Democratizing Innovation by Eric von Hippel - HTML preview

PLEASE NOTE: This is an HTML preview only and some elements such as links or page numbers may be incorrect.
Download the book in PDF, ePub, Kindle for a complete version.

Democratizing Innovation

for in-house use might feel it perfectly acceptable to install a pre-

regarded under US law as also providing an implied warranty of “fit-

cisely right and very cheap computer controller made and promi-

ness for the intended use.” If a product does not meet this criterion,

nently labeled by Lego, a manufacturer of children's toys. (Lego

and if a different, written warranty is not in place, manufacturers can

provides computer controllers for some of its children's building kit

be found liable for negligence with respect to providing a defective

products.) But if that same engineer saw a Lego controller in a

design and failure to warn buyers (Barnes and Ulin 1984). This sim-

million-dollar process machine his firm was purchasing from a spe-

ple difference can cause a large difference in exposure to liability by

cialist high-end manufacturer, he might not know enough about the

innovators and so can drive up the costs of manufacturer-provided

design details to know that the Lego controller was precisely right

solutions relative to user-provided ones.

for the application. In that case, the engineer and his managers

For example, a user firm that builds a novel process controller to

might well regard the seemingly inappropriate brand name as an

185

improve its plant operations must pay its own actual costs if the

indirect signal of bad quality.

self-built controller fails and ruins expensive materials being pro-

182

Manufacturers are often so concerned about a reputation for qual-

cessed. On the other hand, if a controller manufacturer designed

ity that they refuse to take shortcuts that a customer specifically

the novel controller product and sold it to customers, and a fail-

requests and that might make sense for a particular customer, lest

ure then occurred and could be traced back to a fault in the de-

others get wind of what was done and take it as a negative signal

sign, the controller manufacturer is potentially liable for actual user

about the general quality of the firm's products. For example, you

costs and punitive damages. It may also incur significant reputa-

may say to a maker of luxury custom cars: “I want to have a cus-

tional losses if the unhappy user broadcasts its complaints. The

tom car of your brand in my driveway---my friends will admire it.

logical response of a controller manufacturer to this higher risk is

But I only plan to drive it to the grocery store once in a while, so

to charge more and/or to be much more careful with respect to

I only want a cheap little engine. A luxury exterior combined with

running exhaustive, expensive, and lengthy tests before releasing

cheap parts is the best solution for me in this application---just slap

a new product. The resulting increase in cost and delay for obtain-

something together and keep the price low.” The maker is likely to

ing a manufacturer-developed product can tend to tip users toward

respond: “We understand your need, but we cannot be associated

building their own, in-house solutions.

with any product of low quality. Someone else may look under the

Net Result

hood some day, and that would damage our reputation as a maker

186

of fine cars. You must look elsewhere, or decide you are willing to

A net result of the foregoing considerations is that manufacturers

187

pay the price to keep one of our fine machines idle on your drive-

often find that developing a custom product for only one or a few

way.”

users will be unprofitable. In such cases, the transaction costs in-

volved can make it cheaper for users with appropriate capabilities

183

Differing Legal and Regulatory Requirements

to develop the product for themselves. In larger markets, in con-

184

Users that innovate do not generally face legal risk if the product

trast, fixed transaction costs will be spread over many customers,

they develop fails and causes costs to themselves but not to others.

and the economies of scale obtainable by producing for the whole

In contrast, manufacturers that develop and sell new products are

market may be substantial. In that case, it will likely be cheaper for

SiSU

web.mit.edu/evhippel/www/

34

Democratizing Innovation

users to buy than to innovate. As a result, manufacturers, when

duce biased advice, they may attempt to shop around among a

contacted by a user with a very specific request, will be keenly in-

number of suppliers offering different solution types and/or develop

terested in how many others are likely to want this solution or ele-

internal expertise on solution possibilities and/or attempt to write

ments of it. If the answer is “few,” a custom manufacturer will be

better contracts. All these attempts to induce and guard against

unlikely to accept the project.

bias involve agency costs.

An Illustrative Case

190

188

Of course, manufacturers have an incentive to make markets at-

tractive from their point of view. This can be done by deviating from

A case study by Sarah Slaughter (1993) illustrates the impact of

191

precisely serving the needs of a specific custom client in order to

some of the transaction costs discussed above related to users'

create a solution that will be “good enough” for that client but at the

innovate-or-buy decisions. Slaughter studied patterns of innova-

same time of more interest to others. Manufacturers may do this

tion in stressed-skin panels, which are used in some housing con-

openly by arranging meetings among custom buyers with similar

struction. The aspects of the panels studied were related to instal-

needs, and then urging the group to come up with a common solu-

lation, and so the users of these features were home builders rather

tion that all will find acceptable. “After all,” as the representative will

than home owners. When Slaughter contrasted users' costs of in-

say, “it is clear that we cannot make a special product to suit each

novating versus buying, she found that it was always much cheaper

user, so all of you must be prepared to make really difficult compro-

for the builder to develop a solution for itself at a construction site

mises!” More covertly, manufacturers may simply ignore some of

than to ask a panel manufacturer to do so.

the specific requests of the specific user client and make something

A stressed-skin panel can be visualized as a large 4-by-8-foot

192

that they expect to be a more general solution instead.

sandwich consisting of two panels made of plywood with a layer

of plastic foam glued in between. The foam, about 4 inches thick,

189

The contrasting incentives of users and manufacturers with respect

strongly bonds the two panels together and also acts as a layer of

to generality of need being served---and also with respect to the

thermal insulation. In 1989, manufacturing of stressed-skin panels

solution choice issue discussed earlier---can result in a very frus-

was a relatively concentrated industry; the four largest manufactur-

trating and cloudy interaction in which each party hides its best

ers collectively having a 77 percent share of the market. The user

information and attempts to manipulate others to its own advan-

industry was much less concentrated: the four largest constructors

tage. With respect to generality of need, sophisticated users un-

of panelized housing together had only 1 percent of the market for

derstand custom suppliers' preference for a larger market and at-

such housing in 1989.

