh the costs? Let's look at each of these questions in turn.

Are the potential benefits of decentralizing important?
As we saw earlier, decentralization has three general benefits: (1) It encourages motivation and creativity; (2) it allows many minds to work simultaneously on the same problem; and (3) it accommodates flexibility and individualization. The importance of these benefits varies greatly, but they are often especially important in certain industries and business functions. For example, the success of most professional services organizations (such as consulting, software development, and law) hinges on the motivation and creativity of their professionals. Consequently, these organizations are especially good candidates for decentralized decision making. Creativity and innovation are also often particularly important in functions like engineering, sales, product design, and information technology. Here, too, decentralization will often pay off.

But as more work in our economy becomes knowledge work, and as innovation becomes increasingly critical to business success in many industries, the benefits of decentralization are likely to become important in more and more places.5 In fact, in principle, almost any business activity could benefit from having highly motivated, creative people performing it. Much of the early work in the Total Quality Movement, for example, was about encouraging assembly line workers to look for ways to innovate and improve the routine processes they performed.

So the question of whether the benefits of decentralization are important in your situation is not a purely objective one. It is also a matter of your strategic choices. Different people in the same situation can make different choices about how much they want to rely on the advantages of decentralization. Mrs. Fields Cookies tries to systematize and centrally control almost all the decisions needed to operate its local stores, while Wal-Mart tries to give significant autonomy to its local workers.6 Either strategy can work well, but you have to pick one and use it consistently.

Can you compensate for the potential costs of decentralizing?
You may be thinking, "Sure, sure, all this decentralization stuff sounds great in theory, but how often could it actually work? How can you make decisions effectively when no one is really in control? How can you guarantee quality or protect your company against catastrophic losses if no one is watching over things? How can you take advantage of economies of scale or knowledge sharing, if everything is so fragmented?"

These concerns are important—sometimes so important that they'll lead you to reject decentralized structures and stick with rigid hierarchies. Often, though, there are creative ways to deal with the potential downsides. Let's look at the four main problems with decentralization and the possible solutions.

How can you make decisions quickly and efficiently when no one is in control?
Sometimes, it just takes a long time to involve everyone in joint decisions and resolve all their conflicting desires. Cheaper and faster communication, through e-mail, for example, helps temper this problem. But even when the transmission of information is free and instantaneous, it still takes time for people to send and comprehend the information. And no matter how much people communicate, they still won't all agree on every question. Each of the decentralized structures offers different ways to make decision making more efficient.

In loose hierarchies, you, as a manager, can sometimes force decisions on people, even when not everyone agrees. In an economic downturn, for instance, you might decide for yourself which groups to cut, instead of waiting for the groups themselves to make such a difficult decision.

If you're a good manager in a loose hierarchy, you probably won't force decisions very often. Sometimes, you'll have to force a decision, such as when a decision is taking too long, when it looks as if there will never be enough agreement, or when people are spending so much time arguing they're not doing their other work. But the rest of the time, you should let people work things out for themselves.

In democracies, you can make decisions more efficiently in two ways. You can let the employees elect managers to make decisions on their behalf, as the partners of many law firms and consulting firms do in electing managing partners. Or you can let people vote directly (or via opinion polls) on the most important decisions, as the Mondragon cooperatives sometimes do.

In markets, decisions are often made efficiently because only two parties—a buyer and seller—need to agree for a transaction to occur. If an earthquake disables one of your factories, for instance, and your company has an internal market, then pairs of buyers and sellers can start trading with each other right away to solve the problem. They don't need anyone else to agree about what to do.

But for a market to work well, everyone who participates has to agree on the rules of the game. Markets need legal frameworks to resolve disputes between buyers and sellers, and they need regulatory systems to prevent activities (like pollution, price fixing, misleading accounting, or deceptive advertising) that make the whole market less efficient. In external markets, governments usually provide the rules. But, as we saw with Visa International and eBay, other organizations like trade associations, market makers, or standards bodies can also set rules. In internal markets, the rules are established and enforced by the managers of the company.

