In 1958, Leavitt and Whisler predicted that IT would lead to the elimination of middle managers and to greater centralization of decision making. Since then, many others have speculated about how IT will affect centralization and decentralization in organizations; over the years, numerous changes have occurred in both directions. In some cases, IT appears to have led to more centralization; in other cases, to more decentralization; in still others, it appears to have had no effect at all on centralization. Previous research, therefore, gives no clear indication of IT's effect on centralization and decentralization.
Much of this confusion results from lumping together two kinds of decentralization. When we distinguish between decentralized control by unconnected (that is, independent) decision makers and decentralized control by connected decision makers, a clearer pattern emerges. Our research suggests that unconnected, decentralized decision makers should be common when communication costs are high. When communication costs fall, centralized decision making becomes more desirable. When they fall still further, connected, decentralized decision making becomes desirable in many situations.
The logic of this progression is derived from two simple assumptions:
New information technologies will significantly reduce communication costs.
Each stage in this progression requires more communication than the previous one, and in many situations, each stage has some other advantages over the previous stage.
In an era of decreasing communication costs, therefore, eventually each stage will reach a point at which its other advantages will be more important than its (diminishing) communication cost disadvantage.
This simple logic explains some of the most salient aspects of this century's economic history. According to this interpretation, the dramatic rise of large organizations in the past 100 years was motivated partly by the economic benefits of centralized decision making. In many instances, centralized decision makers could integrate diverse kinds of remote information efficiently and thus make better decisions than the unconnected, local decision makers they superseded. Centralized decision making itself was made economically feasible by advances in information technologies (not just computers and telephones but also television, radio, and other innovations). For much of this century, in fact, centralization was the only game in town. And many managers believe it still has significant benefits—witness the recent megamergers of Disney with ABC and Chemical Bank with Chase Manhattan.
In the latter part of the century, however, another kind of change is beginning to occur. Many companies are flattening their organizations by removing layers of middle managers. The remaining managers, who are often supervising significantly more people now than their predecessors did, are delegating more decisions to subordinates—the "empowerment" of the 1990s. More employees find themselves with increased responsibilities, and more managers act like coaches who help employees solve problems, rather than decision makers who issue commands and monitor compliance.
In another transformation, more work is coordinated outside the boundaries of traditional, hierarchical organizations. With large companies outsourcing noncore activities, in many industries, small companies have more important roles. Virtual corporations, networked organizations, and other shifting alliances of people and organizations are performing work that single, large organizations once handled.
Why are these changes happening? Making decisions closer to the point at which they are actually carried out ("closer to the customer", for example) has advantages and provides economic motivations for decentralizing decision making. In many kinds of work, people are more energetic and creative if they have autonomy in both how they work and what they do. Moreover, local decision makers frequently have access to information that helps them make good decisions (customers' unstated preferences, for example) but is difficult to communicate to central decision makers. Yet decentralized decision makers also need the kind of information that helps centralized decision makers make better decisions in the first place. It is precisely the communication of this large amount of information to much bigger groups of decision makers that IT now makes possible at a cost and on a scale never seen before.
Before examining this logic in detail, let's look at one example: the evolution of retailing, especially in small towns in the United States.
For most of this century, the majority of retail stores in small towns were owned and operated at local (or regional) levels."Mom and pop" operations were common, not only as grocery stores and restaurants, but also as clothing, hardware, toy, and many other types of stores. Decision making in such enterprises was necessarily decentralized to the local level. Because there was no higher-level management, each local store owner made key decisions on pricing, promotions, and product selection. And, for the most part, store owners made these decisions without knowing what was happening in other stores outside their area. It was an era of largely unconnected, decentralized decision making.
Into this seemingly placid scene came Sam Walton and Wal-Mart. By centralizing pricing, buying, and promotional decisions on a national level, Wal-Mart was able to deliver better-quality products for lower prices than most of its competitors—with the result that small towns across the United States are now filled with the empty hulls of local retail stores, driven out of business by a Wal-Mart down the street. Other factors played a role, too, but a key factor that enabled Wal-Mart to centralize its decision making was IT. With its famous state-of-the-art electronic ordering and inventory control systems, for instance, Wal-Mart introduced a new level of connected, centralized decision making into small-town retailing.