tempt to argue convincingly that “everyone will want precisely what

I am asking you for.” Manufacturers, in turn, know users have this

In housing construction, stressed-skin panels are generally at-

193

incentive and so will generally prefer to develop custom products

tached to strong timber frames to form the outer shell of a house

for which they themselves have a reasonable understanding of de-

and to resist shear loads (such as the force of the wind). To use

mand. Users are also aware of manufacturers' strong preference

the panels in this way, a number of subsidiary inventions are

for only producing products that embody their existing solution ex-

required. For example, one must find a practical, long-lasting way

pertise. To guard against the possibility that this incentive will pro-

to attach panels to each other and to the floors, the roof, and the

SiSU

web.mit.edu/evhippel/www/

35

Democratizing Innovation

frame. Also, one has to find a new way to run pipes and wires from

A builder was faced with the immediate problem of how to route

197

place to place because there are no empty spaces in the walls to

wires through the foam interior of panels to wall switches located

put them---panel interiors are filled with foam.

in the middle of the panels. He did not want cut grooves or channels

through the surfaces of the panels to these locations---that would

194

Stressed-skin panels were introduced into housing construction

dangerously reduce the panels' structural strength. His inventive

after World War II. From then till 1989, the time of Slaughter's

solution was to mount an electrically heated wire on the tip of a long

study, 34 innovations were made in 12 functionally important ar-

pole and simply push the heated tip through the center insulation

eas to create a complete building system for this type of construc-

layer of the panel. As he pushed, the electrically heated tip quickly

tion. Slaughter studied the history of each of these innovations

melted a channel through the foam plastic insulation from the edge

and found that 82 percent had been developed by users of the

of the panel to the desired spot. Wires were then pulled through

stressed-skin panels---residential builders---and only 18 percent by

this channel.

manufacturers of stressed-skin panels. Sometimes more than one

user developed and implemented different approaches to the same

Table 4.1 Users would have found it much more costly to get cus-

198

functional problem (table 4.1). Builders freely revealed their in-

tom solutions from manufacturers. The costs of user-developed

novations rather than protecting them for proprietary advantage.

innovations in stressed-skin panels were very low.

They were passed from builder to builder by word of mouth, pub-

199

lished in trade magazines, and diffused widely. All were replicated

Function

Average

user

Average

user

N

Minimimum cost of

at building sites for years before any commercial panel manufac-

development

development

waiting for manu-

time (days)

cost

facturer to deliver

turer developed and sold a solution to accomplish the same func-

tion.

Framing of openings in panels

0.1

20

1

1 , 400

Structural connection between panels

0.1

30

2

1 , 400

195

Histories of the user-developed improvements to stressed-skin

Ventilation of panels on roof

0.1

32

2

28 , 000

panel construction showed that the user-innovator construction

Insulated connection between panels

0.1

41

3

2 , 800

firms did not engage in planned R&D projects.

Instead, each

Corner connection between panels

0.2

60

1

2 , 800

innovation was an immediate response to a problem encountered

Installation of HVAC in panels

0.2

60

2

2 , 800

in the course of a construction project.

Once a problem was

Installation of wiring in panels

0.2

79

7

2 , 800

encountered, the innovating builder typically developed and

Connection of panels to roof

0.2

80

1

2 , 800

fabricated a solution at great speed, using skills, materials, and

Add insect repellency to panels

0.4

123

3

70 , 000

equipment on hand at the construction site. Builders reported that

Connect panels to foundation

0.5

160

1

1 , 400

the average time from discovery of the problem to installation of

Connect panels to frames

1.2

377

3

2 , 800

the completed solution on the site was only half a day. The total

Development of curved panels

5.0

1,500

1

28 , 000

cost of each innovation, including time, equipment, and materials,

Average for all innovations

0.5

153

12 , 367

averaged $153.

196

Example: Installing Wiring in a Stressed-Skin Panel

SiSU

web.mit.edu/evhippel/www/

36

Democratizing Innovation

200

N represents number of innovations developed by users to carry

above and below these windows, but panel manufacturers only

out each listed function. Source: Slaughter 1993, tables 4 and

sold flat panels at that time. The builder facing the problem could

5. Costs and times shown are averaged for all user-developed

not simply buy standard flat panels and bend them into curved ones

innovations in each functional category. (The six manufacturer-

at the construction site---completed panels are rigid by design.

developed innovations in Slaughter's sample are not included in

So he bought plywood and plastic foam at a local building supply

this table.)

house and slowly bent each panel component separately over a

curved frame quickly built at the construction site. He then bonded

201

The builder-innovator reported that the total time to develop the

all three elements together with glue to create strong curved panels

innovation was only an hour, and that the total cost for time and

that would maintain their shape over time.

materials equaled $40. How could it cost so little and take so little

time? The builder explained that using hot wires to slice sheets of

To determine whether users' decisions to innovate rather than

205

plastic foam insulation into pieces of a required length is a tech-

buy made economic sense for them, Slaughter calculated, in a

nique known to builders. His idea as to how to modify the slicing

very conservative way, what it would have cost users to buy a

technique to melt channels instead came to him quickly. To test the

manufacturer-developed solution embodied in a manufactured

idea, he immediately sent a worker to an electrical supply house to

panel rather than build a solution for themselves. Her estimates

get some nichrome wire (a type of high-resistance wire often used

included only the cost of the delay a user-builder would incur

as an electrical heating element), attached the wire to a tip of a

while waiting for delivery of a panel incorporating a manufacturer's

pole, and tried the solution on a panel at the building site---and it

solution. Delay in obtaining a solution to a problem encountered

worked!

at a construction site is costly for a builder, because the schedule

of deliveries, subcontractors, and other activities must then be

202

This solution was described in detail in an article in a builder's mag-

azine and was widely imitated. A panel manufacturer's eventual re-

altered. For example, if installation of a panel is delayed, one must

sponse (after the user solution had spread for a number of years)

also reschedule the arrival of the subcontractor hired to run wires

was to manufacture a panel with a channel for wires pre-molded

through it, the contractor hired to paint it, and so on. Slaughter

into the plastic foam interior of the panel. This solution is only

estimated the cost of delay to a builder at $280 per crew per day of

sometimes satisfactory. Builders often do not want to locate switch

delay (Means 1989). To compute delay times, she assumed that a

boxes at the height of the premolded channel. Also, sometimes

manufacturer would always be willing to supply the special item a

construction workers will install some panels upside down in error,

user requested. She also assumed that no time elapsed while the

and the preformed channels will then not be continuous between

manufacturer learned about the need, contracted to do the job,

one panel and the next. In such cases, the original, user-developed

designed a solution, and obtained needed regulatory approvals.

solution is again resorted to.