How can you guarantee quality or protect against catastrophic losses if no one is in control?
Many people assume that quality assurance and risk management require someone to be in control. But that isn't always true. When the right incentives are in place, just sharing information can be enough to maintain quality and temper risk. Suppose that in your company, the bonuses for everyone who deals with customers depend partly on customer satisfaction ratings. And suppose that everyone in the company can easily call up a page on the company intranet to see the customer satisfaction ratings for each store and salesperson. Just by setting up a system like this, many service-quality problems are likely to take care of themselves without any centralized intervention. Social and other pressures will push people to excel.

Sharing information can work in loose hierarchies, democracies, and markets. But each decentralized decision-making structure also offers other ways to manage risk and quality. If you're a manager in a loose hierarchy, you don't have to watch over or sign off on every action your subordinates take. This freedom allows you to focus on controlling the quality of people and measuring results. For instance, you can devote more attention to whom to hire and promote and how to reward them for the results you want.7

In democracies, you can elect managers to watch quality and risk. Or you can let the members of a group vote—taking into account quality and risk, as well as other factors—on whom to hire and promote and how to allocate rewards. Many consulting and law firms, for instance, elect their new partners by a vote of all the existing partners.

In markets, you can control quality in two ways. First, you can use online reputation systems (e.g., those used by eBay, Elance, and Asynchrony) to help people pick high-quality providers in the first place.8 When online reputation systems become widely used, the traditional signifiers of quality, like brand names, are likely to become less important. Actual user ratings give buyers a much more accurate and efficient way of judging quality than relying on their general knowledge of a brand. Which would you rather buy: (a) a television with a well-known brand (e.g., Sony), even though previous buyers and objective raters like Consumer Reports rate the set poorly, or (b) an unknown-brand television (e.g., from Joe's No-Name Appliances) that gets wildly enthusiastic ratings from most previous buyers and objective raters?

In addition to reputation systems, the other primary way to manage quality and risk in markets is with various financial instruments: insurance, performance bonds, pools of risk capital, and other kinds of collateral. One of my former students, for instance, used to work in the credit card area of CapitalOne, a large financial services company with a decentralized, entrepreneurial culture. This student really appreciated the freedom that individual analysts had there to make pricing and credit policy decisions for massive mailings of credit card offers. But in 2002, government regulators forced CapitalOne to institute numerous centralized controls and approval processes designed to reduce the risk of huge credit card losses.9 In my student's view, this involuntary centralization seriously damaged the unique entrepreneurial culture and strengths of the bank.

Could CapitalOne have managed this risk in other—more decentralized—ways? I think so. Here is one possibility: Instead of having a centralized manager sign off on the terms of every mailing, each analyst could have a pool of risk capital. If you were an analyst and wanted to make a mailing in which the total credit offered was below your risk capital limit, you could proceed with no other approvals. And you could still exceed your own limit without centralized approval by assembling a syndicate of peers who together were willing to contribute enough of their own risk capital to cover the mailing. In undertaking a huge risk, you might still have to get approval from a higher-level manager, but you and your peers could manage most of your own risks in a decentralized way.

How can you take advantage of economies of scale if everything is decentralized?
Many times, people assume that just because there are economies of scale in one part of a process, the whole process has to be centralized. But you can often get the benefits of both bigness and smallness by centralizing only those decisions involving important economies of scale and decentralizing everything else. In semiconductor manufacturing, for example, there are major economies of scale—Intel now spends about $2.5 billion to build a fabrication plant.10 But this doesn't necessarily mean that similar economies of scale apply in everything else Intel does. There's no reason, for instance, why the design of semi-conductor chips couldn't be much more decentralized. In fact, some of Intel's competitors, like the Taiwan Semiconductor Manufacturing Company (TSMC), take this idea to an extreme by providing only semiconductor-manufacturing services. Its customers, ranging from tiny start-ups to huge multinationals, design their own chips and then pay TSMC to manufacture them.

Even when economies of scale apply, you can sometimes achieve them with very little centralized control if you follow two key practices: Share information widely, and provide incentives that encourage scale economies. Many companies assume, for instance, that to achieve economies of scale in purchasing, they need to centralize purchasing decisions. By forcing all the different parts of their company to buy from the same vendors, they get much bigger volume discounts.

