Product Development Strategy for the High-tech Startup:
A System Dynamics Post-Mortem
Corey Lofdahl
SAIC
Burlington, MA USA
Clyde Lofdahl
LCG
Littleton, CO USA
Prepared for presentation at
The 19" International Conference of the System Dynamics Society
Atlanta, GA - July 2001
Abstract
This study reviews the case of a failed high-technology product development effort, the Multi-Chip
Module (MCM) project. Specifically it reviews the strategic decisions that led to the project’s demise,
and using simulation it constructs an alternate strategy that would have proven more successful. The
difference between the two strategies centers on the relative timing between product development and
marketing efforts. The first strategy, which was actually pursued, stresses acquiring customers early
and learning about the production process while filling actual orders. The second, hypothetical strategy
centers instead on leaming about the production process first and then, only after production proficiency
has been achieved, shifting attention and resources to the task of customer acquisition. Simulation-
based scenario analysis demonstrates the second strategy’s superior performance as measured by the
MCM project's path to profitability.
Corey Lofdahl holds degrees in electrical engineering, computer science, and international relations
from the University of Colorado, Brown University, and MIT. He is a branch manager with SAIC’s
Simulation and Information Technology Operation.
Clyde Lofdahl holds degrees in mechanical engineering and business administration from the University
of Idaho and Harvard University. He is a management consultant and investor.
Introduction
Many businesses have shut their doors since the dot-com bubble burst in April 2000, so many in fact
that one popular web site - the dot-com deadpool - exists simply to mark (and sometimes gloat over)
their passing. While some failed companies have undoubtedly been the unwitting victims of collapsing
capital markets, others were fatally flawed from the beginning and should never have been conceived,
funded, or staffed. Thus nascent companies come in three flavors: those that are (1) perfectly
conceived, (2) flawed and fixable, and (3) flawed and unfixable. For the purposes of this paper, let us
assume that the majority of new companies are of type (2) or (3) - that is, they are flawed - so the
question then tums to whether or not they are fixable. This is usually determined by funding and
staffing a company and then letting it compete in the marketplace where the staff “fixes” the company in
real-time. If the company makes a profit, then it was fixable; if it doesn’t, then it wasn’t.
Of course the staff, committed as they are to the company’s success, always believe their
situation is salvageable given enough time and money. This is why pulling the life-support on dying
companies is so difficult. This paper explores the application of simulation generally, and system
dynamics specifically, to the crafting of strategies that increase the likelihood of success for new
commercial ventures. Simulation is offered for this task because fixing key policy problems with a
model on a laptop in an afternoon is much less costly and painful than fixing them with real employees
and real money in real-time. In considering the problem of crafting corporate strategy, it is easy to
focus on the intemal workings of the firm. However, it should be recognized that firms exists in
markets where they compete with other firms. In this manner, the crafting of internal policy blurs into
the formulation of competitive strategy. Whether simulation is used to determine a company’s internal
workings or its strategy across a range of competitive and market conditions, the key is to avoid costly
problems and to maximize the company’s chances of profitability and survival. If solutions cannot be
found for likely problems, then the company should be considered unfixable and should go unfunded.
Working in the realm of high-technology complicates matters by increasing the unfriendly
factors of complexity, risk, and uncertainty. High-tech product development, by its very nature, implies
a diminished history on which decisions might be based. Every time, it seems, feels like the first time.
And while the decisions are difficult, the stakes are high: the careers of the project team and millions of
investment dollars are on the line. Given the paucity of data and the high stakes associated with high-
tech product development, it makes sense to spend extra time on those critical decisions, policies, and
strategies that will ultimately determine the fate of the firm.
Focusing of critical decisions is however not as easy as it sounds though. How does one know
that a particular decision is critical? In a high-tech research and development (R&D) environment, the
importance of seemingly simple decisions is often revealed only later, and seemingly important questions
are later revealed to be of only minor significance. Recognizing this, it is similarly impossible to give
high priority to all decisions as “analysis paralysis” inevitably results. Even after acknowledging these
observations as true, it still makes sense to spend extra time on certain key questions and problems.
