THE TOYOTA PRODUCTION SYSTEM AT NUMMI
PAUL S. ADLER
University of Southern California
School of Business Administration
Los Angeles, CA 90089-1421
Draft chapter for Steve Babson (ed.), Globa Labor and Lean Production, Wayne
State University Press, forthcoming.
The human aspects of Japanese manufacturing management techniques,
in particular those embodied in the Toyota production system, are currently
under intense debate. Do they represent a good model for the future of the
American workplace? Some observers applaud the approach’s reliance on
teamwork, workers’ problem-solving and multi-skilling. Other observers
denounce what they see as work intensification, management by stress, and
The present chapter contributes to this debate through a discussion of
New United Motors Manufacturing, Inc. (NUMMI). It draw on several prior
publications (Adler, 1993a, 1993b; Adler and Cole, 1993). NUMMI, a joint
venture between General Motors and Toyota, has implemented the Toyota
approach with but little dilution. The plant is unusual in that its work force is
composed primarily of workers who had been laid off when the GM-Fremont
plant closed in 1982. Moreover, the UAW has retained considerable influence.
The first section summarizes some historical background. I then describe
some of the key policies governing NUMMI’s production system and its labor
relations. The next section summarizes some data on NUMMI’s exceptional
productivity and quality performance. I then characterize workers’
assessments of the quality of work at NUMMI. Some puzzling paradoxes
emerge from this characterization, which I attempt to resolve by suggesting
that NUMMI’s work organization can be understood as an instance of an
unusual but potentially important model of organization that I call “democratic
NUMMI was formed in 1983 as a joint venture between General Motors and
Toyota. For Toyota, setting up plants in the U.S. seemed like a good way to
alleviate trade pressures. NUMMI was conceived as the first step in that strategy,
designed to help Toyota learn about U.S. suppliers and labor. For its part, General
Motors wanted to learn about Japanese manufacturing systems, and they needed a
small car to fill a hole in their product line. Toyota contributed $100 million, and
was responsible for setting up the plant’s day-to-day operations, and provided the
design of the plant main vehicle, the Nova, a variant of the Corolla. (NUMMI later
produced Corollas, Geo Prizms — a variant of Toyota’s Sprinter — and Toyota
compact pick-up trucks). GM contributed the facility and was responsible for
marketing the new vehicle.
The decision was made to house the new company at the old GM-Fremont
plant, which had been shut down in 1982 leaving over 3000 workers on layoff.
Politically, it was inconceivable that the plant reopen without UAW involvement.
A Letter of Intent was thus signed with the UAW in September 1983. The Letter
stipulated that the UAW would be recognized as the bargaining agent for the
venture’s employees; that the company would pay prevailing U.S. auto industry
wages and benefits; and that a majority of the work force would be hired from
among the workers laid off from GM-Fremont (but seniority would not be a
factor). In return, the UAW agreed to support the implementation of a new
production system and to negotiate a new contract. A new collective bargaining
agreement was signed in June 1985.
NUMMI began hiring in May 1984. Of 5300 applications sent to former GM-
Fremont employees, 3200 were returned. Over the next 20 months, 2,200 hourly
team members were hired, approximately 85% of them from the old GM-Fremont
plant, including the entire union hierarchy. Some 300 salaried employees were
also hired. The applicants for hourly jobs — Team Leaders and Team Members —
were evaluated jointly by managers and union officials.
NUMMI’s overall strategy was to build its cars at the lowest cost and the
highest quality. A number of management policies contributed to that goal. The
following two sections highlight two such sets of policies — the policies that
constituted the production system and those that most directly affected labor
THE PRODUCTION SYSTEM
NUMMI’s operations management was modelled directly on the Toyota
production system. The key elements were the following:
• Kanban: NUMMI did not use a computerized scheduling system. Instead, signs
— “kanban” — would be passed to the upstream department whenever inventory
pallets or dollies needed to be replaced. When no kanban arrived, the upstream
department stopped production because no inventory was allowed to build up.
