5—
Turning Inventions into Innovations: Schumpeter's Organizational Sociology Modernized

Introduction

The general purpose of this chapter is to elaborate the fundamental distinction made by Joseph Schumpeter (1942) between invention and innovation. Since Schumpeter was interested in the transformation of the economy by the development of new technology, he was concerned only with innovations that could produce a continuous effect in the market, not with inventions that remained visions of isolated discoverers. He made the distinction between invention and innovation to distinguish those new ideas that revolutionized the economy from those that did not.

The particular lines that Schumpeter drew between inventions and innovations are not very useful in a modern economy. The basic problem is that Schumpeter thought that anything that could be done by large-scale bureaucratic organizations could be done routinely. Since (according to Schumpeter's argument) it was exactly the nonroutine character of innovations that produced the economic effects of monopoly, entrepreneurial profits, business cycles, and economic progress, this specification of the problem led him to predict the disappearance of many of the relevant economic effects as innovation became routinized (1942, 131–134). Even if innovation continues, he argued, its routinization destroys the entrepreneurial function; this destruction in turn undermines the argument in defense of capitalism that entrepreneurs are needed for economic progress (if innovation is only bureaucratic routine, then socialist governments can carry it out just as well). The routinization of innovation also means that innovations can spread rapidly in the market. This creates a lowered expectation of monopoly profits, which previously motivated excess investment in boom times and so decreases the amplitude of business cycles.

The argument below is that building a social system around an inno-

vation is still not routine, even though it is (sometimes) done in large organizations. Not all organizations successfully innovate, but when they do monopoly profit positions and a heady flow of new investment are established, because not everybody can follow their lead. Turning inventions into innovations is difficult, because inventions need a particularly complex information and decision system connected to them to become innovations. For example, all organizations must create incentives so that workers will work hard to do what needs to be done; but creating incentives for the personnel who have to solve all the nonroutine problems of a division producing an innovation is a substantially different, and more complex, problem. Innovating workers and managers have to work into the night, go out and help the salesmen when the solder comes loose, help engineers understand marketing problems, and in general be motivated and authorized to take more responsibility than is needed or tolerable in a more routine production operation. This means that an incentive system that works fine for routinized production will not serve for producing an innovation, which in turn means that the information and decision problems for an incentive system in an innovating part of an organization will need special attention.

We have already argued that routine industrial administration is not so routine as all that (Chapter 3 above; and see Stinchcombe 1974, 3–39). Below we will argue that there are many structural impediments to be overcome in building the social organization that can carry an innovation—that can, in the phrase John R. Commons (1924) adapted from legal terminology, make the innovation into a "going concern."

In the first sections we will ask why it is that innovations create monopolies. That is, we develop variables that determine how long the monopoly created by an innovation is likely to last and, consequently, that determine the size of the economic impact. Schumpeter argued that the monopolies produced by innovation set in motion a disequilibrating cycle of profits, investment, and reestablishment of competition. Under some conditions innovations create larger and more durable monopolies than under other conditions. Longer duration and larger size of monopoly advantage both increase the incentive for introducing an innovation and increase the duration of the market's departure from competitive conditions, making the business cycle effects larger.

Many of the variables we identify have social components, or are technical and economic determinants that take social form. For example, it is quite clear that the learning curve of cost reduction after the introduction of an innovation creates a monopoly advantage for innovators, because at any given time they can expect to be further down the learning

curve and so to have lower costs than potential competitors. But the actual process of cost reduction seems to be mainly one of introducing minor labor-saving and capital-saving innovations into the production process. Introducing such innovations requires not just thinking of them, but introducing them into an ongoing system of activity in which people's interests and skills are invested. Social arrangements determine whether such labor-saving innovations can be introduced easily. Consequently, the rate of "learning" of a production line has social determinants.

For another example, network relations between vendors and buyers that establish the advantages of the current vendor in some product markets have more stability if there are many reasons for the vendor to visit the buyer. In industries in which vendors visit for maintenance or to introduce innovations into the buyer's system, as frequently happens in computer systems, networks will be more stable. An innovator's market advantage partly consists of having stabilized networks that connect it to buyers. An innovator's monopoly advantage will thus tend to be longer lasting when the vendors typically provide maintenance or sell innovations on a system.

After this sociologizing of Schumpeter's basic economic observations on the relations between innovation and monopoly, we turn to sociologizing his view of entrepreneurship. The basic purpose of the middle sections of the chapter is to describe what has to be added to a technical idea in order to make it into a continuously producing going concern. To make an innovation viable, there has to be a market, as well as a social structure that can reach that market. Otherwise the innovation will not create a monopoly that will be worth anything. An investment group with control over the necessary resources has to be convinced that the innovation will work, which usually means in particular that it be convinced there is a market the innovation can reach. A manufacturing (or service producing) process has to be created that can produce the prototype of the innovation at a cost only slightly above what people will pay. This production process has to be started down the learning curve so that, to get the item's price, one need not add a profit to the cost at which the first prototype model was manufactured but instead can add a profit to a product the innovator has learned to make more cheaply. The benefits and profits have to be divided up in such a way that the people who as investors make the innovation viable feel they are sufficiently rewarded, that workers will tolerate the introduction of cost-saving improvements in manufacturing, and the like. And personnel adequate to all these purposes have to be added to the technical group from the R&D department to complete the social system for carrying the innovation.

This theory of what it takes to produce a social system that can carry the innovation, then, provides the basic outline for a theory of the social conditions of innovation in organizations. It is because that theory has so many variables, because there are so many things that can fail in the process of developing that social system, that introducing innovations is not in fact routinized even in large organizations.

In the final section of the chapter we turn to the question of administrative organization for innovation, using the theory extracted from Chandler in Chapter 4 above. The basic observation here is that innovations very often have market uncertainty that ramifies into manufacturing and engineering, and so need to have an autonomous "divisional" organization. Early in building the social system that will carry an innovation, many manufacturing adjustments have to be made in the light of market information, many engineering innovations are involved in solving the manufacturing problems of early production, and many customer service problems of the marketers require information from the engineering or manufacturing staff. Consequently, there is much interdependent uncertainty, requiring rapid information flow among the various parts of the innovation-carrying organization. This is the condition that Chandler argues requires an autonomous, fast-acting, product-oriented administration. That is, innovations must be administered with minidivisions.

But in large organizations it is difficult to create a special administration whose marketing is detached from other marketing, or whose manufacturing is not under the supervision of the vice president for manufacturing. The special reward system needed to get employees to work all night to get things going, to bet their careers on an innovation that may not pan out, is hard to justify to a personnel department trying to get uniform fair standards instituted throughout the corporation. In short, divisionalization on a large scale was not easy for the corporations Chandler studied, and it is not easy on a small scale for building up an organization to turn an invention into a going concern. As a consequence, routinization of innovation is not administratively easy, and large corporations very often fail at it.

Innovation, the Learning Curve of Cost Reduction, and Monopoly

Two main processes account for monopoly advantage: advantages in production, such as cost advantages; and advantages in the market, such as having networks in place through which firm products can be marketed. A central part of the advantages in production, as outlined first by

Schumpeter (1942, 81–106), is the temporary advantage a firm has when it has just introduced a successful innovation. An innovation monopoly advantage consists mainly of better production performance than the earliest imitators.

Let us start with the cost advantage of an innovating firm. The curves in Fig. 4 show the decline of costs with increasing numbers of units produced (the horizontal axis is the units produced by the innovator). The general pattern of decline of costs has been estimated for the production of airplanes and is known by the name "the learning curve." The basic shape is given by the fact that, for airplanes, the manufacturing cost of the second plane of a given model produced is roughly 75–80 percent of that of the first, the cost of the fourth plane is roughly 75–80 percent of that of the second, and so on, with the manufacturing cost of a unit being cut by between 20 and 25 percent each time the number of units produced is doubled. So the cost keeps going down with increasing production (or with time), but at a decreasing rate.

We have imagined that this cost curve describes the experience of the innovator and the first follower. That is, they both learn from experience in production at the same rate, so that, for example, the triangles B and C in Fig. 4 are the same shape. At the time the follower enters the market, the innovator has most of the production problems well in hand and is quite far down the learning curve. Obviously, the larger the distance A

Fig. 4.
Cost Reduction Curves with Increasing Production ("Learning Curves") for an Innovator and a Follower

(the number of units the innovator has already produced before the follower gets into the market), the larger the innovator's cost advantage from learning at that time and at all succeeding times.

We might imagine that at point D , the cost advantage of the innovator over the follower is trivial, so that they are by that point truly competitive. The total size of the monopoly advantage of the innovator, then, is roughly the area bounded by the two curves and the double-headed arrow A , up to point D when the two are essentially even competitors. Between the time the innovator starts production and the time the first follower starts, any other follower may be assumed to be able to start where these two firms started, at some point on A . So up to the point of entry of the first follower, the innovator has an advantage for each unit produced, its size being measured by the distance between its own learning curve and the arrow A . When the first follower enters the market, the innovator's advantage starts to be reduced as the competitor learns to produce more cheaply than a new entrant would. The innovator's loss of advantage is progressive, because the innovator's learning goes slower after many units have been produced, allowing the follower to gain on the innovator. So the distance between the innovator curve and the arrow (before the follower enters) or the competitor curve (after follower entry) measures the cost advantage. The area between the two curves and the arrow is therefore the cost advantage multiplied by the number of units (giving total excess profit up to unit D ) or by the time (giving total excess profit up to time D ).