Following the three-stage pattern I introduced earlier, we might expect that some decisions would return eventually to local store managers. This has occurred, but with a big difference: Local managers now have access to national sales data and other information to help them make decisions. For example, Wal-Mart store managers have considerable autonomy in allocating space and ordering stock. Also, even though most pricing is done centrally, Wal-Mart identifies about 500 to 600 pricesensitive items for which local store managers can set their own prices, depending on what local competitors are doing. Thus it appears that the next wave in retailing may already be happening at Wal-Mart: local managers using global information to make more decentralized decisions. As Wal-Mart's CIO put it: "I think the challenge…is to enable a chain as big as Wal-Mart to act like a hometown store, even while it maintains its economies of scale".
An even more decentralized form of retailing is emerging on the Internet. Almost anyone can now set up a retail sales operation on the Internet and immediately have access to customers worldwide. Picture-Phone Direct, a mail-order reseller of desktop video-conferencing equipment, is one example."When we started our business", reported founder Jeremy Goldstein, "we thought we would concentrate on the northeastern United States. But when we put our catalog on the Internet, we got orders from Israel, Portugal, and Germany. All of a sudden, we were a global company". Another example is the Internet Underground Music Archive; its Internet site provides music samples and information about hundreds of bands and soon expects to sell compact discs on-line for home delivery. The company's rationale, in part, is to provide a distribution channel for musicians whose work is not sold in mainstream music stores such as Tower Records.
In these examples, "local" retailers make their own decisions, without supervision from any national chain or any need to appeal to a mass market. Moreover, initially small retailers have access to global markets and thus the potential to expand rapidly and dramatically.
Why should the pattern I have suggested occur? To better understand my reasoning, it helps to look at the basic information flows for making decisions in different kinds of organizations.
There are three basic types of decision-making structures: independent, decentralized decision makers; centralized decision makers; and connected, decentralized decision makers (see figure 3.1). For simplicity, I call them "cowboys", "commanders", and "cyber-cowboys".
Figure 3.1: Three Decision-Making Structures
Cowboys By definition, independent, decentralized decision makers have relatively low needs for communication. Alone on a horse, a cowboy must make independent decisions based only on what he can see and hear in his immediate environment. Similarly, when local store managers set prices by using only the information available to them locally, they don't need nationwide information systems or long-distance telephone conversations. Independent, local banks make their own loan decisions; they don't need to confer with a national headquarters before approving a loan. Individual farmers who make their own decisions about planting and harvesting don't need to communicate with anyone else either.
The price these independent decision makers pay for simplicity of decision making, however, is that their decisions are relatively uninformed. They don't know what is happening elsewhere; they aren't learning from the experiences of people in other places; and they can't easily pool resources or take advantage of economies of scale.
Commanders Centralized decision makers, on the other hand, have significantly higher communication needs. A military commander who wants to intelligently control troops from a distance needs information from scouts, the battlefield, and other sources. Likewise, "commanders" in companies need information from diverse sources to make informed decisions. For instance, the people who make decisions on national Wal-Mart prices need sales histories for the products they are pricing and detailed information about consumer tastes. Similarly, if a national bank sets its loan policies or advertising strategies at headquarters, it should communicate with local branches in order to do it well.
An obvious advantage of centralized decision making is that, with more information, people can often make better decisions. Managers can test pricing or promotion experiments in a few stores and use the results in others. They can share best practices among stores, identify the best suppliers, and capture economies of scale. Regional managers at Wal-Mart, for instance, share stories every week. As Sam Walton commented, "If they've been to that Panama City Beach store and seen a suntan cream display that's blowing the stuff out the door, they can share that with the other regionals for their beach stores".
In some cases, new technologies make it possible for individualized, local decisions to be made at a national level. For example, until a few years ago, Mervyn's grouped its local stores into a dozen categories based on sales volume, then distributed inventory based on averages for the categories. The problem with this approach was that individual stores varied greatly in the sizes and colors they sold. Some stores sold a lot of black jeans, while others needed traditional blue. To cope with these dilemmas, Mervyn's implemented a highly successful, centralized system that distributes to each store a mix of products, sizes, and colors matched precisely to local sales.
Cyber-Cowboys Connected, decentralized decision makers generally require even more communication than centralized ones. I call them cyber-cowboys because they make autonomous decisions, but based on potentially vast amounts of remote information available through electronic or other networks. These decentralized decision makers sometimes cooperate with each other; other times they compete. In any case, relevant information needs to be brought not just to one central point but to all the decentralized decision makers.