• Production levelling: In Big 3 plants, production schedules are constantly
changing. So they fix the line speed, and adjust output by adding or subtracting
overtime and taking on and laying off shifts. By contrast, NUMMI levelled the
schedule over several months, and made periodic adjustments to output levels by
varying the line speed. The levelling policy also shaped the day’s build schedule:
different models were mixed evenly throughout, rather than batched.
• Kaizen: NUMMI paid enormous attention to kaizen — the continuous
improvement of all aspects of production. All NUMMI workers were given
training in problem-solving for kaizen. Workers participation in the suggestion
program was seen by management as a key measure of the plant’s performance.
• Visual control: This policy was designed to signal abnormal conditions as
rapidly and automatically as possible. Kanban was one form of visual control,
signalling the need to replenish an inventory pallet. Another key element of visual
control at NUMMI was the “andon” board lights that signalled quality problems
on the line.
• Jidoka: This meant assuring quality in the production process itself rather than
allowing inspections at the end of the process to detect problems. Workers pulled
a “line stop” cord when they encountered a quality problem, triggering the andon
board signal that alerted the Team Leader and often the Group Leader. Unless the
problem could be resolved within 60 seconds, the line would stop. “Pake-yoke” or
error-proofing was another element of jidoka — parts packaging, equipment
designs, and tool setups were specified so as to make inadvertent error almost
• Team concept: The team concept at NUMMI encompassed both the cooperative
labor relations system described below and the organization of small teams of four
to six workers. Workers in each team were cross-trained on each others’ tasks and
most workers rotated between tasks. (Rotation was not practiced in Toyota’s
Japanese plants, but it was written into NUMMI’s collective bargaining
• Standardized work: This practice was, in the words of a NUMMI manager, “the
intelligent interpretation and application of Taylor's time and motion studies.”
Each job was analyzed down to its constituent gestures, and the sequence of
gestures was refined and optimized for maximum performance. Every task was
planned in great detail, and person performed that task identically.
The combination of these policies created an extraordinarily disciplined
LABOR RELATIONS POLICIES
NUMMI’s labor relations policies reflected a commitment to what former
NUMMI President and CEO Kan Higashi calls the “team concept”:
The team concept is not just the small groups on the shop floor. It
also applies to the plant as a whole. The bigger team is the workers,
managers, engineers and staff all working together to constantly
improve our product. This way, the workers see that the company
isn't the property of management, but of everyone together. And the
key to this team concept is trust and respect.
This policy was reflected in NUMMI’s dress code – at least in the
production function, everyone from the senior manager down wore the same
uniform – as well as in the absence of management cafeterias and reserved
More substantively, it was reflected in a range of opportunities for worker
representatives not only to give voice to workers’ concerns but also to help shape
the company’s response to those concerns. Every week, month, and quarter,
various levels of the union leadership met with the corresponding levels of
management. The union was consulted in advance on major operating and
Management saw NUMMI’s no-layoff policy as part of its commitment to
the team concept. The collective bargaining agreement of 1985 stated:
New United Motor Manufacturing, Inc. recognizes that job security
is essential to an employee’s well being and acknowledges that it has
a responsibility, with the cooperation of the Union, to provide stable
employment to its workers. The Union’s commitments in Article II of
this Agreement are a significant step towards the realization of stable
employment. Hence, the Company agrees that it will not lay off
employees unless compelled to do so by severe economic conditions
that threaten the long term viability of the Company. The Company
will take affirmative measures before laying off any employees,
including such measures as the reduction of salaries of its officers
and management, assigning previously subcontracted work to
bargaining unit employees capable of performing this work, seeking
voluntary layoffs, and other cost saving measures.
The no-layoff policy was not without some sacrifice for workers. Not only
did the union have to agree to many changes in operating philosophy, but in
exchange for NUMMI’s commitment to avoid layoffs, the contract exempted
NUMMI from contributing the GM’s Supplemental Unemployment Benefits
scheme. The SUB scheme assured GM workers with 10 or more years of seniority
that if they were laid off, they would receive close to full pay until their retirement.
