judged most likely to develop important improvements.
These
sometimes associated with “unauthorized user tinkering.” For
machines are supplied with restrictive interlocks removed so that
example, an automaker might legitimately worry about the user-
the users can easily modify them. In exchange for this research
programmed engine controller chips that racing aficionados and
support, the medical researchers give GE preferred access to
others often install to change their cars' performance. The result
innovations they develop. Over the years, supported researchers
can be findings of eventual commercial value as users explore
have provided a steady flow of significant improvements that have
new performance regimes that manufacturers' engineers might
been first commercialized by GE. Managers consider the policy a
not have considered. However, if users choose to override manu-
major source of GE's commercial success in the MRI field.
facturers' programming to increase engine performance, there is
also a clear risk of increased warrantee costs for manufacturers if
484
Providing Complementary Products or Services
engines fail as a consequence (Mollick 2004).
We have seen that manufacturers can often find ways to profit from
488
485
Many user innovations require or benefit from complementary
user innovation. It is also the case, however, that user innovators
products or services, and manufacturers can often supply these
and user innovation communities can provide many of these same
at a profit. For example, IBM profits from user innovation in open
functions for themselves. For example, StataCorp is successfully
source software by selling the complement of computer hardware.
selling a proprietary statistical software package. User-developed
Specifically, it sells computer servers with open source software
alternatives exist on the web that are developed and maintained
pre-installed, and as the popularity of that software goes up, so
by user-innovators and can be downloaded at no charge. Which
do server sales and profits. A firm named Red Hat distributes a
ownership model will prove more robust under what circumstances
version of the open source software computer operating system
remains to be seen. Ultimately, since users are the customers, they
Linux, and also sells the complementary service of Linux technical
get to choose.
support to users. Opportunities to provide profitable complements
are not necessarily obvious at first glance, and providers often
reap benefits without being aware of the user innovation for which
they are providing a complement.
Hospital emergency rooms,
for example, certainly gain considerable business from providing
medical care to the users and user-developers of physically
83
Democratizing Innovation
489
10 Application: Searching for Lead User Innovations
users sought are at the leading edge of “advanced analog” fields
or at the leading edge of target markets. Searching for the former
490
Users and manufacturers can apply the insights developed in this
is more difficult, but experience shows that the user-developed in-
book to improve their innovation processes. In this chapter, I illus-
novations that are most radical (and profitable) relative to conven-
trate by showing how firms can profit by systematically searching
tional thinking often come from lead users in “advanced analog”
for innovations developed by lead users. I first explain how this
fields.
can be done. I then present findings of a study conducted at 3M to
assess the effectiveness of lead user idea-generation techniques.
Finally, I briefly review other studies reporting systematic searches
for lead users by manufacturers, and the results obtained.
491
Searching for Lead Users
492
Product-development processes traditionally used by manufactur-
ers start with market researchers who study customers in their tar-
get markets to learn about unsatisfied needs. Next, the need infor-
mation they uncover is transferred to in-house product developers
who are charged with developing a responsive product. In other
words, the approach is to find a user need and to fill it by means of
495
in-house product development.
Figure 10.1 Innovations by lead users precede equivalent com-
496
493
These traditional processes cannot easily be adapted to systematic
mercial products.
searching for lead user innovations. The focus on target-market
Identifying Lead Users in Advanced Analog Fields
497
customers means that lead users are regarded as outliers of no
interest. Also, traditional market-research analyses focus on col-
Lead users in advanced analog fields experience needs that are
498
lecting and analyzing need information and not on possible solu-
related to but more extreme than those being faced by any users,
tions that users may have developed. For example, if a user says “I
including lead users, within the target market. They also often face
have developed this new product to make task X more convenient,”
a different set of constraints than those affecting users in the target
market-research analyses typically will note that more convenience
market. These differences can force them to develop solutions that
is wanted but not record the user-developed solution. After all,
are entirely new from the perspective of the target market.
product development is the province of in-house engineers!