Edward D. Jones & Co., a retail brokerage firm based in St. Louis, has 3,100 sales representatives nationwide reporting directly to the national head of sales. This very flat organization makes heavy use of IT. For instance, sales reps update files and download new product information from computers in St. Louis and call headquarters frequently with client problems and questions. They are also in almost daily contact with headquarters via the company's television network, a direct link for new product information, training, motivation, and corporate culture.
One aspect of this sales force structure is its highly motivated people. "The kind of people we attract are self-starters, entrepreneurial, type A personalities, the type who might otherwise be running their own businesses", says Doug Hill, head of sales and marketing, who provides product training and support to the sales force. And what about quotas? "I don't have any quotas", says one sales rep. "I have a profitability responsibility for this territory". At Edward D. Jones, IT has enabled significant decentralization of decision-making authority while retaining the benefits of global information sharing.
More extreme examples of connected, decentralized decision making occur all the time in the interactions among buyers and sellers in a market. Whenever a company chooses to buy a product or a service from an outside supplier, rather than manufacture it internally, for example, it is using the decentralized structure of the marketplace, rather than its own hierarchical structure to coordinate production. In many cases, such market-based structures are cheaper, faster, or more flexible than internal production. For instance, two entrepreneurs compared the advantages of the vendor network in Silicon Valley with those of the larger, more vertically integrated firms on Massachusetts's Route 128:
"One of the things that Silicon Valley lets you do is minimize the costs associated with getting from idea to product. Vendors here can handle everything. If you specify something—or, as is often the case, if the vendor helps you specify it—you can get hardware back so fast that your time-to-market is incredibly short". "There is a huge supply of contract labor—far more than on Route 128. If you want to design your own chips, there are a whole lot of people around who just do contract chip layout and design. You want mechanical design? It's here. There's just about anything you want in this infrastructure".
By making potential markets larger and more efficient, IT can greatly increase the desirability of buying—rather than making—more and more things. In the 1980s, for instance, computerized airline reservation systems allowed airline companies to outsource much of their sales function to independent travel agents. Today, there are on-line markets for all kinds of products—from electronic parts to insurance to consumer appliances. These markets are allowing decentralized decision makers in many autonomous companies to participate in global markets, with access to knowledge and customers from all over the world.
Many factors affect how decision-making power is distributed in organizations: government regulations, national cultures, organizational traditions, and individual personalities, to name a few. Three factors, however—decision information, trust, and motivation—are especially important in determining the economic desirability of making decisions in different places. Let's look at how IT relates to these three factors.
Decision Information Making good decisions requires information. I have discussed how different decision-making situations have different needs for information. By reducing communication costs, IT makes structures that require more communication feasible where they would be impossible otherwise.
IT also makes distance less important in determining where decisions should be made by bringing information to decision makers wherever they are. But this does not mean that all decisions can be made anywhere with equal effectiveness. Some people are better at making certain decisions than others, and some kinds of information are inherently easier to communicate than others. A field salesperson can easily communicate the dollar volume of her sales last month, for example; she finds it much harder to communicate her sense—based on years of experience—of what kinds of new products customers want. It is easy to communicate the temperature of a container in a chemical refinery; it is hard to communicate the chemical reasoning for why a certain temperature is necessary. In general, information is easier to communicate if it is already explicit in some way—already written down, for example, or expressed in quantitative form. Information is more difficult to communicate—or "sticky"—if it is based on someone's experience or on implicit, qualitative impressions.
One implication is that companies should use IT to bring decisions to where the most important sticky information is located. Or, to put it another way, companies should use IT to bring easily communicable information (financial data, news reports, and so forth) to people who have knowledge, experience, or capabilities that are hard to communicate (customer understanding, technical competence, or interpersonal skills).
Trust If I don't trust you, I don't want you to make decisions on my behalf. That very human attitude means that centralized decision makers will avoid delegating important decisions to local decision makers, and if they have to, they will try to control or monitor the local decision makers as much as possible.
IT can increase trust (or deal with the lack of it) in several ways:
IT can make remote decision makers more effective. For example, Mrs. Fields Cookies can hire very young, inexperienced employees in its stores partly because it has centrally developed software that helps manage store operations at a very detailed level. The software helps determine quantities of ingredients and baking schedules based on seasonally and locally adjusted sales projections. It even suggests when store managers should go outside with free samples to entice customers.