That is, if a follower were to enter the market at the same time the innovator produces its first unit, it could presumably make its own first unit at the same cost, and there would be no monopoly advantage. Both innovator and follower would go down the left-hand learning curve and would have equal costs of production. Since, however, the follower is presumed to enter later, after a period A , the innovator reaps all the advantage of learning. It can still sell in a market where a potential competitor's price would have to cover costs at the top of the curve, even though the innovator has learned to produce the product more cheaply. As the innovator comes down the left-hand curve with no follower, its cost advantage (over a fictional follower) increases from zero on the first unit to a quantity q when the follower starts. As it produces the first unit it has zero monopoly advantage; on the second it has, say, 20 percent, on the fourth 36 percent, over the fictional new entrant, and so on up to the advantage q for the (A + 1) unit.

To get the total learning-curve monopoly advantage created by having no entrant up to time A , then, we need to sum up zero times the cost

of the first unit, 20 percent of that cost for the second unit, 36 percent of the initial cost for the fourth, and so on, up to q for the unit produced at the time of the follower's first unit.

If we imagine a series of rectangles one unit wide, each with a length from the arrow A down to the left-hand learning curve, the advantage got from any one unit will be the area of that rectangle (the cost advantage times a width of one unit). Summing up all the rectangles for the total monopoly advantage due to the follower starting after time A will give the area between the left-hand cost curve and line q and below arrow A (which is at the level of the cost of the fictional early entrant).

Similarly, the (A + 2) unit of the innovator (the first after line q ), produced in competition with the second unit of the follower, has a markedly reduced cost advantage. The follower has cut 20 percent (say) off its costs, while the innovator is producing at a stage when it is harder to find production economies of the same size. This would be a cost advantage equal to a rectangle one unit wide and having a length equal to the distance between the two cost curves (extending just to the right of the line q ). This cost advantage is fairly quickly reduced to the distance between the curves at D . By summing up the cost advantage of the innovator at each point, times the number of units to which that advantage applies, we get the area between the curves—between lines q and D —as the total monopoly advantage after the entry of the first competitor. (Complications of different rates of production at different points on the learning curve are ignored here, for the sake of simplicity. They would not change the conclusions.)

Clearly, when learning is fast, the monopoly advantage will be larger. This is shown in Fig. 5. Here we have added to Fig. 4 two learning curves with a flatter slope, one for the innovator and one for the follower. That is, if one compares triangles B and E , it takes more added units produced in triangle E to give the same amount of reduction of cost as in triangle B . Learning is slower. It might intuitively seem that the faster followers could learn, the quicker they would catch up with the innovator (one might, for example, guess that learning is very rapid in the computer industry, as competitors imitate each others' innovations quickly). But if we assume faster learning for the innovator as well as for the follower, the head start of the innovator is larger because it has been going down the learning curve faster. Another way to see this is to compare the total cost savings for the innovator up to the point F from faster learning (the area with vertical hatching) with the total cost savings for the follower to that same point (the area with horizontal hatching). If the learning curve is shallower, the total advantage of being farther along is not as great.

Obviously in either case, the longer the "lead time" between the inno-

Fig. 5.
Graph Showing That the Monopoly Advantage from Innovation Is Lower When the
Learning Rate Is Lower

vator's introduction of the good into production and the beginning of production by the first innovator, the larger the cost advantage of the innovator at any given time and so the larger the total size of innovator advantage. Many of the proposed antitrust remedies against IBM's dominance of the mainframe market have involved demands that IBM give the technical details of an innovation in central computers to possible competitors as soon as these were released to its own peripherals and software divisions, so that at least in manufacturing peripherals and developing software the followers would have an equal start with IBM's own divisions. That is, the proposal was to reduce the distance A in Fig. 4 so as to reduce the monopoly advantage of IBM's software and peripherals divisions (Fishman 1981, 233–245).

Cases in Which the Follower and Innovator Have Learning Curves of Different Shape

So far we have noted that a pair of learning curves can be flatter or steeper, or closer together or farther apart on the x-axis (the axis indicating number of units produced or time spent in production). Both affect

the size of the monopoly advantage, with steeper curves farther apart being advantageous to the innovator. A third difference in the relationship between the curves is that the imitator may learn faster (e.g., if imitation is not as hard as innovation, or if the imitator can hire experienced labor from the innovator, leading to a steeper learning curve). Even if the imitator approaches the learning curve of the innovator as an asymptote, never producing more cheaply than the innovator, faster learning for the follower reduces the distance between the curves and so reduces the total monopoly advantage of the innovator. Ordinarily, faster learning of the follower will eventually bring the follower to a lower overall cost of production than the innovator.

In Fig. 6, the right-hand curve is steeper than the left-hand one, so the follower curve approaches the innovator curve faster than in Figs. 4 and 5. This is shown by triangle G , with its shorter base, indicating that for the same amount of cut in costs (the vertical of the triangle), the follower needs to produce a smaller number of units (or a shorter amount of time has to pass). Because the follower curve approaches the innovator's curve faster, the point at which we judge the innovator advantage to be trivial comes at point H rather than, as before, at point D . Because the follower reaps cost advantages in less time, and because the follower more quickly reaches a point at which the innovator's advantage is trivial, the total area representing the excess profits of monopoly is smaller.

For example, IBM apparently learned something that was essential for success faster than Remington Rand (Univac) in the computer market, since Univac started with the innovation and IBM was a follower (Fishman 1981, 29–47). Relatively soon after IBM entered the market, however, it could compete at least equally with Univac. That is, point H occurred earlier than point D (the point of follower competitiveness with identical learning curves for follower and innovator) because of IBM's more rapid learning than Univac's.

Obviously, if some component of costs is lower for the follower—for example, a follower in Hong Kong may have cheaper labor, making its total curve lower from the beginning—the advantage of the innovator will be smaller after the competitor starts because the right-hand curve is in general shifted downward. Unless the firm with higher labor costs learns faster (it may be that innovation itself is a measure of the capacity "to learn," in the sense measured by the curves), the Hong Kong producer will eventually have lower costs than the innovator and should be able to drive the innovator from the market.

In Fig. 7, the right-hand curve starts lower than the innovator's curve did, because of the lower cost of some factor of production, though at

Fig. 6.
The "IBM Effect": A Steeper Learning Curve by the First Follower Reduces the Monopoly
Advantage of the Innovator

Fig. 7.
The "Hong Kong Effect": A Follower with Cheaper Labor Reduces the Innovator's Monopoly
Advantage, Eventually Driving the Innovator Out of the Market

first it is still above the innovator's curve because production in the cheaper location is not yet routinized. As experience moves the low-cost follower along the learning curve, it becomes essentially competitive with the innovator at point I , rather than at point D as before. Further, with increasing experience the follower in the low-cost location has lower costs than the innovator, eventually perhaps driving the innovator from the market it created.

So the size of the monopoly advantage created by the learning curve increases with: (1) longer delay between the first production of the innovation and the start of production by the first competitor (longer A in Fig. 4) and (2) steeper decline of costs with experience in production (steeper rather than flatter learning curves in Fig. 5). The size of the monopoly advantage decreases with (3) the "IBM effect," faster learning by the first follower than by the innovator (steeper right hand curve in Fig. 6) and (4) the "Hong Kong effect," cheaper labor or other factors of production by the follower (as in Fig. 7).

Innovation, the Marketing Network, and Monopoly

The market advantage of the "first mover" in general consists in having a more solid relation to the client than latecomers. One main way to be a first mover is, of course, by innovating, but there are others (e.g., getting the first contract with a big client). What we want to do now is specify what determines whether a contact in the marketplace is solid (and consequently creates a monopoly advantage for a first mover) or weak (is easily entered into by competitors). The network advantages that tend to maintain an innovator's monopoly position take three main forms: (1) the vendor in place may develop a multiplicity of contacts with clients and serve multiple functions for them, with each such contact reinforcing the others; (2) clients may become technically or economically dependent on the vendor in place, meaning they would pay heavy costs if they changed vendors; and (3) clients may develop such trust in the vendor that information about that company's product is believed while that about a competitor's product is received with suspicion.

The multiplicity of contacts between vendors and clients is increased when the client depends on the vendor for maintenance, for a flow of technical innovations adapted to the installed system, or when the client is involved in continuous buying (including when the equipment is leased from the vendor rather than bought). The responsibility of the vendor for maintenance builds the vendor into the technical system of produc-

tion, so that day-to-day operations of the client require continuous interaction with the vendor. This provides many opportunities for the vendor to find out what the client needs and to give advice about how the vendor can satisfy that need, to exchange information informally about the economic relation and about sources of client dissatisfaction, and otherwise to support the capacity of the vendor to sell the next installation.

The "maintenance" relation established in the computer industry involves not only maintenance, but also access to a continuing supply of technical innovations. When one leases software from a vendor, one ordinarily gets the rights (without cost) to improvements in the system. Similarly, in the days when clients typically leased the machines, technical improvements in the hardware were often installed free of charge. But whether they are free of charge or not, innovations developed for software or hardware around which the client's operations are already built are more useful to the client than are random innovations. IBM's advantage over other mainframe computer manufacturers in selling peripherals was that any peripheral adapted to an IBM machine had a larger market than peripherals adapted to any other machine; their advantage over the plug-compatible manufacturers was that they could earlier and better adapt their improvements to their own systems, as well as make the systems harder for peripherals manufacturers to plug into.