Without this Supplemental pay, laid-off workers would only get Unemployment
Insurance, whose benefits were substantially lower.
NUMMI’s “team concept” was also reflected in its organizational structure.
There was only one classification for Division 1 personnel, as opposed to over 80
in the GM-Fremont contract. The number of skilled trades classifications had also
been reduced from 18 to two – general maintenance and tool-and-die.
Team Leaders were non-exempt employees. They played a role somewhere
between the old-style “utility man” (multi-skilled worker able to fill in for a broad
range of positions) and foreman. The Team Leader filled in for absent workers;
trained new workers; assisted workers having difficulty in their jobs; recorded
attendance; assigned work when the line stopped; assisted team members in
minor maintenance and housekeeping; assessed new team members; led Kaizen
efforts; and organized social events outside the plant. In March 1988, in response
to workers’ concerns with favoritism in the selection of Team Leaders,
management agreed to a joint union/management selection process based on
explicit and objective criteria.
The productivity of the NUMMI plant has been extensively analyzed by
Krafcik (Krafcik, 1986). Some key indicators of NUMMI’s performance are
summarized in Exhibit 1.
! • hourly *
! • salaried
NUMMI PRODUCTIVITY COMPARISONS
[Source: Krafcik , 1986]
* excluding stamping, molding, and seat assembly personnel
** corrected for number of welds, welding automation, product size, relief time, and option content
Labor productivity, both corrected and uncorrected for differences in
product and technology, was much higher at NUMMI than at the old GM-Fremont
plant in 1978 and at the GM-Framingham plant. (Krafcik chose Framingham
because it was a GM plant somewhat comparable in product and technology mix
to NUMMI.) By 1986, NUMMI was almost as productive as its sister plant in
Takaoka and more productive than any other GM plant. This performance is all
the more impressive when it is recalled that the NUMMI work force was on
average some 10 years older than Takaoka’s, and younger workers are in general
better equipped to deal with the pressures of assembly line work.
The comparisons with the Takaoka plant are particularly useful because
one of the factors contributing to NUMMI’s productivity and quality performance
was the “producibility” (“manufacturability”) of their vehicles’ designs. Not only
were the designs already in production — which meant that most of the
producibility problems that the original designs may have had were already
ironed out — but Toyota was renowned for its ability to assure a high level of
producibility in its original designs. The fact that NUMMI’s overall performance
had reached a level so close to Takaoka’s suggests that this performance level was
not due exclusively to the products’ producibility.
More recent data indicate that these extraordinary results persisted into
1992. The J.D. Power and Associates Initial Quality Study of the number of
problems per 100 vehicles experienced by customers within 90 days of purchase
show that NUMMI progressed from 116 per 100 vehicles in 1989 (compared to an
industry average of 148 for all cars sold in the U.S.) to 93 in 1991, and to 83 in 1992
(versus an industry average of 125 for all cars sold in the U.S., an average of 105
for Asian nameplates, 136 for U.S. nameplates, and 158 for Europeans.)
THE QUALITY OF WORK AT NUMMI
One cannot but ask oneself if these efficiency and quality outcomes were
not been obtained at the expense of workers’ well-being. NUMMI’s extraordinary
productivity results clearly had something to do with the intensity of work.
Standard task times at GM-Fremont were set to as to occupy the experienced
worker approximately 45 seconds out of a hypothetical cycle time of 60 seconds.
NUMMI aimed to occupy the worker 60 seconds out of 60 and in practice
averaged about 57 seconds out of 60.
This work intensity was the direct result of the Toyota production system
and of the organization of work it specified: no buffer inventories, detailed work
The J.D. Power Report, June 1992. Revised data for earlier years provided by J.D.