She then asked panel manufacturers to estimate how long it would

take them to simply construct a panel with the solution needed

203

Example: Creating a Curved Panel

and deliver it to the construction site. Delay times computed in this

204

A builder was constructing a custom house with large, curved win-

manner ranged from 5 days for some innovations to 250 days for

dows. Curved stressed-skin panels were needed to fill in the space

the longest-term one and averaged 44 days.

SiSU

web.mit.edu/evhippel/www/

37

Democratizing Innovation

206

The conservative nature of this calculation is very clear. For ex-

user problem. We also assume that both user firms and manufac-

ample, Slaughter points out that the regulatory requirements for

turer firms will incur the same costs to solve a specific user problem.

building components, not included, are in fact much more strin-

For example, they will have the same costs to monitor the perfor-

gent for manufacturers than for user-builders in the field of res-

mance of the designer employees they hire. In this way we simplify

idential construction. Manufacturers delivering products can be

our innovate-or-buy problem to one of transaction costs only.

required to provide test data demonstrating compliance with local

If there are no transaction costs (for example, no costs to write

211

building codes for each locality served. Testing new products for

and enforce a contract), then by Coase's theorem a user will be

compliance in a locality can take from a month to several years,

indifferent between making or buying a solution to its problem. But

and explicit code approval often takes several additional years.

in the real world there are transaction costs, and so a user will

In contrast, a builder that innovates need only convince the lo-

generally prefer to either make or buy. Which, from the point of

cal building inspector that what he has done meets code or per-

view of minimizing overall costs of obtaining a problem solution, is

formance requirements--- often a much easier task (Ehrenkrantz

the better choice under any given circumstances?

Group 1979; Duke 1988).

Let Vij be the value of a solution to problem j for user i. Let Nj be the 212

207

Despite her very conservative method of calculation, Slaughter

number of users having problem j. Let Wh

found the costs to users of obtaining a builder solution to be at

j be the cost of solving

least 100 times the actual costs of developing a solution for them-

problem j, where W = hourly wage and hj = hours required to solve

selves (table 4.1). Clearly, users' decisions to innovate rather than

it. Let Pj be the price charged by a manufacturer for a solution

buy made economic sense in this case.

to problem j. Let T be fixed or “setup” transaction costs, such as

writing a general contract for buyers of a solution to problem j. Let

208

Modeling Users' Innovate-or-Buy Decisions

t be variable or “frictional” transaction costs, such as tailoring the

209

In this section I summarize the core of the argument discussed in

general contract to a specific customer.

this chapter via a simple quantitative model developed with Carliss

To explore this problem we make two assumptions. First, we as-

Baldwin. Our goal is to offer additional clarity by trading off the

213

sume that a user firm knows its own problems and the value of

richness of the qualitative argument for simplicity.

a solution to itself, Vij. Second, we assume that a manufacturer

210

Whether a user firm should innovate or buy is a variant of a well-

knows the number of users having each problem, Nj, and the value

known problem: where one should place an activity in a supply

of solutions for each problem for all users, V

chain. In any real-world case many complexities enter. In the

ij.

model that follows, Baldwin and I ignore most of these and consider

These assumptions are in line with real-world incentives of users

214

a simple base case focused on the impact of transaction costs on

and manufacturers, although information stickiness generally pre-

users' innovate-or-buy considerations. The model deals with man-

vents firms from getting full information. That is, users have a high

ufacturing firms and user firms rather than individual users. We

incentive to know their own problems and the value to them of a

assume that user firms and manufacturer firms both will hire de-

solution. Manufacturers, in turn, have an incentive to invest in un-

signers from the same homogeneous pool if they elect to solve a

derstanding the nature of problems faced by users in the target

SiSU

web.mit.edu/evhippel/www/

38

Democratizing Innovation

market, the number of users affected, and the value that the users

Nj (Whj - t) - Whj > 0,

224

would attach to getting a solution in order to determine the potential

or equivalently (provided Wh

profitability of markets from their point of view.

j > t)

225

N

215

We first consider the user's payoff for solving a problem for itself.

j > Whj / (Whj - t) > 1.

226

A user has no transaction costs in dealing with itself, so a user's

Again, users will not call on upstream manufacturers to solve prob-

227

payoff for solving problem j will be Vij - Whj. Therefore, a user will

lems unique to one user.

buy a solution from an upstream manufacturer rather than develop

one for itself if and only if P ≤ Wh

The findings from the simplified model, then, are the following:

228

j.

Problems unique to one user will always be solved efficiently by

216

Next we consider payoffs to a manufacturer for solving problem j.

users hiring designers to work for them in house. In contrast, prob-

In this case, transaction costs such as those discussed in earlier

lems affecting more than a moderate number of users, n, which is

sections will be encountered. With respect to transaction costs as-

a function of the transaction costs, will be efficiently solved by the

sume first that t = 0 but T > 0. Then, the manufacturer's payoff for

manufacturer hiring designers to develop the needed new product

solving problem j will be Vij - Whj, which needs to be positive in

or service and then selling that solution to all users affected by the

order for the manufacturer to find innovation attractive:

problem. However, given sufficient levels of T and/or of t, problems

affecting more than one but fewer than n users will not be solved by

217

Nj Pj - Whj - T > 0.

a manufacturer, and so there will be a market failure: Assuming an

218

But, as we saw, Pj ≤ Whj if the user is to buy, so we may substitute

institutional framework consisting only of independent users and

Whj for Pj in our inequality. Thus we obtain the following inequality

manufacturers, multiple users will have to solve the same problem

as a condition for the user to buy:

independently.

219

Nj (Whj) - Whj - T > 0,

As illustration, suppose that t = 0.25Whj and T = 10Whj. Then, 229

combining the two expressions and solving for n yields

220

or

n = (11Wh

221

Nj > (T / Whj) + 1.

j /0.75Whj) = 14.66.