When the client is buying a continuing stream of goods (as when an assembler buys from parts manufacturers) or a continuing stream of services (as when the client leases a building or a computer), continuity of supply is an important consideration. If the present supplier provides continuity, while changing suppliers would create an interruption, then the present supplier has an advantage. Further, the continuous flow of goods tends to be accompanied by a continuous flow of information, giving the vendor an information advantage in future sales. The vendor can in extreme cases make what Oliver Williamson calls "transaction specific investments," which are especially adapted to efficient and satisfactory delivery of what the client wants. A common case is remodeling by the landlord in a leased building, but transaction-specific investments may involve innovations made especially for a client, such as modifications on software for a specific client application.

The technical and economic dependence of clients on the vendor is increased when clients' operations are adapted to specific features of the product, such that they would have to reorganize many of their routines in order to go over to a competitor. For example, as long as the only people who use the computer are in the computer center, the costs of going over to a competitor system tend to be small. But when many

people in the accounting department use a particular software program for invoicing, accounts payable, and so on, it can be very expensive to go over to a slightly superior or slightly cheaper competitive software system. Since software itself often depends on the hardware, it is often the case that building software into an organization's routines makes the organization technically dependent on the hardware as well.

One of the most common forms of technical and economic dependence is moving costs in real estate relations. Because it would cost us a good deal in time and effort, and often a good deal of money as well, we do not easily choose to move to gain small advantages in the real estate market. The costs of moving are even more substantial for a business, for it must move a social system with recalcitrant parts. (Moving a clientele is often extremely difficult; cf., on moving a labor force, Mann 1973.) There are often similar costs in changing equipment. For example, IBM for a number of years simultaneously rented two functionally equivalent memory systems, the "Mallard" and an older one, with the latter costing more per unit of performance. Although clients renting the older, more expensive system were free to change and save money, many kept it simply because finding out about the new one or moving the old one out would have been too much of a disruption (Fishman 1981, 235). If such moving costs are required when switching vendors, the first vendor in the market has a monopoly advantage in the amount of the moving costs, whether it got its first mover advantages from innovation or not.

Economic dependence can take the form of special credit arrangements with the current vendor. Because the current vendor may enjoy an information advantage, it may be willing to give credit when other vendors would not. Or the vendor may have got inadvertently into an exposed credit position with a client which another vendor would not rationally get into. Nevertheless, the client now depends on the vendor.

An obvious source of stability in vendor-client relations is mutual buying. IBM, of course, buys a great many telephone services, and arranges for its clients to buy many more. Bell was IBM's largest customer for computers when it was a single system, though this relation has been somewhat undermined by Bell's breakup. If either company is being difficult in its buying behavior (e.g., threatening to go over to competitors), it may face the sanction of the other being difficult in its .

In continuing client-vendor relations, trust grows out of a history of trustworthy behavior. In many cases the trustworthy behavior is making the system in which the product functions run, which may mean getting the bugs out of that system (e.g., out of the interfaces between the vendor's machine and the rest of the production line). Clearly, the vendor already on the scene has an advantage in having got the bugs out and is

more likely to be able to deliver adequate system performance than another vendor. The long-run delivery of such adequate bug-free performance (or minor-bugs-rapidly-fixed performance) creates an atmosphere of trust.

In addition, the vendor's representatives can deliver information useful to the client on the vendor's own probable future behavior (e.g., products coming out, early warning of delays in release of those products), on its financial situation, on who to call to get action in the maintenance department, and the like, which may not improve the vendor's behavior but at least renders it more predictable by the client. Insofar as trust is based on predictability, it will tend to grow in an information-rich relationship with the current vendor rather than in a new relationship.

Further, because the vendor on the scene is in a continuing relationship itself, its own behavior will in general be more trustworthy. One is less likely to imagine that a continuing relationship can be maintained by lying to one's clients than one is to imagine that one can sell for the first time by misrepresentation. The client may be able to use continuing small sanctions on the vendor to improve its performance, and so to make the vendor in fact more trustworthy (Rogers and Larson 1984).

The sum of these advantages in creating trust results, for example, in managers of a company or university computer center using as a clinching argument a phrase like "I feel comfortable with IBM."

It is important to remember that one can destroy one's own advantages. Apparently Univac sent out engineers to maintain its machines—engineers perhaps too honest about Univac's problems. IBM sent maintenance people supervised by marketing specialists, who, whatever their other disadvantages as maintenance supervisors, were not excessively honest about technical details they did not know (Fishman 1981, 38, 44).

Thus one would predict that the more the vendor-client relationships in a market involve a multiplicity of contacts, technical or economic dependence of the client on the vendor, and developed trust between the client and the vendor, the greater the innovator's monopoly advantage in that market would be. The argument to this point has been that the monopoly advantage of an innovator is also a social system advantage. To gain and preserve a monopoly advantage by being farther down the learning curve, an innovator must first build, then constantly improve, a production social system. To gain a monopoly advantage in marketing, an innovator must build and solidify ties with clients as the first vendor in the market. The next section specifies what kinds of information-processing and decision-making structures it takes to create social systems that can exploit an invention by building and maintaining such monopoly advantages.

The Theory or Doctrine of an Innovation

In this section we hope to describe what sort of a social system an "innovation" is. Schumpeter (1942) is very clear that an "innovation" with a potential economic impact is quite different from an "invention" that might be patented. In the first place, a number of things that cannot be patented are innovations, such as the marketing in the big Eastern cities of lettuce from irrigated desert farms by packing it in ice and getting through service on the trains. In the second place, an innovation is what John R. Commons (1924, 143–213) has analyzed as a "going concern"; that is, it is a social system that can in fact regularly produce and market the innovation.

An innovation, then, is the unit of analysis, and we are talking about things like introducing word processors into an office, bringing out a new computer (Kidder 1981), designing and building a wide-body passenger airplane (Newhouse 1982), developing hybrid seed-corn (Grilliches 1957), and introducing multidivisional administrative structures (Chandler 1962). I have written elsewhere about some of the variables describing administrative innovations, whose essence is social, which determine how much opposition they are likely to meet (Stinchcombe 1986b), such as whether they involve the redistribution of status symbols. These variables described innovations by variations in the impact those innovations had on the interests embedded in the old regime, and consequently predicted whether or not they would be adopted at all. But here we want to describe what kind of social system it takes to make innovations that have already got past this adoption barrier into going concerns.

Obviously, the social systems created to administer an innovation differ a lot in size and complexity. When IBM embeds a number of improvements in a new board to slip into an old machine, they are very careful to design that board so that it interfaces with the rest of the machine and with the software having the same structure, so that the maintenance workers can just haul out the old one and put in the new one, and the users do not have to change a single line of code in their programs, appoint or fire anyone, change any human routines, or learn anything new. The innovation requires some modification of IBM's system, but nothing from the user. This is an innovation where the client's adaptation involves only lower costs and fewer mistakes, and it has already been paid for (in the old days, at least) in the maintenance contract that went with the lease. For the replacement board, the value of all the variables we describe below—what one has to do to make the social system attached to the innovation go—is very near to zero for the user; very little

social system with this innovation needs to be made to go. So what we will try to describe with the variables below is a series of ways that innovations can depart socially from this ideal-type of an innovation with zero social system requirement, the replacement board.

I will organize the description of the social requirements of an innovation around six elements. Since a proposal to institute an innovation (or to implement a decision to introduce an innovation) must have a theory of the social requirements of the going concern, or be forced to develop one as it goes along, I will call each of these elements a theory of the social requirements in question. I have previously used many of these ideas to describe the theory one has to have to create an organization (Stinchcombe 1973). Here I use the same approach to describe the elements that must go into a theory of an innovation for that theory to be sufficient to build a going concern that can create and maintain a monopoly advantage. The complete theory of an innovation, then, requires the following elements:

1. A core theory of the innovation, of what is centrally involved technically in the design of the innovation, and a corresponding theory group (sometimes a "group" of one) that develops and defends this core theory

2. The theory of the investment in the innovation (and the risks and profits theory that justifies it), and the investment group responsible for collecting, safeguarding, and spending the money or other resources

3. The technical-costs part of the theory of the innovation, which is about what it will take to produce the innovation and what that will cost, and a corresponding engineering or other technical-costs controlling group that designs the technical productive system

4. The market or benefits part of the theory, about who wants the good or the effects produced by the technical-costs part, what price will they pay for the benefits, how the organization can reach those who want the benefits, and the concrete sales apparatus or clientele networks that the organization needs to build up for these

5. The division-of-benefits part of the theory, about what promises of future profits or interest payments can motivate the investor group, and about how benefits are to be distributed so as to provide careers for motivation of personnel

6. The personnel part of the theory, and in particular

answers to the problems of competence of personnel, trustworthiness of personnel, motivation of personnel, and the arrangements for the corresponding personnel flows.

A "Zero Resources Innovation" Described in Detail

Let us describe briefly how these elements enter into our "zero innovation" of the improved slip-in board, but now looking at it from the point of view of IBM, for which this is a "small" rather than a zero innovation. (The details of the following picture come mostly from Fishman 1981, 51–151, 379–411.)