Power and Associates.
methods closely calibrated to the real work situation, level production, rapid
response to any breakdowns, error-proofing the process. The work
organization embedded in this system was almost the mirror image of the Volvo
model that has become for many, especially in the labor movement, the yardstick
of quality of worklife. The Volvo model is best known through the innovations
introduced first at the Kalmar plant and then given a more radical extension in the
Uddevalla plant before both plants were both closed due to a collapse of Volvo’s
sales (see Berggren, 1992).
NUMMI’s operations contrasted most directly with the Volvo model in the
degree of autonomy experienced by workers. A key feature of the Volvo model
was the idea that the work team should be as autonomous as possible. Teams were
thus accorded considerable discretion in how they performed their work and in
how they scheduled their work from period to period. To achieve this autonomy,
buffers were created upstream and downstream. Teams were also given extensive
management responsibilities in such domains as scheduling overtime and
At NUMMI, the standardized work principle meant that workers had no
autonomy whatsoever in how they performed their tasks. The kanban principle
meant that there were no buffers upstream or downstream and so no autonomy in
how they scheduled their work. And NUMMI’s team concept gave teams only a
very narrow range of managerial responsibilities.
What evidence do we have, then, of workers’ reactions to working in this
intensely disciplined and regimented environment? There is no doubt that
workers at NUMMI had complaints. Assembling automobiles at NUMMI was
enormously demanding work physically. Some workers I have interviewed were
angry at what they perceived to be management’s reluctance to recognize work-
related injuries. Some Team Leaders and some management personnel were seen
as “playing favorites” or as simply incompetent. Some workers expressed
frustration because they felt they had been unfairly overlooked for special projects
or for promotions.
But how did workers overall evaluate their experience of work at NUMMI?
None of the available aggregate indicators are, in my opinion, compelling. But
they all seem to point in a common, favorable direction:
• In numerous interviews conducted between 1987 and 1993, I have yet to find
anyone who would rather work in the GM-Fremont system.
• NUMMI surveys workers’ opinions every two years. “Overall work satisfaction”
grew from 76% in 1987 to 90% in 1991. In 1987, only 70% of team members
expressed satisfaction with job security. This was primarily because sales of the
Nova were very sluggish — capacity utilization rates fell to under 60% — and
workers weren't sure they could trust management’s commitment to the contract’s
no-layoff clause. But management lived up to that commitment by putting
workers in training classes, onto facilities maintenance tasks, and into kaizen
teams, By 1991, satisfaction with job security reached 89%.
• Participation in the suggestion program grew from 26% in 1986 to 94% in 1992.
By 1992, workers were contributing an average of nearly six suggestions a year.
• Absenteeism rates held steady at an exceptionally low level of around 3%, and
turnover remained very low at less that 6%.
Why, then, do workers seem to respond rather positively to a system of
such great regimentation as the Toyota production system? I believe that at least
part of the answer lies in the feeling among workers that this regimentation was
not imposed from above but jointly decided as the best way to accomplish a
GM-Fremont had some 80 Industrial Engineers who developed work
methods and standards from handbooks in their offices. By contrast, at NUMMI,
there were no Industrial Engineers performing such duties. Team Members and
Team Leaders were taught how to analyze their own jobs using a stop watch and
assessing alternative work procedures proposed by their colleagues. Each team
would compare notes with the teams upstream and downstream to assure the best
allocation of tasks and line balancing, and then compare the result with the
analyses performed on the other shift. The resulting methods had to be applied
with great consistency, but workers were strongly encouraged to suggest
improvements for efficiency, quality or safety, and management typically followed
up on such suggestions very promptly. The resulting methods were seen as a
better way to do the job.
Let me quote one of the UAW Team Leaders:
The GM system relied on authority. People with rank, the managers,
ruled regardless of their competence or the validity of what they
were saying. It was basically a military hierarchy. At NUMMI rank
doesn't mean a damn thing. Standardized work means we all work
the objectively best way to do the job, and everyone does it that way.