As an example, consider the relationship between the braking re-
499
494
We are therefore left with a question: How can manufacturers build
quirements faced by users of automobiles (let's call auto users the
a product-development process that systematically searches for
target market) and the braking requirements faced by large com-
and evaluates lead user-generated innovations? (See figure 10.1.)
mercial airplanes as they land on an airport runway (the advanced
It turns out that the answer differs depending on whether the lead
analog market). Clearly, the braking demands on large airplanes
84
Democratizing Innovation
are much more extreme. Airplanes are much heavier than autos
for a long time, and often have searched beyond the target market
and land at higher speeds: their brakes must rapidly dissipate hun-
for information.
dreds of times more energy to bring the vehicle to a stop. Also, the
Networking from innovators to more advanced innovators in this
situational constraints are different. For example, auto drivers are
502
way is called pyramiding (von Hippel, Thomke, and Sonnack 1999).
often assisted in braking in winter by the application of salt or sand
Pyramiding is a modified version of the “snowballing” technique
to icy roads. These aids cannot be applied in the case of aircraft:
sometimes used by sociologists to identify members of a group
salt would damage aircraft bodies, and sand would be inhaled into
or accumulate samples of rare respondents (Bijker 1995). Snow-
jet engines and damage them.
balling relies on the fact that people with rare interests or attributes
tend to know others like themselves. Pyramiding modifies this idea
500
The result of the more extreme demands and additional constraints
placed on solutions to aircraft braking was the development of an-
by assuming that people with a strong interest in a topic or field can
tilock braking systems (ABS) for aircraft. Auto firms conducting
direct an enquiring researcher to people more expert than them-
searches for valuable lead user innovations regarding auto brak-
selves. Experiments have shown that pyramiding can identify high-
ing were able to learn about this out-of-field innovation and adapt
quality informants much more efficiently than can mass-screening
if for use in autos---where it is common today. Before the devel-
techniques under many conditions (von Hippel, Franke, and Prügl
opment of ABS for autos, an automobile firm could have learned
2005). Pyramiding was made into a practical industrial process by
about the underlying concept by studying the practices of users with
Mary Sonnack, a Division Scientist at 3M, and Joan Churchill, a
a strong need for controlling skidding while braking such as stock
psychologist specializing in the development of industrial training
car auto racing teams. These lead users had learned to manually
programs.
“pump” their brakes to help control this problem. However, auto
Identifying Lead Users in Target Markets
503
company engineers were able to learn much more by studying the
automated solutions developed in the “advanced analog” field of
In general it is easier to identify users at the leading edge of tar-
504
aerospace.16
get markets than it is to identify users in advanced analog fields.
Screening for users with lead user characteristics can be used.
501
Finding lead users in advanced analog markets can be difficult be-
When the desired type of lead user is so rare as to make screening
cause discovering the relevance of a particular analog can itself be
impractical---often the case---pyramiding can be applied. In ad-
a creative act. One approach that has proven effective is to ask the
dition, manufacturers can take advantage of the fact that users
more easily identified lead users in target markets for nominations.
at the leading edge of a target market often congregate at spe-
These lead users tend to know about useful advanced analogs, be-
cialized sites or events that manufacturers can readily identify. At
cause they have been struggling with their leading-edge problems
such sites, users may freely reveal what they have done and may
learn from others about how to improve their own practices still fur-
16ABS braking is intended to keep a vehicle's wheels turning during braking.
ther. Manufacturers interested in learning from these lead users
ABS works by automatically and rapidly “pumping” the brakes. The result is that
the wheels continue to revolve rather than “locking up,” and the operator
can easily visit the sites and listen in. For example, sports equip-
continues to have control over steering.
ment companies can go to sporting meets where lead users are
85
Democratizing Innovation
known to compete, observe user innovations in action, and com-
Methods
509
pare notes.
3M first began using the lead user method in one division in 1996.