1. The core theory involved in a replacement board is built up out of a series of maintenance problems brought in by service personnel, special fixes in the operating system to get around a limitation in the old board collected from folks who consult on software (some even from IBM's plug-compatible competitors), and perhaps some engineering innovations that came after the original design. There will be a small group of engineers and others who collect these, look them over, design modifications in the board, test them to make sure the "interfaces" are the same and that new bugs are not built in. That is, there is a flow of detailed information about problems with the old board flowing to a group dedicated to improvements of the design of the hardware, a "theory improvement" group.

2. The "investor group" is of course ultimately IBM stockholders, who allow IBM to retain earnings to invest at IBM's internal rate of return. More immediately, the group within IBM responsible for the system will have allocated such "upgrading investment" funds as part of their normal budget, probably financed out of returns from the particular machine being upgraded. IBM does this because they know that one main thing users buy is the reliable operation of the system, which is why IBM sells maintenance contracts. (When technical change is as fast as in the computer industry clients cannot provide reliable maintenance for themselves; see Stinchcombe and Heimer 1988.) A system executive committee of some sort will invest in the upgrading.

3. The technical-costs part of the theory is an engineering cum software theory about how the various glitches can be alleviated by rewiring the board.

4. The market theory of IBM has changed over time, especially in response to antitrust suits. The basic IBM market philosophy is that a cus-

tomer buys the service of a computer system, not the hardware. For a long time they would not even sell the client the hardware. Originally the client bought (i.e., rented) the system package, which included the maintenance agreement (which included the upgrading). Part of the system package was general access to all sorts of technical innovations—that is, the client bought a flow of improvements as well as the latest system. Much of IBM's advantage in the market consisted of the faith (assiduously maintained by the marketing person in charge of the account) that an IBM user would not be trapped with a big investment in an obsolete machine.

A second part of the market theory is that if IBM makes clients rewrite their software and reorganize their data-processing department when IBM introduces an innovation, that is the time they will go over to Control Data or DEC or some other company. So the interface stability criterion for the new board is part of the IBM market theory.

5. The central part of IBM's division-of-benefits theory is that 20–30 percent is a reasonable expected return on an investment. A second is that the company should not lay anybody off—the worker has a job at IBM regardless of business conditions. Market uncertainty is to a large degree put off onto the suppliers; when demand goes down, IBM is likely to start manufacturing parts that it once bought from suppliers, in order to keep its own workers busy (on this strategy in general, see Sabel 1982, 47–56). Related to the security of employment, however, is the practice of demotion. While employees are more or less guaranteed a job, they are not guaranteed the job they have now or one at the same rank (Goldner 1965).

A third part is market criteria dominance in IBM's reward system. At IBM, marketing dominates over engineering much more than at Data General (Kidder 1981). People get rewarded in IBM if their product does well in the market, not if it is only technically superior. The division of benefits is then made up of administration and profit criteria and career-structure rules that are traditional at IBM.

6. The personnel part of the theory at IBM is embedded in the company's general personnel system, by which people's careers organizationally depend on satisfying the users of the systems they sell. So the engineers in IBM will be part of a group dominated by the contact people who service the big users. The contact people, not the engineers, will be close to the managers for the system involved. In IBM the engineers are closely controlled so they give the users what they want, not merely a state-of-the-art product. These engineers, even more than other IBM engineers, are dominated by market people, whose big aim is

to hold customers. Bell, for example, in going competitive has imported a lot of IBM marketing people, modeled their new marketing departments on the IBM marketing structure (IBM has systems people responsive to marketing people, who are divided according to type of customer and type of system). Obviously, the personnel also have to understand electrical engineering, some computation theory, and so on.

Overall, then, this innovation process makes use of parts of IBM's general structure, which is set up to introduce client-serving innovations on a regular basis.

The Multidivisional Structure of Chandler As an Innovation

Let us now go through these elements with Chandler's innovation (1962), the multidivisional structure.

1. We note Chandler's comment about the innovations coming faster if people took time off from operational tasks to study organization, if they had a general tradition of organizational thinking, and if an engineering-rationalizing tradition dominated the organization. These are characteristics of a "theory group" that is likely to develop the sort of theory involved in an organizational innovation. But we also note that the problems could be misdiagnosed if a company had experts from outside without access to the details of the problems.

In general, though, adaptive response—or "muddling through"—was a main way of developing a theory of the new situation. For example, if Du Pont executives theorized too much about organization in the abstract, they tended to come up with the traditional theoretically optimal solution according to the organizational theory of the time, namely centralized, departmentally specialized administration. Chandler argues that a theory group with a decentralizing innovation was less likely to develop among older executives and more likely to form among younger executives (which means people about forty to fifty years old, apparently), and it was less likely to form when the organizational arrangements were strongly determined by family ties (though Pierre du Pont and Henry Ford II were leading advocates of the decentralized system).

2. "Investment" for administrative reform has to be the time and attention of authoritative people. Only a few managers are inclined to be reflective, to develop a theory. Sloan was very unusual in this respect. Top managers in general like to live in a bustle of activity, to be on the phone, to pay no more than five minutes' attention to a given subject (Mintzberg 1973). So it is not an easy thing to develop an "investment group" when the time of top administrators is the investment required.

3. The technical-costs theory has to break down the idea that minimizing administrative costs is the most efficient approach and also that part-time generalists (the "cabinet ministers" system) are good enough for strategic management. In Chandler's analysis, generalists who also had operating responsibility, especially responsibility for a functional department that was likely to develop goals of its own and a distinctive worldview, were not sufficiently attentive to general problems. Generalists with specialized responsibilities were less likely to depend only on market forecasts and an analysis of the distinctive competence of the organization and less likely to invest without fear or favor in different company activities; instead they tended to favor investments in more of their own specialty. Operative executives, too, were generally likely to invest more money in trying to solve their problems, rather than deciding that their area of operations was so full of problems that it ought to be scuttled.

Further on in the technical theory of the innovation the innovators had to decide what sort of abstraction techniques could be built into the information system so that the general office would not find the general contours of the strategic problems buried in a mass of operative detail. A crucial intellectual innovation in all this was to see that committees of generalists are good for making some kinds of ("strategic") decisions, while individuals (with advice from committees subordinate to them) are better for making operational ("routine") decisions. Different kinds of decisions required different administrative structures.

The technical-costs theory described by Chandler was embedded in many details about how particular decisions are to be made, what information about the basis and quality of decisions goes on to the general office, what controls the general office puts on those decisions, and so on. In short, as a theory, administrative decentralization was pretty helter-skelter, which is why Chandler had such a hard time explaining exactly what it was.

4. In Chandler's description, the benefits of the innovation were supposed to go ultimately to the stockholders. What the stockholders supposedly wanted was managerial control of investment in the activities of the corporation's subsidiaries by the criterion of long-run profits. Ultimately the innovators had to sell the active part of the board of directors—that is, the active top managers—that the benefits would show up in long-run profitability. The market for the innovation was, thus, internal to the company.

5. The benefits of innovation mostly went into the pot of general corporation profits. There may have been career benefits, such as becoming general officers or having full charge of a division, but Chandler does not

discuss this. Sloan has endowed many programs in business schools, so he must have got something out of his work of introducing the innovation at General Motors.

6. The main personnel part of the theory of administrative decentralization had to do with the full-time dedication of central office people to central office problems. The comparable thing in government is the relative role of the White House staff, especially the Office of Management and Budget (playing the role of the general office) versus the cabinet. If we imagine putting the White House staff on top of the cabinet, we have the personnel arrangement that Chandler saw as central to the innovation he was describing: a full-time top staff with general and abstract responsibilities supervising a group of heads of operative units. In general, the hierarchical superiority of the central office tends to lead to "problems of subordinates" gobbling up the superiors' time (see Stinchcombe 1974, 81–90). This had to be avoided by limiting communication between the general office and the divisions to general matters.

To put it another way, the innovation theory has to figure out how to make a hierarchical superior in the general office into a "staff" advisor on the implications of strategic policies on operative decisions, rather than a "line" executive. It involves having technical executives in the general office make reports to a committee, rather than making decisions. The crucial personnel question, then, was to create "general officers." The companies Chandler analyzed already had experience producing operative executives, so that was not a new personnel question.

Social Predictors of Success in Introducing Innovations

The basic point of the preceding sections is that if we want to describe an innovation in social terms, so as to bring sociological insights to bear on what happens to it, we can use this simple six-point outline (theory, investment, technical costs, marketing, division of benefits, and personnel) for descriptive purposes. We will next specify some variables that make it difficult to introduce an innovation. Obviously this bears directly on the question of monopoly, because the longer a successful innovator has the innovation while its competitors scramble to build the social system needed to carry it, the larger the monopoly advantage of the innovator will be. As preparation for this task, let us briefly reconceive the requirements of making an invention into a going concern as a set of information and decision requirements.

The first requirement is that the information system be set up so as to

receive the information that needs to be processed in all six areas. The most common way this fails is to have engineers or inventors build a system that is equipped only to make technical decisions. That is, information leading to sensible investment, marketing, production management, division of benefits, and personnel decisions is not incorporated into the plan of the innovation, so these decisions get made by default. We will call the failure to satisfy this requirement "technological utopianism" below.