IT can help control and monitor remote decision makers more effectively. Several years ago, Otis Elevator Company replaced its decentralized service system with a centralized one, so trouble calls bypassed the field service offices. This allowed executives to spot a number of chronically malfunctioning elevators whose poor records had been buried for years in field-office files.
IT can help socialize remote decision makers and engender loyalty. Edward D. Jones managers use the company's business television network to inculcate feelings of corporate identity and team spirit. By enabling personal contacts over long distances, electronic communication technologies (from telephones to e-mail to video-conferencing) can also inspire a spirit of community and a sense of loyalty in geographically dispersed organizations.
Of course, not all such uses of IT are desirable in all situations, but they illustrate how IT can help increase trust or deal with the lack of it. For example, IT can help central decision makers trust the local decision makers to implement their decisions more faithfully, or to make more decisions themselves. In that way, IT helps centralized systems become more decentralized. On the other hand, if a system is so decentralized that local decision makers make the major decisions, then IT can help local decision makers trust central decision makers (such as their centralized suppliers) more. In cases like that, IT enables more centralized systems, with some important decisions "delegated" to central decision makers.
Overall, therefore, the factor of trust leads to ambiguous predictions about the effects of IT on centralization. With regard to trust, IT can either increase or decrease centralization. In general, IT should lead to more mixed systems, with some important decisions made by central decision makers and others by local decision makers.
Motivation The kind of energy and creativity that people bring to their work often depends on who makes the decisions about what they will do. For certain kinds of work (highly routine or purely physical work, for example), people may work harder when others tell them what to do. But, in general, a big factor that makes jobs more enjoyable is some degree of autonomy. When people make their own decisions about how to do their work and how to allocate their time, they usually enjoy their jobs more and put more energy and creativity into their work. An important part of entrepreneurial motivation, for instance, is not just that you get to keep the rewards of your work but also that you make your own decisions.
Increased motivation, then, is one advantage of decentralizing decisions (and rewards) to local decision makers. Increased motivation, in turn, often leads to higher quality and more creativity in what people do. As more work becomes knowledge work and as innovation becomes increasingly critical to business success, this factor probably will become more important. Because IT can enable either more centralized or more decentralized systems, its effect on motivation is ambiguous.
The characteristics of different situations that are likely to make centralization or decentralization desirable can be used as a kind of checklist to decide which kind of decision making is desirable in a given situation (see table 3.1). Of all three factors, decision information has the clearest implications for costs and benefits. (table 3.2 summarizes the relative costs and benefits of the three different decisionmaking structures.) In general, cowboys should incur the lowest communication costs because they do the least communicating, followed by commanders, then cyber-cowboys. In addition, both commanders and cyber-cowboys enjoy the benefits of remote information, whereas cowboys do not.
Centralization is desirable when…
Decentralization is desirable when…
Using remote information is valuable in decision making, and the information can be communicated to central decision makers at moderate cost
Local decision makers have access to important information that cannot be easily communicated to central decision makers
Central decision makers don't want to (or cannot) trust local decision makers for important decisions
Local decision makers don't want to (or cannot) trust central decision makers for important decisions
Local decision makers work harder or better when told what to do by someone else (likely to be less common in the future)
Local decision makers work harder or better when they make decisions for themselves (likely to be more common in the future)
Costs of Communicating Remote Decision Information
Benefits of Considering Remote Decision Information
All Other Costs (Trust, Motivation, etc.)
Independent, Decentralized (Cowboys)
Connected, Decentralized (Cyber-Cowboys)
The costs of the other two factors, trust and motivation, are more situationdependent. The costs of lack of trust do not depend primarily on the type of decisionmaking structure but on how extensively important decisions are delegated. Similarly, the costs resulting from lack of motivation, initiative, and creativity depend on the kind of work being done. Because they are somewhat ambiguous, I have included these other two factors as part of the uncertainty concerning "all other costs". That category might also include the costs of actually making decisions (for example, the cost of salaries for decision makers) and the costs of economies of scale (or the lack thereof) that are realized by a particular decision-making structure.