510
505
Essentially the same thing can be done at virtual sites. For ex-
By May 2000, when data collection began, five divisions of 3M had
ample, recall the practices of StataCorp, a supplier of statistical
completed seven lead user (LU) idea-generation projects and had
software. Stata sells a set of standard statistical tests and also a
funded further development of the product concepts generated by
language and tools that statisticians can use to design new tests to
five of these. These same five divisions also had 42 contempo-
serve their own evolving needs. Some Stata users (statisticians)
raneously funded projects that used “find a need and fill it” idea-
took the initiative to set up a few specialized websites, unaffili-
generation methodologies that were traditional practice at 3M. We
ated with StataCorp, where they post their innovations for others
used these two samples of funded ideas to compare the perfor-
to download, use, comment on, and improve. StataCorp person-
mance of lead user idea-generation projects with traditional idea-
nel visit these sites, learn about the user innovations, and observe
generation projects. Although 3M cooperated in the study and per-
which tests seem to be of interest to many users. They then de-
mitted access to company records and to members of the product-
velop proprietary versions of the more generally useful tests as
development teams, the firm did not offer a controlled experimen-
commercial products.
tal setting. Rather, we as researchers were required to account for
any naturally occurring differences after the fact.
506
When specialized rendezvous sites for lead users don't exist in a
particular field, manufacturers may be able to create them. Techni-
Our study methodology required a pre-post/test-control situation,
511
con Corporation, for example, set up a series of seminars at which
with at least quasi-random assignments to treatment cells (Cook
innovating users of their medical equipment got together and ex-
and Campbell 1979). In other words, our goal was to compare
changed information on their innovations. Technicon engineers
samples of development projects in 3M divisions that differed with
were free to listen in, and the innovations developed by these users
respect to their use of lead user idea-generation methods, but
were the sources of most of Technicon's important new product im-
that were as similar as possible in other respects. Identifying,
provements (von Hippel and Finkelstein 1979).
understanding, and controlling for the many potential sources of
difference that could affect the natural experiment involved careful
507
The 3M Experiment
field explorations.
Thus, possible differences between project
508
To test whether lead users in advanced analog fields can in fact
staffing and performance incentives applied to LU and non-LU
generate information that leads to commercially valuable new prod-
idea-generation projects were assessed. We looked for (and did
ucts, Lilien, Morrison, Searls, Sonnack, and von Hippel (2002)
not find) differences in the capabilities or motivation of LU and
studied a natural experiment at 3M. That firm was carrying out
non-LU project team members with respect to achieving a major
both lead user projects and traditional market research-based idea-
new product advance. 3M managers also said that there was
generation projects in the same divisions at the same time, and
no difference in these matters, and a content analysis of formal
in sufficient numbers to make statistical comparisons of outcomes
annual performance goals set for the individual LU and non-LU
possible.
team members in a division that allowed access to these data
86
Democratizing Innovation
from project to project. Market data collected by outside organiza-
tions were sometimes used, as were data from focus groups with
major customers and from customer panels, and information from
lab personnel. Non-LU teams collected market information from
target markets users but not from lead users.
Findings
515
Our research compared all funded product concepts generated by
516
LU and non-LU methods from February 1999 to May 2000 in each
of the five 3M divisions that had funded one or more lead-user-
developed product concepts. During that time, five ideas gen-
erated by lead user projects were being funded, along with 42
ideas generated by non-LU idea-generation methods. The results
of these comparisons can be seen in table 10.1. Product con-
cepts generated by seeking out and learning from lead users were
found to be significantly more novel than those generated by non-
LU methods. They were also found to address more original or
newer customer needs, to have significantly higher market share,
to have greater potential to develop into an entire product line,
and to be more strategically important. The lead-user-developed
product concepts also had projected annual sales in year 5 that
were greater than those of ideas generated by non-LU methods
by a factor of 8---an average of $146 million versus an average
of $18 million in forecast annual sales. Thus, at 3M, lead user
idea-generation projects clearly did generate new product concepts
with much greater commercial potential than did traditional, non-LU
methods (p < 0.005).
Table 10.1 Concepts for new products developed by lead user
517
project teams had far more commercial promise than those devel-
oped by non-lead-user project teams.