The social origin of this deformation of an information and decision system for producing an innovation is that one needs the invention first. The social requirements for producing an invention do not in general lead to a theory adequate to produce an innovation, because the theory built by inventors does not create a structure that allows social information to get through. If the inventing group were good at processing social information and making social decisions, they would likely be managers instead of inventors.

That is, the variable we will be describing as "technological utopianism" describes a specific common type of inadequacy of the information and decision system: confinement to technical matters and to administrative structures useful for getting the right technical answer. This deformation is particularly likely to happen in a part of an organization dedicated to introducing an innovation, precisely because innovations tend to start with inventions. It is a deformation because a theory adequate to describe an invention will not produce an information and decision system adequate to introduce an innovation.

In the following sections we will develop a theory of the information and decision difficulties that inventions get into, by discussing the information requirements of each part of the necessary social structure and the information problems they are likely to create by ignoring the requirements of the rest of the going concern being built.

Technological Utopianism

The basic argument of this section is that a common feature of theories of innovations is that they are seriously incomplete about what it will take to make the innovation a going concern. Let me illustrate this by giving a somewhat idiosyncratic reading of Engels's Socialism, Utopian and Scientific ([1882] 1935). The main thesis of that work, as I see it, is that the utopian socialists want to give a theory of how society might be organized better; Saint-Simon, for example, had scientists running everything. The utopians think of the process of introducing socialism as basi-

cally one of teaching people the advantages of socialism, whereupon, being convinced, they will institute it. But Engels says a "scientific" socialism has to specify the social forces that will bring this ideal state into being: specifically, the proletariat, subject to oppression in a capitalist market, and class conflict as a way to organize this group into a political force. That is, Engels's argument involves a theory of what the social system has to be to make it a likely carrier of an innovation like socialism. And that theory does not look like schoolteachers giving lectures from socialist texts.

The big first variable about theories involved in innovations, then, is "technological utopianism," by which we mean the degree to which the theory is merely describing the technical way to achieve some end, with no analysis of the social and economic forces needed to bring the innovation into being. A very good example is Apple's scheme a few years back to give one computer to each of one hundred thousand schools. It is not quite, but almost, fair to say that the scheme said nothing about how six hundred children in a school were going to practice using one computer, how teachers were going to plan their science or mathematics or writing curricula around it, how the teachers themselves would learn how to do it, who would pay the costs of their training, which children would be able to do their homework on a computer at home, what it means to have an academic training program that cannot have homework, and so on. There was, in short, no analysis of the investment and investor group, the benefits and who gets them, the training and motivation of the personnel, how the social system relates to who gets promoted in the schools, and so on. It was a very utopian theory, worthy of Saint-Simon or Fourier. Of course, if it was not intended to be a theory of an innovation, but merely a way of distributing samples to schools for advertising purposes, then the technological utopianism of the theory behind it was perhaps not a business disadvantage—merely a disadvantage for the adopting school.

A good many technologically utopian theories like this can be found in education, ones not based at all in the realities of thirty children in a room six hours a day for 180 days with one teacher (or five teachers one at a time) and no money for extras. People proposing innovations in education rarely say how many minutes their innovation should take, whether the time should come from the time now spent on arithmetic, reading, social science, or whatever. They are often utopian in the even narrower sense of not even having a technology for how to do it and when. When social scientists interview teachers about why they did not follow up on some jazzy program, the most common response is that after they got

their enthusiasm up, nobody told them how to institute it concretely. So the educational theories tend to go skating over the surface of the schools, earning education professors lecture fees (small ones) and ending up being entertainment for teachers who get an hour off to go hear them.

The social significance of utopian theorizing is, first of all, that the computer sits in a closet or the reform movement disappears without a trace. The first consequence, then, is failure of the innovation to be truly introduced. It remains a technological utopia, without social flesh.

A second consequence is that if the technical solution is a real advantage but is nevertheless embedded in a technologically utopian scheme for bringing it to reality, there tends to be a long period of thrashing about until a viable social vehicle for it is found (a good description of such thrashing about, which has informed the discussion here, is Smelser 1959, 50–179; see also Hirschman 1963, 12–91). In such situations the innovation in a particular organization often depends on some enthusiast working eighty hours a week to put together the pieces to make the innovation go. For all innovations there is a long period of "social debugging" of the scheme for the innovation, with someone needed to carry the scheme through when unexpected difficulties (mostly market, personnel, division-of-benefits, and investment difficulties) arise and when the benefits of the technical innovation are delayed.

But there is more muddle in the muddling through for innovations introduced with utopian theories. If the plan has not specified the social forces needed to bring the going concern into being or described the market and who benefits from the innovation, how to divide the benefits that return to the organization, or who invests how much and why they do it, or the training and motivation of personnel, or the measures of success to be used, then the organization has only a "neural itch" of an innovation, not a theory.

One generalization in this field comes from Karl Mannheim, in a great book, Ideology and Utopia (1929, 117–122, 211–219, 229–239). The basic argument of that book is that theories of activities already embedded in ongoing practices tend to be disintegrated theories about how to solve particular problems, emphasizing the wisdom of institutions and the value of experience, with no overall scheme of how the whole thing works. Theories associated with movements for innovation ("utopias"), in contrast, tend to cover great areas of social life with one simple theory, one master mechanism to drive it all, involving a cataclysmic crisis or revolution that will quickly create a new society that runs on different principles, with no hard work of building that society up piece by piece involved.

To put it another way, the conservatives (he calls them "ideologists") typically believe that no reform can ever work because it would upset all the specific solutions we have worked out to deal with the various problems of coping with human nature, while the utopians or radical reformers typically believe there is no work to do after the revolution, that things will all fall into place. If one takes the advice of checking out the theory of the innovation implicit in the above analysis, and asks the reformer about who benefits, how much it all costs and who pays, who is going to run it and how we can arrange things so that we can trust them, one is likely to be ranged on the side of the conservatives, the "ideologists." Obviously Engels did not think he was a conservative, but the anarchists and other types of utopians did.

So our first big social variable about a prospective innovation is "utopian versus scientific," the degree to which the social structure necessary to bring the innovation into being is analyzed in the theory.

Investment Approval

In general, an "investment group" is a set of people who have to believe in the innovation enough to decide to put money and time into it. The first variable describing investment is the relation between authority to invest in an innovation and authority to keep running the routines. The core variable is how much harder it is to get authority to innovate than to get authority to keep going. To be sure, the ultimate source of money may not be the group that controls its investment, for instance the stockholders, who never hear of the innovation until it has either succeeded or failed. But the yearly report of the corporation and whatever appears in Business Week or the Wall Street Journal have to be convincing to the stockholders, even though the power to invest has been delegated to a set of managers.

An analysis more detailed than the one that may appear in the annual report has to be approved by a management committee high in the hierarchy, a strategic management group. Sometimes the plan for the innovation must go outside, as when the airline that is going to buy the new airplane has to approve the manufacturer's plans and to promise to buy a certain number before the manufacturer can borrow money to make them (Newhouse 1982).

Often even quite small innovations involving "capital expenditures" have to go much higher in the organization than does approval of how to spend already budgeted resources. For example, a manager who controls ten thousand dollars worth of labor per day (in 1989 that might be roughly

one hundred factory workers) may have to go to a higher-management committee for all capital expenditures over, say, five thousand dollars. The higher management have already said they believe in the theory that explains how the ten thousand dollars per day is spent, but not the theory for new capital above five thousand. In the civil service or a university, the control is often much more minute, so a person with subordinates worth ten thousand dollars per day (somewhat fewer than one hundred because civil servants and professors are more expensive than factory workers) may have to ask for approval by line item of all capital expenditures whatever.

In a certain sense, then, investment groups are ideological groups. The reason nonprofits often have tighter investment controls is that almost everything they do is ideological, so nearly all expenditures are "investments." One has to believe in a university before one will invest money in it, even as a client, because one is not buying a well-specified good. New research projects (presumably all innovations) whose total budget will amount to less than a year's salary plus overhead of a leading professor may be reviewed by several levels within the university before going to be reviewed by a body of peers at the National Science Foundation (NSF), while decisions to renew the contract for that leading professor for another year are not ordinarily reviewed at all, or only reviewed if he or she gets an offer from another university. But when businesses are making investments, they conduct themselves much more like nonprofit organizations and universities, with endless committees and discussions among high-level executives about small expenditures.

The general point is that a clear functional distinction is made even at the lowest hierarchical levels between "operational" and "strategic" decisions, with the strategic decisions getting kicked upstairs, and innovations are almost always considered strategic if they involve any substantial capital investment at all. The approval of investments in innovations usually takes place in a committee-type decision-making structure. People ask whether they believe the theory involved in the innovation enough to put some freely disposable resources into it, resources not already committed to running the operations on which there is already consensus. The formation of beliefs in a theory quite generally goes on more in discussion groups ("congregations") than in an individual mind as a person reads a memo on the innovation.