But what if, instead of forcing everyone to buy from the same vendors, you just provide incentives for people to form voluntary purchasing groups? If I don't much care, for instance, what kind of personal computer I have, I could just delegate my personal computer purchasing decision to a PC purchasing specialist and automatically get whatever volume discounts that person can negotiate. If I do care, I could look at an online database of the different PC purchasing plans available in my company and decide which one is best for me. In this scenario, the central purchasing people could still have a job organizing voluntary coalitions of buyers, maintaining a database of available purchasing plans, and negotiating volume discounts for the people who choose to participate.

Of course, if the incentives aren't right, this arrangement won't work well. I might, for instance, choose my own favorite PC vendor, even when this is really not the best choice from the company's point of view. But if I am measured and rewarded on the basis of my contributions to corporate profit, then I can balance the potential cost savings for the company with all the other factors that are important to me.

In general, the three decentralized structures allow individuals to make their own decisions about economies of scale. But, in each structure, you sometimes need to restrict individual decisions to encourage economies of scale (e.g., with utilities) or to prevent abuses of power (e.g., with monopolies). In loose hierarchies, the managers do this. In democracies, it's done by elected managers or popular vote. In markets, some kind of regulatory infrastructure does it. For instance, in an internal production-capacity market with only a single factory, the corporation might regulate the factory like a public utility rather than letting its managers charge whatever price the internal market would bear.

How can you enjoy the benefits of knowledge sharing without centralized control?
One of the most important advantages of being in a big organization is having ready access to many sources of knowledge. If you're the owner of an isolated hamburger stand in a small town in New Mexico, you have only your own ideas and experience to guide you in running your restaurant. But if you're the manager of a McDonald's in the same town, then—at least theoretically—you have access to the best burger-selling knowledge available anywhere in the world.

Of course, big companies don't always take advantage of their full potential for knowledge sharing. And even when they do, gaining the benefit of knowledge sharing doesn't require centralized control at all. It just requires the widespread sharing of knowledge. When communication is difficult and expensive, the best way to share knowledge may be to have centralized managers find and spread the best ideas from different parts of their organizations. But when communication is cheap and easy, it's often better to have people share knowledge directly through many different channels. For example, independent restaurant managers can share knowledge with each other through trade association meetings, online discussion groups, best-practice databases, and so forth.

In each decision-making structure, you have different ways of encouraging broad knowledge sharing. In loose hierarchies and democracies, one of your most important roles as a manager is often to provide the channels and incentives for sharing knowledge. For instance, when some consulting firms do their annual employee performance evaluations, they take account of individuals' contributions to corporate knowledge bases.

To share knowledge in markets, you need effective ways to buy and sell it. The effective sharing of knowledge is the purpose of intellectual property laws, like patents and copyrights. Some people are surprised to hear that intellectual property laws are meant to help share knowledge. They think, for instance, that if you can't share music for free over the Internet, the laws are decreasing your opportunities for knowledge sharing. But when designed well, legal infrastructures like copyright laws give people much stronger economic incentives to create knowledge and package it for sharing.

Do the benefits of decentralizing outweigh the costs?
Once you've worked out the benefits and costs, you need to weigh them to decide whether decentralization will pay off. Here, again, however, there are no simple answers—much depends on your particular situation. But some simple rules of thumb can help you think through the choice.

Decentralize when the motivation and creativity of many people is critical. We've already seen examples of this principle in action in AES, consulting firms, internal markets for allocating semiconductor manufacturing, and so on. This principle becomes especially important in certain situations. When your company is growing rapidly, for instance, it is often a good idea to encourage people to creatively explore new possibilities. When your industry is in the midst of rapid change, the best way to figure out how to respond is often to let many highly motivated people try many experiments. And when small groups in your company work independently from other groups (e.g., in a consulting firm or a research university), it's often a good idea to decentralize most decisions to these small groups to spur their innovativeness.

To share knowledge in markets, you need effective ways to buy and sell it.