I might make some minor adjustments because of my height, for
example, but I follow the procedure as laid out because it makes
sense. We're more like a special forces unit than the regular military
hierarchy. Management's delegated responsibility to the people who
do the work, and that gives workers a sense of pride in their jobs.
This does indeed sound like “the intelligent interpretation and application
of Taylor's time and motion studies.” I submit that NUMMI’s standardized work
process represents something we might call “democratic Taylorism,” in contrast to
the more traditional “despotic” form of Taylorism. The contrast can be described
along a number of dimensions:
• At NUMMI, workers actively participated in defining work methods; traditional
Taylorism assumed that the methods would be imposed by the Methods
• NUMMI’s standardized work process focused on work methods and assumed
that better time standards would emerge from the discovery of better methods;
traditional Taylorism focused on time standards, and assumed that failure to meet
them was due not to inadequate methods but to insufficient effort by the worker.
• NUMMI taught the standardized work techniques in the same program as the
kaizen process because NUMMI’s Taylorism was devoted to collective learning in
the plant; traditional Taylorism was designed to coerce work effort from a
recalcitrant work force, and the resulting methods and standards were rigidly
fixed, prisoners of the balance of power on the shopfloor.
• At NUMMI, the central role of workers in the standardized work process forced
management to share power with workers; traditional Taylorism was often a
means for asserting management’s power over the shop floor.
Precedents for such a vision of “democratic Taylorism” could perhaps be
found in the Amalgamated Clothing Workers under Sidney Hillman (Fraser, 1991,
p. 187). This version of Taylor’s doctrine of “scientific management” was
advanced by Taylorites such as Morris L. Cooke, intellectuals such as J.R.
Commons, and socialists such as the young Walter Lippmann (for unsympathetic
surveys of such interpretations, see Haber, 1964; Merkle, 1980; Hughes, 1989; also
see Jacoby 1983). It was also perhaps what Lenin (1991, p. 83) was gesturing
towards when he suggested that the Soviets should learn Taylorism and apply it
to their collective goals.
This view of Taylorism suggests that we should see Taylorism as a kind of
organizational technology, and like equipment technology, it can be designed and
implemented so as to empower or to enslave (extending the model proposed by
Walton, 1989). Management’s traditional assumption is that employees are
recalcitrant and irresponsible. As a result, they design both equipment
technologies and organizational technologies to ensure compliance, minimize
employees’ scope of discretion, and reduce their reliance on employees’ skills. And
management should then not be surprised when employees respond by apathy
and antagonism — a result which in turn comforts management in their initial
assumption that employees are recalcitrant and irresponsible.
The NUMMI example suggests that if you begin with the assumption that
workers want to contribute to the goals of that organization, you can design these
organizational technologies so that they capture, diffuse, and suggest
improvements to standard practices, and in that way, you can maintain and elicit a
high level of commitment on the part of the work force — a result which in turn
comforts this alternative initial assumption.
But how does an organization create and then sustain these initial
conditions, so workers want to contribute? What are the conditions under which
this democratic version of Taylorism is feasible? I would highlight three
First, the criteria used to select the first group of 16 GM managers at
NUMMI tell us something about the prerequisite management attitudes and
behavior. A pool of candidates was created from GM personnel files, selecting
individuals with the appropriate experience, education, work evaluations and age.
These candidates were invited to interview in Detroit. Jacobson (1986, p. 47)
quotes from two GM managers responsible for conducting these interviews their
key selection criteria: “honesty, humility, groupism [group orientation],
sensitivity, listening ability, and communication ability.”
A second pre-requisite would seem to be a commensurate set of changes in
workers’ attitudes and behavior. One of the managers I interviewed assessed this
The production people bought into standardized work very easily.