How do these different kinds of costs trade off against each other for different decisions? Let's look first at the two dimensions about which we have the least ambiguity: (1) the value of the remote decision information used (that is, the cost of not considering it) and (2) the costs of communicating the remote decision information. Any decision can be plotted on a graph, depending on the average value of the remote information available and the average costs of communicating that information (see figure 3.2). For different regions of the graph, different decision-making structures are desirable. Of course, the exact shapes and locations of the regions depend on the nature of the various costs in the different decision-making structures. But the shapes and relative positions of the regions, shown in figure 3.2, follow mathematically from the assumptions in table 3.2, with one additional assumption: that the other costs of the cyber-cowboys are less than those of the commanders.
Figure 3.2: Desirable Decision-Making Structures for Different Kinds of Decisions
That additional assumption would be true, for instance, in any situation in which the motivational advantages of having entrepreneurial, local decision makers make autonomous decisions are important. These motivational factors are usually important in all kinds of management situations and in most knowledge work (sales, marketing, finance, product development, and consulting). They are even important in many physical jobs, such as assembly line work, when creativity and innovation are valuable. This additional assumption would also be true whenever local decision makers have important information that is sticky (hard to communicate), such as knowing what customers really want or understanding subtle but critical aspects of new technologies.
If the "other costs" of cyber-cowboys are higher than the costs of commanders, however, then cyber-cowboys are never desirable, and the commanders' region extends all the way to the vertical axis. For instance, if the local decision makers are very unskilled, such as the young workers in Mrs. Fields stores, then it may never be desirable to decentralize some decisions.
In using figure 3.2, we see that decisions in which remote information is too expensive to communicate relative to its value for decision making should generally be left to local cowboys who already have the information. Even in centralized, national retail chains, for instance, local store managers usually decide whom to hire as clerks. But if the remote information is valuable enough, it may be worth paying significant communication costs to transmit it somewhere else for decision making. Accounting information about the amount of money received and spent in each store, for example, is of significant value in many kinds of business decisions and is almost always communicated elsewhere, whether for centralized decision making in a single place or decentralized decision making in multiple places.
The recognition that an important effect of IT is to reduce the costs of communicating many kinds of information produces a key insight (see figure 3.2). In general, we can expect decisions to move gradually leftward in the figure as the unit costs decline for communicating the information that people use. Thus the graph suggests that many decisions will pass through a stage of being centralized before eventually moving to a structure with decentralized, connected decision makers.
This progression will not always occur. For instance, in situations in which the remote information is of only moderate value (and the other costs of centralized control are high), we might see a transition from the cowboy structure directly to cyber-cowboy. Instead of creating a chain of their own local truck-repair shops, for example, Caterpillar developed a PC-based service that lets independent truckrepair shops use a national database of repair histories for individual truck engines. Similarly, in situations where the remote information is even less valuable (and the costs of connected, decentralized decision making are also relatively high), the cowboy structure may be the most desirable, even when communication costs become zero. In the same way, if the other costs of cyber-cowboys are higher than those of commanders, then the cyber-cowboys would never be desirable and would not even appear on the graph.
In general, however, decreasing communication costs leads to movement along the path described when local decisions can be significantly improved with remote information and when either or both of the following are true:
Local decisions can be significantly improved by considering local information that is sticky or hard to communicate.
Local decision makers are significantly more enthusiastic, committed, and creative when they have more autonomy in their work.
While not true for all important decisions, these conditions appear to be true for many. Therefore, we can expect a significant long-term migration along the path described.
Leavitt and Whisler 1958.
For summaries of previous research, see, for example, Attewell and Rule 1984; George and King 1991.
For a previous paper that makes this distinction, see Anand and Mendelson 1995.
See, for example, Chandler 1977.
Stevenson 1994; see also Anand and Mendelson 1995.
Walton and Huey 1993.
Dvorak, Dean, and Singer 1994.
J. Kalb, quoted in Saxenian 1994, x.
See, for example, Malone, Yates, and Benjamin 1987.
von Hippel 1994.
For useful discussions of these issues, see, for example, Jensen and Meckling 1973; Gurbaxani and Whang 1991.
Stoddard 1986; see also Bruns and McFarlan 1987.
For a description of how electronic and other communications media are used in different ways, see Daft and Lengel 1986.
See, for example, Hackman and Oldham 1980.
The mathematical proof of this result is given in Wyner and Malone 1996.