230

The condition for the user to buy the innovation rather than inno-

222

In other words, Baldwin and I find that the absolute lower bound

231

on N is greater than 1. This means that a single user will always

vate itself becomes Nj ≥ 15. For a number of users less than 15

prefer to solve a unique problem j for itself (except in Coase's world,

but greater than 1, there will be a wasteful multiplication of user

where T = 0, and the user will be indifferent). If every problem is

effort: several users will invest in developing the same innovation

unique to a single user, users will never choose to call on upstream

independently.

manufacturers for solutions.

In a world that consists entirely of manufacturers and of users that

232

223

Now assume that T = 0 but t > 0. Then the condition for the user

do not share the innovations they develop, the type of wasteful du-

to buy rather than to innovate for itself becomes

plicative innovation investment by users just described probably

SiSU

web.mit.edu/evhippel/www/

39

Democratizing Innovation

will occur often. As was discussed earlier in this chapter, and as

being developed. Make-or-buy evaluations typically include fac-

was illustrated by Slaughter's study, substantial transaction costs

tors such as the time and materials that must be invested to de-

might well be the norm. In addition, low numbers of users hav-

velop a solution. These costs are then compared with the likely

ing the same need---situations where Nj is low---might also be the

benefits produced by the project's “output”---the new product or

norm in the case of functionally novel innovations. Functionally

service created---to determine whether the project is worth doing.

novel innovations, as I will show later, tend to be developed by

This was the type of comparison made by Slaughter, for example,

lead users, and lead users are by definition at the leading (low-N

in assessing whether it would be better for the users to make or to

j)

edge of markets.

buy the stressed-skin panel innovations in her sample. However, in

the case of individual user-innovators, this type of assessment can

233

When the type of market failure discussed above does occur, users

provide too narrow a perspective on what actually constitutes valu-

will have an incentive to search for institutional forms with a lower T

able project output. Specifically, there is evidence that individuals

and/or a lower t than is associated with assignment of the problem

sometimes greatly prize benefits derived from their participation in

to an upstream manufacturer. One such institutional form involves

the process of innovation. The process, they say, can produce

interdependent innovation development among multiple users (for

learning and enjoyment that is of high value to them.

example, the institutional form used successfully in open source

software projects that I will discuss in chapter 7). Baldwin and Clark

In the introductory chapter, I pointed out that some recreational ac-

236

(2003) show how this form can work to solve the problem of waste-

tivities, such as solving crossword puzzles, are clearly engaged in

ful user innovation investments that were identified in our model.

for process rewards only: very few individuals value the end “prod-

They show that, given modularity in the software's architecture, it

uct” of a completed puzzle. But process rewards have also been

will pay for users participating in open source software projects to

found to be important for innovators that are producing outputs that

generate and freely reveal some components of the needed inno-

they and others do value (Hertel, Niedner, and Herrmann 2003;

vation, benefiting from the fact that other users are likely to develop

Lakhani and Wolf 2005). Lakhani and Wolf studied a sample of

and reveal other components of that innovation. At the limit, the

individuals (n = 684, response rate = 34 percent) who had written

wasteful duplication of users' innovative efforts noted above will be

new software code and contributed it to an open source project.

eliminated; each innovation component will have been developed

They asked the programmers to list their three most important rea-

by only one user, but will be shared by many.

sons for doing this. Fifty-eight percent of respondents said that an

important motivation for writing their code was that they had a work

234

Benefiting from the Innovation Process

need (33 percent), or a non-work need (30 percent) or both (5 per-

235

Some individual users (not user firms) may decide to innovate for

cent) for the code itself. That is, they valued the project's “output”

themselves rather than buy even if a traditional accounting evalua-

as this is traditionally viewed. However, 45 percent said that one

tion would show that they had made a major investment in time and

of their top three reasons for writing code was intellectual stimu-

materials for an apparently minor reward in product functionality.

lation, and 41 percent said one of their top three reasons was to

The reason is that individual users may gain major rewards from

improve their own programming skills (Lakhani and Wolf 2005, ta-

the process of innovating, in addition to rewards from the product

ble 6). Elaborating on these responses, 61 percent of respondents

SiSU

web.mit.edu/evhippel/www/

40

Democratizing Innovation

said that their participation in the open source project was their

5 Users' Low-Cost Innovation Niches

239

most creative experience or was as creative as their most creative

experience. Also, more than 60 percent said that “if there were

The Problem-Solving Process

240

one more hour in the day” they would always or often dedicate it to

Product and service development is at its core a problem-solving

programming.

241

process. Research into the nature of problem solving shows it to

237

Csikszentmihalyi (1975, 1990, 1996) systematically studied the

consist of trial and error, directed by some amount of insight as to

characteristics of tasks that individuals find intrinsically rewarding,

the direction in which a solution might lie (Baron 1988). Trial and

such as rock climbing. He found that a level of challenge some-

error has also been found to be prominent in the problem-solving

where between boredom and fear is important, and also that the

work of product and process development (Marples 1961; Allen

experience of “flow” gained when one is fully engaged in a task

1966; von Hippel and Tyre 1995; Thomke 1998, 2003).

is intrinsically rewarding. Amabile (1996) proposes that intrinsic

motivation is a key determining factor in creativity. She defines a

Trial-and-error problem solving can be envisioned as a four-phase

242

creative task as one that is heuristic in nature (with no predeter-

cycle that is typically repeated many times during the development

mined path to solution), and defines a creative outcome as a novel

of a new product or service. Problem solvers first conceive of a

and appropriate (useful) response to such a task. Both conditions

problem and a related solution based on their best knowledge and

certainly can apply to the task of developing a product or a ser-

insight. Next, they build a physical or virtual prototype of both the

vice.

possible solution they have envisioned and the intended use en-

vironment. Third, they run the experiment---that is, they operate

238

In sum, to the extent that individual user-innovators benefit from

their prototyped solution and see what happens. Fourth and fi-

the process of developing or modifying a product as well as from

nally, they analyze the result to understand what happened in the

the product actually developed, they are likely to innovate even

trial and to assess the “error information” that they gained. (In the

when the benefits expected from the product itself are relatively

trial-and-error formulation of the learning process, error is the new

low. (Employees of a firm may wish to experience this type of in-

information or learning derived from an experiment by an exper-

trinsic reward in their work as well, but managers and commer-

imenter: it is the aspect(s) of the outcome that the experimenter

cial constraints may give them less of an opportunity to do so. In-

did not predict.) Developers then use the new learning to modify

deed, “control over my own work” is cited by many programmers

and improve the solution under development before building and

as a reason that they enjoy creating code as volunteers on open

running a new trial (figure 5.1).

source projects more than they enjoy coding for their employers for

pay.)