A second variable in the investment process is the relation between the theory group and the investment group. Ordinarily, an investment group will require more in a theory than a technological utopia. Suppose we examine the difference between an NSF proposal (addressed to an

"investment group") and a seminar paper delivered early on in a line of scientific work. The seminar paper is supposed to be a technical utopia, outlining the possibilities of a line of work in a way not much constrained by practical realities of money and time. The NSF proposal, in contrast, will have (1) a specific description of activities proposed (though the concreteness will be to some extent specious, and NSF specifically provides that within reasonable limits researchers do not have to pursue a line of activities that turns out not to be productive, but can invest the resources in another that appears more promising); (2) a budget; (3) a set of vitas of the people in charge; and (4) a statement of how this research will advance the science. The proposal is in a sense a routine extension of the technological utopia presented in the seminar paper into a theory of the social system that will accomplish this utopia, what its benefits will be, and so on. The additions are sometimes called "boilerplate" in the survey research business, are quite often already in standard files in the word processor, and are very often written or calculated primarily by administrative rather than technical personnel. In general there will be a conflict in points of view between a technical theory group and the business group that makes the investment decision, a technical-business conflict (Veblen 1919 makes this conflict central to modern capitalism) about adding to the technical theory enough social (especially financial and market) information so that the investors can judge the viability of the innovation as a social system. The technical group will often resist doing so much "irrelevant" work on the details of administration and finance.

A third general variable in the investment process is that investment groups have to trust each other with large quantities of money or other resources. There is a strong tendency for people to trust only those others whose behavior they understand well—people of the same culture, often the same people they see socially (Kanter 1977, 47–63). If one knows seven of a person's friends, one knows whether he or she plays the horses too much or likes women or men who are too expensive for his or her personal income.

One consequence is that people of different cultural backgrounds (for example, blacks, women, or foreigners) may have a harder time breaking into an investment group than into a technical theory group (see Stinchcombe 1986b, 255–264). There may be women executives at the middle levels (e.g., department heads and buyers in a department store) long before there are women at the top investment levels. There may be women selling houses long before there are women selling industrial real estate, because the industrial real estate broker is a central participant in an investment group. There may be minority groups in a country (for ex-

ample, Chinese in Southeast Asia) who control most of the country's business because they trust each other enough to carry out the investments but do not trust members of the majority (for example, Malays). Social distance between the innovator group and the investor group reduces the likelihood that the innovation will be carried out, because it undermines the process of convincing the investment group of the theory of the innovation.

A big variable among investment groups is whether they are already a part of an organization (in the case of the IBM board, for example, the investment group is an integral part of the IBM organizational structure) or whether they have to be assembled from scratch to start a new company. The crucial people in the first case are decision-making executives, usually acting in a committee. The crucial people in the second case are brokers, who go around from one to another investor or approval body with a partially worked out investment coalition and persuade one after another to come in until the coalition is complete (cf. Banfield 1961). If a structure with money available for investment already approves of an innovation, it is ordinarily easier to bring the innovation into being. On the other hand, if the distance one has to go in the internal structure of the organization before reaching those with authority to invest is too long, then assembly of new enterprises, with short distances between investors and innovators, may work better.

Cost Reduction and Manufacturing Improvements

Let us now turn to the creation of a technical system to carry out the job of introducing an innovation—the process of coming down the learning curve. As we have already discussed, a crucial thing that creates a monopoly position from an innovation is being ahead of other firms on the learning curve. "Learning" here is the process of developing routines; of using cost statistics and other knowledge derived from experience to improve the efficiency of a productive system; of setting up incentives and social mechanisms so that workers who figure out a faster way of doing the work will let the company know about it (quality control circles as discussed by Cole and Siegel 1979, 134–155, 160–168, 202–211, 222–223; the idea that worker cooperation in technical improvements is more crucial for workers in pilot plants is suggested in Whyte 1961, 198–217); of making opportunistic improvements in capital equipment when it is down for repair (see Stinchcombe 1974, 3–41, for a series of managerial strategies used in a steel rolling mill to reduce down time). That is, learning as described in the learning curve is a set of innovations associated

with the initial product innovation that improve the efficiency with which the innovation itself can be produced.

This translation of theories into technical action may show the technical-costs theories on which the innovation is based to be mistaken. The engineering theories that allow people to estimate the net benefits of an innovation by estimating its costs are very unreliable, especially when the innovation itself is more fundamental. The theories may, for example, be false or radically incomplete only in a specific environment, as I illustrated earlier in discussing Granick's book on the Soviet Union (Granick 1967, 161–164): a mass production theory of machine assembly lines depends on being in an environment that can standardize parts. If a company's parts suppliers do not have adequate quality control, the company cannot run an assembly plant, but instead needs assembly craftsmen or craftswomen with files and calipers to make the pieces of the machine fit together.

The bugs in engineering theories are often in the details, as when a machine tool that is supposed to work to the nearest thousandth in fact works to the nearest two thousandths, or when the clay that looks solid turns to jelly if it gets too wet (and so causes an earth dam to come sliding down the valley). Generally, though, the theory is likely not to work quite right at first not because crucial details were wrong, but because they were simply left out.

A big first variable that causes variation in turning technical theories into production systems is the engineering adequacy of the theory—its truth and completeness, how many bugs it has in it. Obviously, then, the speed with which the theory improves and the adequacy of feedback about mistakes are crucial as well.

This brings up a second feature of the whole startup process. The engineers and factory managers for an innovation need to be got down into the shop looking at the details. But this means that production workers who are used to running things for themselves have to get used to an engineer or manager always looking over their shoulders. If they have been used to starting work at 8:15, after fifteen minutes of joking and bullshitting, and the big boss comes down at 8:05 to a shop with no machine noise at all, the situation becomes strained. All sorts of illegitimate little arrangements (though they may be agreed to by lower management and tolerated through careful not-noticing by higher management) for making the work easier and more pleasant, the "indulgency pattern" (Gouldner 1954), tend to be undermined. Gouldner found this attentiveness of an innovating management to be a cause of wildcat strikes in a plasterboard factory he studied.

When the theory is known to be uncertain, the company quite often builds a pilot plant, say one-quarter of the projected efficient size. In general, the relations of workers to managers, the variety and excitement of work, the chances of promotion, the pressure for production, are better in the pilot plant. Hence, the creation of the actual production system involves making the pilot plant workers give up their close relations to engineers and higher bosses, making them do more boring work, cutting down their speed of promotion, and putting them under productivity standards that are more demanding. This transition from a sort of test laboratory for the theory on which the innovation is based to the production system that is actually going to produce the innovation thus creates strains that are the reverse of those documented by Gouldner when the innovation was introduced on the assembly line that he studied (Gouldner 1954; see Whyte 1961, 202–217).

Markets and Innovation Success

The market structure with which an innovation is faced determines how much profit the firm in which the innovation might take place may expect to make. This in turn depends on (a) whether the competitive costs are within or outside the organization—that is, whether the destruction of other economic opportunities, which Schumpeter (1942, 81–86) calls "creative destruction," also destroys economic opportunities and advantages that the innovating organization has (as in the case of the cheaper disk drives that IBM developed under the code name Mallard, mentioned above), or whether only economic advantages held by other , competitive, organizations are affected; (b) whether there is an easy market niche to pay for the period when costs are exceptionally high (cf. Grilliches 1957); and (c) whether the profits of monopoly can be retained by continuous improvements .

a. If the costs of competitive damage to unattractive alternatives by the greater attractiveness of the innovation are outside the innovating organization, the innovation is more likely to be introduced. Schumpeter (1942) argued that the central virtue of capitalism was that it could carry out "creative destruction" of technically and economically archaic social structures, primarily because the losses from an innovation occur in a different place than the gains, and in particular because there is no (in actual practice, very little) political communication from one to the other. People hurt by competition with an innovation cannot in general sue those who hurt them for damages, and they cannot stop those benefitting from the introduction of the innovation from introducing it (for a more formal analysis, see Commons 1924, 97–100).

We can immediately see some ways in which this virtue of capitalist social structure is limited in the United States, and we would expect lower rates of innovation because of these limitations. (i) Competition may be politically vulnerable. For example, innovations by the Japanese in the steel industry or by Hong Kong in the textile and apparel industry that damage their competitors in the United States are often limited through political constraints on free competition: tariffs, "voluntary" restraints, and the like. When At&T was the common carrier for telephone traffic, they defended their monopoly on the manufacture of transmitting instruments by not allowing instruments made by competitors to be hooked up to the telephone lines. Political vulnerability of an innovator can transfer costs of competitive damages to the innovator by political means.

(ii) Benefits may be displaced automatically outside the organization that makes the innovation. For example, no one inside the garbage collection department would ordinarily get the benefit of a more rationalized work plan that saved labor—only taxpayers would be benefited. The "Scanlon Plan" of measuring base-rate efficiency and then giving the workers half the savings from labor-saving innovations introduced at their suggestion (or the rather similar arrangement of the West Coast Longshoremen's Union with the longshore employer's association [Finlay 1987]) retains some benefits for the workers, and so increases the rate of innovation insofar as worker resistance would significantly impede innovation. That is, the Scanlon Plan uses the benefits of innovations within public organizations to motivate innovation. When all costs of the innovation's introduction are costs within the innovating organization, then innovation is likely only if some of the benefits of the innovation can be transferred in, for example by special incentive plans.