They understood the technique, because it had been done to them for
years; and they liked the idea, because now they had a chance to do
it for themselves. Their biggest problem was that many of them don't
have a lot of education, so some of the math [for statistical quality
control, for example] is a bit challenging and maybe threatening. So
you have to work on that. A second challenge comes from the fact
that you're changing things. At NUMMI, we want people to
constantly improve their standardized work and a lot of people just
aren't used to that much change. So you have to work with people so
they come to see that change not just as a disturbance but as an
opportunity to improve things. That's hard sometimes.
The third pre-requisite condition for democratic Taylorism is more
structural. George Nano, the head of the Bargaining Committee expressed his
view this way:
Standardized work gives workers the right to set up their own jobs
and that means that management has to share power and cooperate
with us. [...] The key to NUMMI's success is that management gave
up some of its power, some of its traditional prerogatives. If
managers want to motivate workers to contribute and learn, they
have to give up some of their power. If management wants workers
to trust them, we need to be 50-50 in making the decision with them.
Don't just make the decision and say 'trust me.'
If managers want the powerful apparatus of Taylorism to serve learning ends
rather than coercive ends, they will need to reconcile themselves to some real loss
Which leads to the question: why should managers relinquish power? And
if managers have no incentive to relinquish power, is not democratic Taylorism an
ephemeral epiphenomenon? There are indeed many structural and institutional
forces that should makes us sanguine about the prospects for democratic
Taylorism, such as the overall configuration of power in the U.S. today and the
peculiarities of the inherited institutional framework (of labor law in particular.)
But NUMMI’s successes incline me to modest optimism: the loss of management’s
power over workers seems to be more than compensated by the competitive
benefits of the associated increase in the organization’s power to accomplish joint
This essay has attempted to tease out of the NUMMI case a new model of
work organization. I have proposed calling the kind of work organization that we
find in this combination of “lean production” with a strong union presence
“democratic Taylorism.” I do not believe that NUMMI lives up to this model every
day in every way. Nor is this model the only useful way of describing NUMMI’s
overall characteristics. But if it helps account for NUMMI’s combination of world-
class productivity and quality and high worker morale and commitment, it is a
model that deserves our attention.
Adler, P.S., “The Learning Bureaucracy: New United Motors Manufacturing, Inc.”
In Barry M. Staw and Larry L. Cummings (eds.) Research in Organizational
Behavior, vol. 15, pp. 111-194. Greenwich, CT: JAI Press, 1993a.
Adler, P.S., “Time-and-Motion Regained.” Harvard Business Review, Jan-Feb
1993b, pp. 97-108.
Adler, P.S. and R. Cole, “Designed for Learning: A Tale of Two Auto Plants.”
Sloan Management Review, Fall 1993, vol. 34, no. 3, pp. 85-94.
Berggren, C., Alternatives to Lean Production: Work Organization in the Swedish
Auto Industry, Ithaca, NY: ILR Press, 1992
Fraser, Steven, Labor Will Rule: Sidney Hillman and the Rise of American Labor,
New York, Free Press, 1991.
Haber, S., Efficiency and Uplift: Scientific Management in the Progressive Era,
Chicago: University of Chicago Press, 1964.
Hughes, T.P., American Genesis: A Century of Invention and Technological
Enthusiasm 1870-1970, New York: Viking, 1989.
Jacoby, S. M., “Union-Management Cooperation in the United States: Lessons
from the 1920s,” Industrial and Labor Relations Review, 37, 1, Oct. 1983, pp. 18-33.
Jacobson, E.K., NUMMI: A Model in Human Relations, BSIE thesis, General
Motors Engineering and Management Institute, 1986.
Krafcik, J., “Learning from NUMMI,” MIT International Motor Vehicle Program,
Lenin, V.I., “The Immediate Tasks of the Soviet Government,” Izvestia, 28 April,
1918, translated in Selected Works, 2 volumes, Moscow, 1967, vol. 2.
Merkle, J.A., Management and Ideology: The Legacy of the International Scientific
Management Movement, Berkeley, Calif.: University of California Press, 1980.
Walton, R.E., Up and Running: Integrating Information Technology and
Organization, Boston: Harvard Business School Press, 1989.