Trial-and-error experimentation can be informal or formal; the un-

243

derlying principles are the same. As an example on the informal

side, consider a user experiencing a need and then developing

what eventually turns out to be a new product: the skateboard.

In phase 1 of the cycle, the user combines need and solution in-

formation into a product idea: “I am bored with roller skating. How

SiSU

web.mit.edu/evhippel/www/

41

index-45_1.png

Democratizing Innovation

can I get down this hill in a more exciting way? Maybe it would be

the flame in the cylinders is a possible solution direction, and I think

fun to put my skates' wheels under a board and ride down on that.”

that changing the shape of the spark plug electrodes will improve

In phase 2, the user builds a prototype by taking his skates apart

this.” In phase 2, the engineer builds a spark plug incorporating

and hammering the wheels onto the underside of a board. In phase

her new idea. In phase 3, she inserts the new spark plug into a lab

3, he runs the experiment by climbing onto the board and heading

test engine equipped with the elaborate instrumentation needed to

down the hill. In phase 4, he picks himself up from an inaugural

measure the very rapid propagation of a flame in the cylinders of an

crash and thinks about the error information he has gained: “It is

auto engine and runs the test. In phase 4, she feeds the data into

harder to stay on this thing than I thought. What went wrong, and

a computer and analyzes the results. She asks: “Did the change

how can I improve things before my next run down the hill?”

in spark plug design change the flame front as expected? Did it

change fuel efficiency? How can I use what I have learned from

this trial to improve things for the next one?”

In addition to the difference in formality, there is another important

247

difference between these two examples. In the first example, the

skateboard user was conducting trial and error with a full prototype

of the intended product in a real use environment---his own. In the

second example, the experimental spark plug might have been a

full prototype of a real product, but it probably consisted only of that

portion of a real spark plug that actually extends into a combustion

chamber. Also, only aspects of the use environment were involved

in the lab experiment. That is, the test engine was not a real auto

engine, and it was not being operated in a real car traveling over

real roads.

Experimentation is often carried out using simplified versions---

248

models--- of the product being designed and its intended use en-

244

vironment. These models can be physical (as in the example just

given), or they can be virtual (as in the case of thought experiments

245

Figure 5.1 The trial-and-error cycle of product development.

or computer simulations). In a computer simulation, both the prod-

246

As an example of more formal experimentation, consider a product-

uct and the environment are represented in digital form, and their

development engineer working in a laboratory to improve the per-

interaction is tested entirely within a computer. For example, one

formance of an automobile engine. In phase 1, need and solution

might make a digital model of an automobile and a crash barrier.

information are again combined into a design idea: “I need to im-

One could then use a computer to simulate the crash of the model

prove engine fuel efficiency. I think that a more even expansion of

car into the model barrier. One would analyze the results by calcu-

SiSU

web.mit.edu/evhippel/www/

42

Democratizing Innovation

lating the effects of that crash on the structure of the car.

cultures, animal models, etc.) before finally seeking to test its effect

on real human patients during clinical trials (Thomke, von Hippel,

249

The value of using models rather than the real thing in experimen-

and Franke 1998).

tation is twofold. First, it can reduce the cost of an experiment-

--it can be much cheaper to crash a simulated BMW than a real

Sticky Information

252

one. Second, it can make experimental results clearer by making

Any experiment is only as accurate as the information that is used

253

them simpler or otherwise different than real life. If one is trying to

as inputs. If inputs are not accurate, outcomes will not be accurate:

test the effect of a small change on car safety, for example, it can

“garbage in, garbage out.”

be helpful to remove everything not related to that change from

the experiment. For example, if one is testing the way a particular

The goal of product development and service development is to

254

wheel suspension structure deforms in a crash, one does not have

create a solution that will satisfy needs of real users within real

to know (or spend time computing) how a taillight lens will react in

contexts of use. The more complete and accurate the information

the crash. Also, in a real crash things happen only once and hap-

on these factors, the higher the fidelity of the models being tested.

pen very fast. In a virtual crash executed by computer, on the other

If information could be transferred costlessly from place to place,

hand, one can repeat the crash sequence over and over, and can

the quality of the information available to problem solvers would

stretch time out or compress it exactly as one likes to better under-

or could be independent of location. But if information is costly

stand what is happening (Thomke 2003).

to transfer, things are different. User-innovators, for example, will

then have better information about their needs and their use context

250

Users and others experimenting with real prototypes in real use

than will manufacturers. After all, they create and live in that type

environments can also modify things to make tests simpler and

of information in full fidelity! Manufacturer-innovators, on the other

clearer. A restaurant chef, for example, can make slight varia-

hand, must transfer that information to themselves at some cost,

tions in just a small part of a recipe each time a customer calls

and are unlikely to be able to obtain it in full fidelity at any cost.

for it, in order to better understand what is happening and make

However, manufacturers might well have a higher-fidelity model of

improvements. Similarly, a process machine user can experiment

the solution types in which they specialize than users have.

with only a small portion of machine functioning over and over to

It turns out that much information needed by product and service

test changes and detect errors.