(iii) Innovations discovered by one organization that would be in the jurisdiction of a competing organization tend not to be developed. For example, navy research rarely develops weapons that would fall under the jurisdiction of the air force, although their technical expertise in aircraft to be based on carriers might make that otherwise likely. If only a competitive organization can exploit an innovation, that innovation is not likely to be pushed.

b. One can classify market segments by how easily they can be penetrated by an innovation. Two sorts of market segments are often reached only by reducing the levels of profit from monopoly over the innovation. First, some segments of the market can be entered only at increasing cost, such as smaller ecological niches for hybrid corn or smaller users of computers requiring larger marketing costs than OEMs (OEMs are

"original equipment manufacturers," a description of computer brokers invented in government bureaucracies—they sell large systems to large computer users by integrating other manufacturers' parts into a system). Second, some sections of the market do not gain particularly large benefits from the innovation, so they buy only when the innovation is cheap. An innovation, then, may depend for early success on being able to identify and reach those sections of the market that will give quick and lush returns.

(i) Government markets, especially the military, are often the first users of expensive innovations. Electronically read Hollerith punch cards (the starting innovation of IBM) were first sold to the Census Bureau; the first computers were sold to the government; the Navy Advanced Research Projects Administration is now a main client for the most advanced computers. Similarly, many transport planes for civilian freight were first developed either for military transport or for civilian passenger traffic, while the electronics for landing and taking off in bad weather was first developed for military use and only later applied to civilian air transport. The government payoff for having the innovation, especially the military payoff, may be very high, making it a lush market for innovations.

(ii) Raymond Vernon has argued (1960, 1966) that rich countries and rich cities are more likely to introduce innovations than poor ones. The argument has three parts. First, many innovations can be sold as luxury goods before they become cheap enough to be sold to a mass market. In a world where "rich" feudal landowners in poor South American countries often have about a fifth the income of North American steelworkers, the real luxury market tends to be concentrated in rich countries, and within those countries in the cities where the rich and powerful live, such as New York, Paris, or Tokyo. Innovators in those countries and cities, then, have the best chance of reaching the lush luxury markets that can sustain the costs of the high part of the learning curve. Second, advanced countries are advanced in part because they have developed ways of marketing innovations and of concentrating that marketing in great metropolitan centers; the rich of poor South American countries may often be more easily reached from New York than by an innovator in the South American country itself. The Marcoses and the Shah of Iran used to come to the United States or Europe to spend their billions. Third, the productivity of an innovation in capital equipment is likely to be proportional to the total volume of production to which it is applied, and that total volume is likely to be higher in countries with more production per capita.

(iii) Rank, prestige, and the network density of a consumer is associ-

ated in many fields with likelihood of adopting an innovation, even when someone else is paying for it (see Coleman, Katz, and Menzel 1957). If a firm or its members are close, in network terms, to the center of the network of consumers of the innovation, then the firm can reach the likely early adopters more easily. Thus we would expect innovations that are made at leading teaching hospitals to be diffused earlier among physicians, or those made at prestigious departments of a scientific discipline to become known earlier by people doing related work. Innovations connected to the centers of consumer networks should find the richest market niches more quickly.

(iv) In some lines of business, geographical similarity between the locus of the innovation and that of the adopting unit facilitates adoption of the innovation. Thus, in early modern times a pattern of crop rotation, artificial fertilization, and other innovations whose center was Flanders increased the yield-to-seed ratio (a rough measure of productivity in grain agriculture) from around 4:1 to around 12:1. This complex of innovations spread across the geographically similar rich plains of northern France at a more or less steady rate of about ten to fifteen miles per decade (Slicher Van Bath 1964, 330–333), but stopped at the borders of granite and sandy soils: the hills and mountains of Brittany, the massif central , the Ardennes, and the Alps. Just as many plants native to the isolated valleys of the Himalayas or Ethiopia only spread outside them into protected gardens and houseplant pots elsewhere (e.g., many roses, African violets), so innovations can have a limited radius because they are developed in an environment inhospitable to their spread. Such innovations may not find a niche rich enough to sustain them because they are not fitted for other geographical environments into which they would have to spread (Wallerstein 1974, 1980).

c. The monopoly advantage of innovators may be sustained by introducing new innovations that are tied to or elaborate on the first innovation. Boeing, for instance, often seems to come down the learning curve to manufacture a new airplane faster than its competitors by introducing production innovations more quickly. IBM manages to continue to sell its computers partly by making it easy for its customers to get new software and peripherals that will increase the productivity of their current installations. Hughes Tool in the oil business can sell innovations better than many of its competitors because it can service the tools anywhere in the world through an existing service apparatus (Boeing has the same advantage).

All these companies can market their secondary innovations relatively easily because the links they have with their customers have induced

those customers to trust the vendor. Such trust is particularly crucial in selling an innovation, because the client cannot turn to alternative sources—such as competitors—to check out the vendor's sales information. Put another way, companies that have innovated successfully in the past are most likely to have access to that part of the market that will buy innovations.

The general point here is that features of the information-processing and decision-making structure attached to an innovation can determine the market success of that innovation. If that social structure does not have to take into much account the losses created by investments in the goods it replaces; if its competition is not protected by tariff or other barriers; if the benefits of the innovation do not automatically accrue to clients; if the innovating social structure's contacts are with the richest markets in government, in metropolitan cities, in innovative and educated groups, in the geographical area where the innovation applies; if the innovation has a stream of other innovations connected to the first one—then the innovation brings prosperity. If the innovating social structure's market contacts are poor, isolated, behind a prohibitive tariff or other barriers, and uninterested in buying improvements, the innovation fails.

The Division of Benefits

The market theory of an innovation says that there will be benefits to the innovating organization that must be divided up, but it does not say how. At least in economic organizations, the basic notion is that the total benefits that come back to the organization itself (the appropriable benefits) have to be divided up to motivate the activities that go into the innovation. I will analyze this under two headings: (d) whether rewards are closely tied to the growth rate of one's section or organization, and consequently generally whether the innovation is a new product (growth inducing, hence rewarding) or labor-capital saving (decline inducing, hence punishing); and (e) whether there are powerful residual claimants on the profits or benefits of the innovation—people whose interest is in reducing costs and taking the profits out rather than in motivating and rewarding everyone generously out of monopoly benefits.

d. Growth dependence of rewards is actually a complex variable. It takes a positive growth in the size of the innovating organization, plus some "firm-specific human capital" or "job rights" to tie the rewards to the people already employed in the enterprise, to produce from an innovation career advantages for presently employed people. It is because civil servants in the garbage department or West Coast longshoremen

(Finlay 1987) cannot be fired to allow the work to be hired out more cheaply that the Scanlon Plan or the sharing of the benefits of labor-saving changes by longshoremen motivates acceptance of innovations. If the workers expected that as soon as the innovations pushed their wage rate above the market rate they would be replaced by new workers at the market rate, they would not be inclined to support innovations; hence, job rights in the innovating organization are essential to turn growth into an incentive for workers. If a whole labor market area is involved in the innovations, as Houston was (once) for offshore drilling in the Gulf of Mexico, then engineers may participate in the boom even if their own firm goes under. In such cases the job rights are in the industry rather than the firm.

In some cases, labor-saving innovations may produce enough of a competitive advantage to maintain positive growth of the labor force, and so to maintain job rights—for instance, Japanese and Korean steel plants have at times expanded production when the world steel market as a whole was in glut.

Product innovations may not produce growth if they are nonpatentable and easily imitated. In general, then, the connection of innovation to growth of the firm is through monopoly, either monopoly of lower production costs through labor-saving innovations or monopoly of the market niche through product innovations and established market position. And the connection of growth of the firm to career rewards for workers is through worker control of their jobs, which in turn can come through indispensability, through labor union power, or through civil service regulation or other legal or customary tenure provisions. These interacting forces on career incentives for building an invention into an innovation are outlined in Table 2.

e. "Residual claimants" are those people who take up the excess of revenues over costs, and in particular those who can be expected to get most of the benefits of establishing a monopoly through introducing an innovation. These residual claimants can range in power from the anonymous and disorganized clients of a city's garbage system, who will pay one-tenth of a mill lower tax rate if garbage workers work more efficiently, to the family-owner majority of Du Pont's board of directors in the days Chandler was talking about. The basic and easy generalization is that the more powerful the residual claimants are, the more likely a profitable innovation will be introduced.

The probability of an innovation's succeeding may be drastically increased by changing the residual claimants—for example, when the developers of an innovation leave the firm in which they have developed

the competence to introduce an innovation, to exploit that competence in a firm of their own. Leaving is especially likely if the company where the developers were employed has already disapproved the innovation. Many large computer and software firms were started in this way (for example, the Data General firm analyzed by Kidder 1981; for software, see Stinchcombe and Heimer 1987). Passing out equity interests to leading technical innovators is one way computer firms have tried to stay at the leading edge of technology.

Merton has argued that the capacity to establish priority through publishing makes the person who publishes the residual claimant for innovations in science—that is, the one whose fame is increased by the innovation. His argument is that priority established by publication increases the motivation to carry out an innovation (Merton [1942] 1973b, 273–275). Joseph Ben-David (1968–1969) has extended Schumpeter's argument above by arguing that prestige competition among universities is crucial for the development of leading scientific nations, for then the costs of older discoverers whose fame is decreased are displaced outside the university introducing the innovation. He holds that nations with many regional universities, such as Germany and the United States, tend to be more scientifically productive, particularly in new fields, than countries with more centralized university systems, such as England and France.