518
LU product con-
Non-LU
product
Significance
cepts (n =5)
concepts (n = 42)
87
Democratizing Innovation
supported their views.
LU product con-
Non-LU
product
Significance
cepts (n =5)
concepts (n = 42)
Factors related to value of concept
512
We also found no major differences in the innovation opportuni-
ties teams faced. They also looked for Hawthorne or placebo ef-
Novelty compared with competition a
9.6
6.8
0.01
fects that might affect the project teams differentially, and found
Originality/newness of customer needs
8.3
5.3
0.09
addressed
none. (The Hawthorne effect can be described as “I do better be-
% market share in year 5
68%
33%
0.01
cause extra attention is being paid to me or to my performance.”
Estimated sales in year 5 (deflated for
146 m
18 m
0.00
The placebo effect can be described as “I expect this process will
forecast error)
work and will strive to get the results I have been told are likely.”)
Potential for entire product family a
10.0
7.5
0.03
We concluded that the 3M samples of funded LU and non-LU idea-
Operating profit
22%
24.0%
0.70
generation projects, though not satisfying the random assignment
Probability of success
80%
66%
0.24
criterion for experimental design, appeared to satisfy rough equiva-
Strategic importance a
9.6
7.3
0.08
lence criteria in test and control conditions associated with natural
Intellectual property protection a
7.1
6.7
0.80
or quasi-experimentation. Data were collected by interviews and
Factors related to organizational fit of
concept
by survey instruments.
Fit with existing distribution channels a
8.8
8.0
0.61
513
With respect to the intended difference under study---the use of
Fit with existing manufacturing capabili-
7.8
6.7
0.92
ties a
lead user methods within projects---all lead user teams employed
Fit with existing strategic plan a
9.8
8.4
0.24
an identical lead user process taught to them with identical coach-
ing materials and with coaching provided by members of the same
small set of internal 3M coaches. Each lead user team consisted of
Source: Lilien et al. 2002, table 1.
519
three or four members of the marketing and technical departments
a. Rated on a scale from 1 to 10.
of the 3M division conducting the project. Teams began by identi-
Note that the sales data for both the LU and non-LU projects are
fying important market trends. Then, they engaged in pyramiding
520
forecasts. To what extent can we rely on these? We explored this
to identify lead users with respect to each trend both within the tar-
matter by collecting both forecast and actual sales data from five
get market and in advanced analog markets. Information from a
3M division controllers. (Division controllers are responsible for au-
number of innovating lead users was then combined by the team
thorizing new product-development investment expenditures.) We
to create a new product concept and business plan---an “LU idea”
also obtained data from a 1995 internal study that compared 3M's
(von Hippel, Thomke, and Sonnack 1999).
sales forecasts with actual sales. We combined this information to
514
Non-lead-user idea-generation projects were conducted in accor-
develop a distribution of forecast errors for a number of 3M divi-
dance with traditional 3M practices. I refer to these as non-LU idea
sions, as well as overall forecast errors across the entire corpora-
generation methods and to teams using them as non-LU teams.
tion. Those errors range from forecast/actual of +30 percent (over-
Non-LU teams were similar to lead user teams in terms of size and
forecast) to --13 percent (underforecast). On the basis of the infor-
make-up. They used data sources for idea generation that varied
mation just described, and in consultation with 3M management,
88
Democratizing Innovation
we deflated all sales forecast data by 25 percent. That deflator is
524
consistent with 3M's historical experience and, we think, provides
Incremental product improve-
Major new product lines
conservative sales forecasts.17 Deflated data appear in table 10.1
ments
and in the following tables.
LU method
0
5
521
Rather strikingly, all five of the funded 3M lead user projects cre-
Non-LU method
41
1
ated the basis for major new product lines for 3M (table 10.2). In
contrast, 41 of 42 funded product concepts generated by non-LU
methods were improvements or extensions of existing product lines
Source: Lilien et al. 2002, table 2.
525
(χ2 test, p < 0.005).
To illustrate what the major product line innovations that the LU
526