255

designers is “sticky.” In any particular instance, the stickiness of

251

Sometimes designers will test a real experimental object in a real

a unit of information is defined as the incremental expenditure re-

experimental context only after experimenting with several gener-

quired to transfer that unit of information to a specified location in a

ations of models that isolate different aspects of the real and/or

form usable by a specified information seeker. When this expendi-

encompass increasing amounts of the complexity of the real. De-

ture is low, information stickiness is low; when it is high, stickiness

velopers of pharmaceuticals, for example, might begin by testing

is high (von Hippel 1994). That information is often sticky has been

a candidate drug molecule against just the purified enzyme or re-

shown by studying the costs of transferring information regarding

ceptor it is intended to affect, then test it again and again against

fully developed process technology from one location to another

successively more complex models of the human organism (tissue

with full cooperation on both sides. Even under these favorable

SiSU

web.mit.edu/evhippel/www/

43

Democratizing Innovation

conditions, costs have been found to be high---leading one to con-

pacity. A firm's or an individual's capacity to absorb new, outside

clude that the costs of transferring information during product and

technical information is largely a function of prior related knowl-

service development are likely to be at least as high. Teece (1977),

edge (Cohen and Levinthal 1990). Thus, a firm knowing nothing

for example, studied 26 international technology-transfer projects

about circuit design but seeking to apply an advanced technique

and found that the costs of information transfer ranged from 2 per-

for circuit engineering may be unable to apply it without first learn-

cent to 59 percent of total project costs and averaged 19 percent---

ing more basic information. The stickiness of the information about

a considerable fraction. Mansfield et al. (1982) also studied a num-

the advanced technique for the firm in question is therefore higher

ber of projects involving technology transfer to overseas plants, and

than it would be for a firm that already knows that basic informa-

also found technology-transfer costs averaging about 20 percent of

tion. (Recall that the stickiness of a unit of information is defined

total project costs. Winter and Suzlanski (2001) explored replica-

as the incremental expenditure required to transfer a unit of infor-

tion of well-known organizational routines at new sites and found

mation to a specified site in a form usable by a specific information

the process difficult and costly.

seeker.)

Total information stickiness associated with solving a specific prob-

258

256

Why is information transfer so costly? The term “stickiness” refers

only to a consequence, not to a cause. Information stickiness can

lem is also determined by the amount of information required by a

result from causes ranging from attributes of the information itself to

problem solver. Sometimes a great deal is required, for two rea-

access fees charged by an information owner. Consider tacitness-

sons. First, as Rosenberg (1976, 1982) and Nelson (1982, 1990)

--a lack of explicit encoding. Polanyi (1958, pp. 49--53) noted that

point out, much technological knowledge deals with the specific

many human skills are tacit because “the aim of a skilful perfor-

and the particular. Second, one does not know in advance of prob-

mance is achieved by the observance of a set of rules which are

lem solving which particular items will be important.

not known as such to the person following them.” For example,

An example from a study by von Hippel and Tyre (1995) illustrates

259

swimmers are probably not aware of the rules they employ to keep

both points nicely. Tyre and I studied how and why novel produc-

afloat (e.g., in exhaling, they do not completely empty their lungs),

tion machines failed when they were first introduced into factory

nor are medical experts generally aware of the rules they follow in

use. One of the machines studied was an automated machine

order to reach a diagnosis of a disease. “Indeed,” Polanyi says,

used by a computer manufacturing firm to place large integrated

“even in modern industries the indefinable knowledge is still an es-

circuits onto computer circuit boards. The user firm had asked an

sential part of technology.” Information that is tacit is also sticky be-

outside group to develop what was needed, and that group had

cause it cannot be transferred at low cost. As Polanyi points out,

developed and delivered a robot arm coupled to a machine-vision

“an art which cannot be specified in detail cannot be transmitted by

system. The arm, guided by the vision system, was designed to

prescription, since no prescription for it exists. It can be passed on

pick up integrated circuits and place them on a circuit board at pre-

only by example from master to apprentice. . . .” Apprenticeship is

cise locations.

a relatively costly mode of transfer.

Upon being installed in the factory, the new component-placing

260

257

Another cause of information stickiness is related to absorptive ca-

machine failed many times as a result of its developers' lack of

SiSU

web.mit.edu/evhippel/www/

44

Democratizing Innovation

some bit of information about the need or use environment. For

The important point is that this omission was not due to poor prac-

example, one day machine operators reported that the machine

tice; it was due to the huge amount of information about the need

was malfunctioning---again---and they did not know why. Investi-

and the use environment that was potentially relevant to problem

gation traced the problem to the machine-vision system. This sys-

solvers. Note that the use environment and the novel machine con-

tem used a small TV camera to locate specific metalized patterns

tain many highly specific attributes that could potentially interact

on the surface of each circuit board being processed. To func-

to cause field problems. Note also that the property of the board

tion, the system needed to “see” these metalized patterns clearly

causing this particular type of failure was very narrow and specific.

against the background color of the board's surface. The vision

That is, the problem was not that the board had physical proper-

system developed by the machine-development group had func-

ties, nor that it had a color. The problem was precisely that some

tioned properly in their lab when tested with sample boards from

boards were yellow, and a particular shade of yellow at that. Since

the user factory. However, the field investigation showed that in

a circuit board, like most other components, has many attributes in

the factory it failed when boards that were light yellow in color were

addition to color (shape, size, weight, chemical composition, reso-

being processed.

nant frequency, dielectric constant, flexibility, and so on), it is likely

that problem solvers seeking to learn everything they might need

261

The fact that some of the boards being processed were sometimes

to know about the use and the use environment would have to col-

light yellow was a surprise to the machine developers. The factory

lect a very large (perhaps unfeasibly large) number of very specific

personnel who had set the specifications for the machine knew that

items of information.

the boards they processed varied in color; however, they had not

volunteered the information, because they did not know that the de-

Next, consider that the information items the problem solver will ac-

263

velopers would be interested. Early in the machine-development

tually need (of the many that exist) are contingent on the solution

process, they had simply provided samples of boards used in the

path taken by the engineer designing the product. In the example,

factory to the machine-development group. And, as it happened,

the problem caused by the yellow color of the circuit board was con-

these samples were green. On the basis of the samples, devel-

tingent on the design solution to the component-placing problem

opers had then (implicitly) assumed that all boards processed in

selected by the engineer during the development process. That is,

the field were green. It had not occurred to them to ask users

the color of the circuit boards in the user factory became an item

“How much variation in board color do you generally experience?”

the problem solvers needed to know only when engineers, in the

Thus, they had designed the vision system to work successfully

course of their development of the component placer, decided to

with boards that were green.

use a vision system in the component-placing machine they were

designing, and the fact that the boards were yellow became rel-

262

In the case of this field failure, the item of information needed to

evant only when the engineers chose a video camera and light-

understand or predict this problem was known to the users and

ing that could not distinguish the metalized patterns on the board

could easily have been provided to the machine developers---had

against a yellow background. Clearly, it can be costly to transfer the

the developers thought to ask and/or had users thought to volun-

many items of information that a product or service developer might

teer it. But in the actual evolution of events this was not done.

require---even if each individual item has low stickiness---from one

SiSU

web.mit.edu/evhippel/www/

45

Democratizing Innovation

site to another.

their technical expertise to improve the products along known di-

mensions of merit, such as accuracy.