Centralize when resolving conflicts is critical. When the key issue for your organization is not encouraging creativity, but resolving conflicts, then centralizing may be the better choice. For instance, Goran Lindahl, former CEO of highly decentralized ABB, says that you often need to centralize when an organization is shrinking.11 In times of contraction, hard choices need to be made about where to cut and what to change. Although it's not impossible to make these decisions in a decentralized way, getting people to agree to give up things—including their jobs—is usually much harder than handing down orders from on top. AES, for example, appears to be moving to more centralized decision making now that its whole industry is shrinking.12

Centralize when it's critical to have lots of detail—down to a very low level—united by a single vision. Even though sometimes thousands of people are involved in filming and editing a movie, the director usually needs to make many very detailed decisions to be sure that the final film embodies a single artistic vision. Sometimes, this principle applies in decisions about business strategy as well. For example, you could argue that Microsoft's successful strategic shift to embrace the Internet in the mid-1990s was possible because Bill Gates both understood the details of his business and had significant centralized power.

Centralize when only a few people are capable of making good decisions. Sometimes people think decentralization means automatically pushing all decisions lower in an organization, whether or not the people at the bottom have the skills to make the decisions wisely. But such an approach is not decentralization; it's just stupidity. With some decisions, no matter how much information you have, you still need special skills or knowledge to make the right choice. Most hospital patients, for instance, want well-trained physicians to make the key decisions about their medical care, even though the nurses on the floor may have much more detailed patient information. Similarly, some business decisions benefit from the kind of judgment that comes only from years of experience. Such decisions often need to be centralized. Even in such cases, however, a business may benefit from developing good decision-making capabilities in more people. You might be surprised at what some people can do when given the right opportunities to develop their skills.

Excerpted with permission from The Future of Work: How the New Order of Business Will Shape Your Organization, Your Management Style, and Your Life, by Thomas W. Malone. Harvard Business School Press, 2004.

[ Buy this book ]

Thomas W. Malone is the Patrick J. McGovern Professor of Management at the MIT Sloan School of Management and the Founder and Director of the MIT Center for Coordination Science.

Evaluating Decision-Making Structures

by Thomas W. Malone

Evaluating Decision-Making Structures

Excerpted with permission from The Future of Work: How the New Order of Business Will Shape Your Organization, Your Management Style, and Your Life, by Thomas W. Malone. Harvard Business School Press, 2004.


1. Louis V. Gerstner Jr., Who Says Elephants Can't Dance? Inside IBM's Historic Turnaround (New York: HarperBusiness, 2002), 12-13, 57-62, 68-70.

2. Ibid., 22.

3. Ibid., 248-252.

4. For related work that contributed to the development of this approach, see Jay R. Galbraith, Designing Organizations: An Executive Guide to Strategy, Structure, and Process, 2nd ed. (San Francisco: Jossey-Bass, 2002); K. A. Merchant, "The Control Function of Management," Sloan Management Review 23, no. 4 (spring 1982): 43-55.

5. On the growing importance of knowledge work, see, for example, Peter F. Drucker, Post-Capitalist Society (New York: HarperBusiness, 1993).

6. Thomas W. Malone, "Is 'Empowerment' Just a Fad? Control, Decision-Making, and Information Technology," Sloan Management Review 23: 38, no. 2 (1997): 23-35; M. Stevenson, "The Store to End All Stores," Canadian Business Review, May 1994; B. Fox, "Staying on Top at Wal-Mart," Chain Store Age Executive 70, no. 4 (1994): 47; Thomas Richman, "Mrs. Fields' Secret Ingredient," INC. Magazine, October 1987, 65-72.

7. Merchant, "The Control Function of Management."

8. See, for example, Chrysanthos Dellarocas, "The Digitization of Word-of-Mouth: Promise and Challenges of Online Reputation Mechanisms," working paper, MIT Sloan School of Management, Cambridge, MA, 2002; Paul Resnick et al., "Reputation Systems," Communications of the ACM 43, no. 12 (2000): 45-48.

9. Capital One Financial Corporation, "Securities and Exchange Commission Form 8-K," 16 July 2002, available at (accessed 2 June 2003).

10. Brent Schlender, "Intel's $10 Billion Gamble," Fortune, 11 November 2002, 90.

11. Goran Lindahl, conversation with author, Carmel, CA, 24 June 2001.

12. Rebecca Smith, "AES, Calpine Post Losses for the Quarter: Results Reflect Electric-Power Industry's Tight Credit, Instability, Declining Margins," Wall Street Journal, 14 February 2003.