Examples of Incentives for Innovation

We can specify which ends of these marketing and career reward variables predict easy introduction of an innovation by considering the account given by Tracy Kidder (1981) of the design and introduction of a new computer at Data General. (a) DEC VAX, the competitor to the machine with a soul, takes the losses if the Data General computer comes up to its promised level of performance. Contrast this situation with the introduction of a new computer at IBM, where most of the users who buy the new machine will be replacing an IBM machine. (b) The original equipment manufacturers (OEMs) provided an easy market for technically better products, which they in turn marketed as systems to the military and other large users; Data General could start making money on the innovation before any general marketing was required. (c) A series of "downward compatible" and "upward compatible" related models, improved software, and the like enable computer companies in general to maintain their monopolies created by providing an element crucial to the whole system, such as a main computer, so Data General expected to benefit from the innovation for a long time.

(d) Career rewards for computer folks are in general tied to their firm's growth rates. As we suggested above, innovations in the product itself tend to spur firm growth, whereas innovations on the production line tend to depress growth. The Data General innovation was a product innovation, and most of the people Kidder describes as being crucial in the innovation were employees rather than owners or stockholders; they could thus expect their rewards to depend on growth in the size of the labor force rather than, say, on increased profits from cost reductions. And (e) the powerful group in the company was the group who would collect what was left over of the benefits of the innovation after costs were deducted—the residual claimants. The chief executives at Data General were large equity holders and had killed the last innovation, one carried out by some of the people who implemented the innovation studied by Kidder. That is, the control over the introduction of this particular computer was not in the hands of those most committed to it and most motivated to see that innovation introduced—the engineers who worked on it. The top executives did not care whether this one or the one being developed in a competitive company group at another installation in North Carolina was ultimately put on the market. The technical group had to convince the money group. But once convinced the innovation would be profitable, they were strongly motivated to introduce it.

In contrast we can consider a labor-saving innovation in schools, such

as teaching two hundred children at once by television. (a) Competitive costs are paid by teachers who are laid off. (b) There is no special market for easily automatable classes; students will not flock to required lecture-demonstration classes such as civics, American history, and English because they are automated and so cheaper. (c) Schools are already virtual monopolies, and there is no monopoly advantage of a more efficient school that could be maintained by spreading the innovation to related areas, because the school already has all the monopoly it will ever get by law. (d) Saving labor reduces the growth rate of the school labor force, and so reduces the number of supervisory or other lucrative and prestigious positions that will be created. And (e) there are no powerful claimants of the excess profits; any benefits go to the taxpayers. But since the constraining law on schools is that they should have equal expenditures , not equal effects, if they save money one way they will probably have to spend it another and there will probably not be any benefits for the taxpayers anyway.

It would be difficult to set up a social system less hospitable to laborsaving innovations than American elementary and secondary schools, and it would be difficult to set up a situation more encouraging to product innovations than that described by Kidder for Data General—except perhaps by making the engineers in the innovation group shareholders in the company.

Divisionalization and Innovation

Innovations not only require a complete social system with markets, division of benefits, personnel policies, and the like, just as any manufacturing enterprise does; but they require rapidly adjusting social systems as well. Because the market for an innovation is untested, new information about what will sell and at what price is always coming in, and manufacturing and engineering have to be adjusted to that rapid flow of new market information. Marketing people selling machines need, for example, machines that will actually work, and if the machines have been improperly soldered at the factory (because the soldering is also a procedure carried out under new conditions for this innovation), they need to be able to call on engineering and manufacturing people to come out and solder them correctly at the client's place of business (see Fishman 1981, 43). Since getting production costs down requires introducing innovations on the production line, and since innovations on the production line also have bugs, engineers have to come down and take a hand in manufacturing. That is, the introduction of a product innovation typi-

cally requires the sort of intimate relationship between manufacturing, engineering, and marketing that Chandler argued produced the requirements for divisional organization in Du Pont, General Motors, and other chemical, electrical, and automobile manufacturing organizations (1962, 52–162, 363–378).

Of course, Chandler was usually arguing that once a corporation began to market to distinct markets it would be under pressure to introduce a divisional organization. Here we are asking not whether an innovation that is successful will cause divisionalization of the corporation, but whether an innovation will be introduced successfully in the first place. What interests us, then, is not why the chemicals industry, which often introduces innovations requiring responses to new markets, is divisionalized, but rather what it is that makes the chemicals industry capable of introducing many innovations.

The functional argument of the last chapter can however be reversed to say that it will be easier to build a system that can introduce an innovation if it is easy in a given organization to organize a new division, largely autonomous from central manufacturing, marketing, and engineering, so as to allow rapid response as the news needed to make the new social system go comes in.

We noted in Chapter 4 that the Du Pont divisionalization was based partly on an analysis of Du Pont experience in introducing innovations. Let us quote again from Harry Haskell of Du Pont (in Chandler 1962, 68) on why a minidivisional arrangement was needed for the dyes business, which was at that time an innovating organization, before Du Pont became generally organized into product divisions:

It may be that it would be better for a few years to carry on the dye business as a separate entity. I think it would because it is a developing, unstandardized industry and should merit independent attention just as the Parlin chemical mixtures business was better by itself until standardized—when it was merged with the regular sales and operating departments.

And then the conclusion of the committee based on this analysis (Chandler 1962, 70):

[Dyes should have one] individual in control of both production and sales, because the relation of the product and its qualities is so mixed up with the demands of the market for the product that to divorce them and segregate the business into a clearly defined production

department and an independent sales department, would be detrimental to the business. Later on when the production of dyes becomes standardized it will no doubt follow the evolution of other portions of the business.

In the final analysis, then, the dyes business required what Chandler calls (when it happens later) a divisional administrative structure because it is carrying an innovation. But this means not only that Du Pont had the category of division in its organizational analysis before it introduced divisionalization on a large scale; they were also used to introducing autonomous divisions to manage product innovations, and although this was treated as an exception to normal good centralized management, it was a "normal exception."

Presumably the causal relation between innovation and decentralized minidivisions goes both ways. Innovating companies learn to create autonomous minidivisions easily; companies that create minidivisions easily are more successful at innovation. Stated more generally: one can easily create a new social system to fit a given niche only if allowed considerable autonomy, and larger social orders that allow considerable autonomy to new social units are therefore advantaged in introducing innovations. This is one of Schumpeter's main arguments in favor of capitalism—that it allows innovating entrepreneurs enough autonomy to produce economic progress. Thus Schumpeter's advice on how to organize an economy can be turned into advice to bureaucratic corporations: only routinize and subordinate an innovating minidivision of a large bureaucracy when its adaptation to its niche and its viability as a going concern are established.

Conclusion

Our first task in this chapter was to show that the monopoly advantage from innovations, which Schumpeter saw as the driving force of capitalist progress and of business cycles, is increased and decreased by social structural variables. The profitability of innovations depends on how fast the innovator and potential competitors come down the learning curve, and on how solid the network connections an innovator builds to its clients are.

Our second task was to show that turning an invention into an innovation—into a going concern that can regularly produce benefits for an innovating firm—is in fact creating a social system. A social system has to meet its functional requirements in order to produce benefits over the

long run. This means it has to nurture technical ideas, to make investments in risky situations, to build a production system that can produce effectively and can come down the learning curve rapidly, to reach the market that can afford the innovation while it is still expensive, and to arrange the division of benefits so that both investors and personnel will be motivated to develop the competences needed to do all these things and then do them.

In short, because routinizing innovations is still a difficult and risky process, even in large bureaucratic organizations, it still produces excess profits when well done, creates investment booms as followers try to catch up, and induces creative destruction of archaic social forms during the recession that follows the investment boom. Furthermore, since there are special problems in administering viable innovative social systems within large bureaucracies, in that adaptation occurs on a shorter time scale than in more routine production and marketing, divisionalization serves in part as a social process for making innovation possible; it is the organizational creation of entrepreneurial social structures, taking the place of the heroic individual in Schumpeter's model.

In one sense, this chapter is merely a specialized application of the theory developed in previous chapters, especially Chapter 4. It says that in introducing innovations, market uncertainty ramifies into manufacturing and engineering and so tends to produce divisionalization. But turning an invention into an innovation also involves the special task of creating information-processing and decision-making routines. It therefore has distinctive pathologies. As an invention, for example, the normal state of the information-processing system is that of a "technological utopia." In general, the people who developed this technological utopia are not good at collecting and processing the information needed to satisfy an investment group, knowing little about "business plans." They may well not know where the lucrative markets are to be found (unless they developed the innovation for an engineer who is also a colonel in the Pentagon). They may be bad at estimating how fast the bugs can be got out of a production line for the product. And so on.

When we describe what is lacking in a technologically utopian plan for introducing an innovation, we are describing the information-processing requirements for a new product-line division. Each of those information criteria in turn requires skills in the people (as analyzed above in Chapter 2: innovation is a situation of low routinization and hence requires a high average skill level) and the building up of all sorts of departmental information systems (as analyzed in Chapter 3), responding to the distinctive technical uncertainties of the innovation. For

example, the cost-accounting system needed to find the cost bottlenecks in a production line for an innovation would have to be much more flexible than would the cost system serving the overall firm in which the innovation is likely to be embedded.

In Chapter 6, we will analyze some of the distinctive problems of building a system to produce innovations through networks of contracts and subcontracts (see also Stinchcombe and Heimer 1988); and in Chapter 7 we will focus on the information system that has to be embedded in the incentive system for workers. As should be becoming clear, what this book is developing a theory of is the things that have to be added to a technologically utopian theory to turn it into an organized going concern.