264

How Information Asymmetries Affect User Innovation vs.

Manufacturer Innovation

Table 5.1 Users tend to develop innovations that deliver novel

268

functions.

265

An important consequence of information stickiness is that it results

in information asymmetries that cannot be erased easily or cheaply.

269

Different users and manufacturers will have different stocks of infor-

Type of improvement provided by innovation

User

Manufacturer

n

mation, and may find it costly to acquire information they need but

New functional capability

82%

18%

17

do not have. As a result, each innovator will tend to develop inno-

Sensitivity, resolution, or accuracy improvement

48%

52%

23

vations that draw on the sticky information it already has, because

Convenience or reliability improvement

13%

87%

24

that is the cheapest course of action (Arora and Gambardella 1994;

Total sample size

64

von Hippel 1994). In the specific case of product development, this

means that users as a class will tend to develop innovations that

draw heavily on their own information about need and context of

use. Similarly, manufacturers as a class will tend to develop in-

Source: Riggs and von Hippel 1994, table 3.

270

novations that draw heavily on the types of solution information in

The variation in locus of innovation for different types of innova-

271

which they specialize.

tions, seen in table 5.1 does fit our expectations from the point of

view of sticky information considerations. But these findings are

266

This effect is visible in studies of innovation. Riggs and von Hippel

(1994) studied the types of innovations made by users and manu-

not controlled for profitability, and so it might be that profits for new

facturers that improved the functioning of two major types of scien-

functional capabilities are systematically smaller than profits ob-

tific instruments.

tainable from improvements made to existing functionality. If so,

this could also explain the patterns seen.

267

They found that users tended to develop innovations that enabled

the instruments to do qualitatively new types of things for the first

Ogawa (1998) took the next necessary step and conducted

272

time. In contrast, manufacturers tended to develop innovations that

an empirical study that did control for profitability of innovation

enabled users to do the same things they had been doing, but to do

opportunities.

He too found the sticky-information effect---this

them more conveniently or reliably (table 5.1). For example, users

time visible in the division of labor within product-development

were the first to modify the instruments to enable them to image

projects. He studied patterns in the development of a sample of

and analyze magnetic domains at sub-microscopic dimensions. In

24 inventory-management innovations. All were jointly developed

contrast, manufacturers were the first to computerize instrument

by a Japanese equipment manufacturer, NEC, and by a user firm,

adjustments to improve ease of operation. Sensitivity, resolution,

Seven-Eleven Japan (SEJ). SEJ, the leading convenience-store

and accuracy improvements fall somewhere in the middle, as the

company in Japan, is known for its inventory management. Using

data show. These types of improvements can be driven by users

innovative methods and equipment, it is able to turn over its

seeking to do specific new things, or by manufacturers applying

inventory as many as 30 times a year, versus 12 times a year for

SiSU

web.mit.edu/evhippel/www/

46

Democratizing Innovation

competitors (Kotabe 1995). An example of such an innovation

rough terrain such as mountain trails. It may also involve various

jointly developed by SEJ and NEC is just-in-time reordering,

other extreme conditions, such as bicycling on snow and ice and

for which SEJ created the procedures and NEC the hand-held

in the dark (van der Plas and Kelly 1998).

equipment to aid store clerks in carrying out their newly designed

Mountain biking began in the early 1970s when some young cy-

tasks. Equipment sales to SEJ are important to NEC: SEJ has

276

clists started to use their bicycles off-road. Existing commercial

thousands of stores in Japan.

bikes were not suited to this type of rough use, so early users put to-

273

The 24 innovations studied by Ogawa varied in the amount of sticky

gether their own bikes. They used strong bike frames, balloon tires,

need information each required from users (having to do with store

and powerful drum brakes designed for motorcycles. They called

inventory- management practices) and the amount of sticky solu-

their creations “clunkers” (Penning 1998; Buenstorf 2002).

tion information required from manufacturers (having to do with

Commercial manufacture of mountain bikes began about 1975,

new equipment technologies). Each also varied in terms of the

277

when some of the early users of mountain bikes began to also build

profit expectations of both user and manufacturer. Ogawa deter-

bikes for others. A tiny cottage industry developed, and by 1976

mined how much of the design for each was done by the user firm

a half-dozen small assemblers existed in Marin County, California.

and how much by the manufacturer firm. Controlling for profit ex-

In 1982, a small firm named Specialized, an importer of bikes and

pectations, he found that increases in the stickiness of user infor-

bike parts that supplied parts to the Marin County mountain bike as-

mation were associated with a significant increase in the amount

semblers, took the next step and brought the first mass-produced

of need-related design undertaken by the user (Kendall correla-

mountain bike to market. Major bike manufacturers then followed

tion coefficient = 0.5784, P < 0.01). Conversely he found that

and started to produce mountain bikes and sell them at regular bike

increased stickiness of technology-related information was asso-

shops across the United States. By the mid 1980s the mountain

ciated in a significant reduction in the amount of technology de-

bike was fully integrated in the mainstream bike market, and it has

sign done by the user (Kendall correlation coefficients = 0.4789,

since grown to significant size. In 2000, about $58 billion (65 per-

P < 0.05). In other words, need-intensive tasks within product-

cent) of total retail sales in the US bicycle market were generated in

development projects will tend to be done by users, while solution-

the mountain bike category (National Sporting Goods Association

intensive ones will tend to be done by manufacturers.

2002).

274

Low-Cost Innovation Niches

Mountain biking enthusiasts did not stop their innovation activi-

278

275

Just as there are information asymmetries between users and

ties after the introduction of commercially manufactured mountain

manufacturers as classes, there are also information asymmetries

bikes. They kept pushing mountain biking into more extreme en-

among individual user firms and individuals, and among individual

vironmental conditions, and they continued to develop new sports

manufacturers as well.

A study of mountain biking by Lüthje,

techniques involving mountain bikes ( Mountain Bike 1996). Thus,

Herstatt, and von Hippel (2002) shows that information held

some began jumping their bikes from house roofs and water tow-

locally by individual user-innovators strongly affects the type of

ers and developing other forms of acrobatics. As they did so,

innovations they develop. Mountain biking involves bicycling on

they steadily discovered needs for improvements to their equip-

SiSU

web.mit.edu/evhippel/www/

47

Democratizing Innovation

ment. Many responded by developing and building the improve-

.

Mean

Median

Very high or

high

agree-

ments they needed for themselves.

ment

"I had it due to my professional background."

4.22

4

47.5%

279

Our sample of mountain bikers came from the area that bikers call

the North Shore of the Americas, ranging from British Columbia to

"I had it from mountain biking or another hobby."

4.56

5

52.4%

Washington State. Expert mountain bikers told us that this was a

"I learned it to develop this idea."

2.11

2

16%

current “hot spot” where new riding styles were being developed

and where the sport was being pushed toward new limits. We

used a questionnaire to collect data from members of North Shore

Source: Lüthje et al. 2003. N = 61. Responses were rated on a

283

mountain biking clubs and from contributors to the mailing lists of

seven-point scale, with 1 = not at all true and 7 = very true.

two North Shore online mountain biking forums. Information was

obtained from 291 mountain bikers. Nineteen percent of bikers re-

Discussion

284

sponding to the questionnaire reported developing and building a

To the extent that users have heterogeneous and sticky need and

new or modified item of mountain biking equipment for their own

285

solution information, they will have heterogeneous low-cost inno-

use. The innovations users developed were appropriate to the

vation niches. Users can be sophisticated developers within those

needs associated with their own riding specialties and were het-

niches, despite their reliance on their own need information and so-

erogeneous in function.

lution information that they already have in stock. On the need side,

280

We asked mountain bikers who had innovated about the sources

recall that user-innovators generally are lead users and generally

of the need and solution information they had used in their problem

are expert in the field or activity giving rise to their needs. With re-

solving. In 84.5 percent of the cases respondents strongly agreed

spect to solution information, user firms have specialties that may

with the statement that their need information came from personal

be at a world-class level. Individual users can also have high lev-

needs they had frequently experienced rather than from informa-

els of solution expertise. After all, they are students or employees

tion about the needs of others. With respect to solution information,

during the day, with training and jobs ranging from aerospace en-

most strongly agreed with the statement that they used solution in-

gineering to orthopedic surgery. Thus, mountain bikers might not

formation they already had, rather than learning new solution infor-

want to learn orthopedic surgery to improve their biking equipment,

mation in order to develop their biking equipment innovation (table

but if they already are expert in that field they could easily draw on

5.2).

what they know for relevant solution information. Consider the fol-

lowing example drawn from the study of mountain biking discussed

281

Table 5.2 Innovators tended to use solution information they al-

ready had “in stock” to develop their ideas. Tabulated here are

earlier:

innovators' answers to the question “How did you obtain the infor-

I'm a human movement scientist working in ergonomics and biome-

286

mation needed to develop your solution?”

chanics. I used my medical experience for my design. I calcu-

282

lated a frame design suitable for different riding conditions (down-

hill, climb). I did a CAD frame design on Catia and conceived a

SiSU

web.mit.edu/evhippel/www/

48

Democratizing Innovation

spring or air coil that can be set to two different heights. I plan to

priate to developing and testing possible solutions to those new

build the bike next year.

problems.

Of course, these same considerations apply to user firms as well as

289

287

Users' low-cost innovation niches can be narrow because their de-

to individual users. A firm that is in the business of polishing mar-

velopment “labs” for such experimentation often consist largely of

ble floors is a user of marble polishing equipment and techniques.

their individual use environment and customary activities. Con-

It will have a low-cost learning laboratory with respect to improve-

sider, for example, the low-cost innovation niches of individual

ments in these because it can conduct trial-and-error learning in

mountain bikers. Serious mountain bikers generally specialize in

that “lab” during the course of its customary business activities.

a particular type of mountain biking activity. Repeated specialized

Innovation costs can be very low because innovation activities are

play and practice leads to improvement in related specialized skills.

paid for in part by rewards unrelated to the novel equipment or tech-

This, in turn, may lead to a discovery of a problem in existing moun-

nique being developed. The firm is polishing while innovating---and

tain biking equipment and a responsive innovation. Thus, an in-

is getting paid for that work (Foray 2004). The low cost innovation

novating user in our mountain biking study reported the following:

niche of the marble polishing firm may be narrow. For example, it is

“When doing tricks that require me to take my feet off the bike ped-

unlikely to have any special advantage with respect to innovations

als in mid-air, the pedals often spin, making it hard to put my feet

in the polishing of wood floors, which requires different equipment

back onto them accurately before landing.” Such a problem is en-

and techniques.

countered only when a user has gained a high level of skill in the

very specific specialty of jumping and performing tricks in mid-air.

Once the problem has been encountered and recognized, how-

ever, the skilled specialist user can re-evoke the same problematic

conditions at will during ordinary practice. The result is the cre-

ation of a low-cost laboratory for testing and comparing different

solutions to that problem. The user is benefiting from enjoyment of

his chosen activity and is developing something new via learning

by doing at the same time.

288

In sharp contrast, if that same user decides to stray outside his

chosen activity in order to develop innovations of interest to oth-

ers with needs that are different from his own, the cost properly

assignable to innovation will rise. To gain an equivalent-quality con-

text for innovation, such a user must invest in developing personal

skill related to the new innovation topic. Only in this way will he

gain an equivalently deep understanding of the problems relevant

to practitioners of that skill, and acquire a “field laboratory” appro-

SiSU

web.mit.edu/evhippel/www/