This study presents a case study of a high-tech, product development effort that pursued a strategy that
ultimately led to its demise. It does so using system dynamics simulation to provide an example of what
a cnucial, initial, and incorrect strategy looks like. An alternative strategy is then developed that shows
how a different set of decisions might have yielded a superior result. This study is organized into three
parts: first, a background of the motivations and facts underlying the product development example is
provided; second, a model is developed that captures the relevant relationships of the product
development process; third, a dynamic analysis is performed with an eye towards developing a better
guiding strategy that leads to a better final outcome. Concluding thoughts are presented in the final
section.
Part 1 - Product Background
The project took place at a Silicon Valley firm from 1984 to 1992 and centered on the development of
Multi-Chip Modules (MCM), products that concentrate computational capability in a smaller area than
had been possible until that time. The idea was to put a whole computer on a single silicon wafer, but
the parent company had little previous experience in this area. In fact, the parent company’s primary
expertise was in wire, cable, and electrical interconnection products, and it had watched the integrated
circuit (IC) and computer industries grow up around it. It had never participated in either of these
important and lucrative markets; the MCM project was to provide its entry.
Figure 1. Example Multi-Chip Module (MC M) Products
The development strategy was to use established IC manufacturing equipment and methods to
produce MCMs, which were in effect a printed circuit boards shrunk down to 1/10 scale. The MCMs
were used to connect ICs and associated discrete electrical components -- i.e., capacitors, resistors, and
other devices. At the time, printed circuit board standard conductors were 10 mils, or 10 thousandths
of an inch wide. Conductors in the MCM were intended to be 1 mil or one tenth as wide, a significant
technical leap. However, the project was launched with an initial development team that did not include
IC expertise. The possibility of hiring an expert was discussed and several were interviewed, but their
salary demands were substantially higher than the company’s salary structure allowed. Thus the
development team was formed entirely with in-house personnel.
The product development philosophy of the company for many years had been, “Take a big order
and learn to make it on the run!” This strategy had proven successful on many occasions and was a
matter of pride within the company. Time spent in R&D by a small team was viewed with suspicion.
Management's view was, “There is nothing like a big customer yelling for his order to focus effort and
speed development!”
The first MCMs produced were test pieces that proved manufacturing capability, but within 90
days a senior company executive booked an order with a large computer manufacturer to, “give the
MCM team something to chew on.” It took over 18 months to deliver this order. During this time,
additional orders were booked with major computer manufacturers and large military electronic
contractors in the US and Europe. Orders were not the problem; deliveries were.
The foremost delivery problem was always manufacturing yield, and the essence of the yield
problem centers on geometry. If a 5-inch diameter wafer is used in IC manufacturing, it will product
fifty 0.5 inch square ICs. If there are 5 defects (i-e., electrical opens or shorts in the circuitry) in the
wafer, there can be up to 5 defective ICs. 45 good ICs out of 50 produced means a 90% yield, which is
quite acceptable. If a 5-inch diameter wafer is used in MCM manufacturing, it will produce five 1.5 in
square MCMs. If there are the same 5 defects in this wafer, there can be up to 5 defective MCMs, fora
yield of zero. So when wafer (IC-like) manufacturing methods are used, the larger the final product, the
smaller the yield.
The following list outlines the number of MCM products per wafer for customer orders during the
MCM project:
1 per wafer 13% of orders
2 per wafer 41% of orders
3 per wafer 22% of orders
4 per wafer 14% of orders
5 or more per wafer 10% of orders
The “book a big order and learn to make it on the run” philosophy was ruinous for the MCM project.
The pressure from customer orders with firm delivery dates meant that there was never sufficient
manufacturing capacity to produce development orders from which the group could leam to improve
yields. During the first year manufacturing yields were about 5%. Eight years later they were in the 20-
25% range. To break even financially required yields of 50%, and this was never achieved.
The high outflow of cash or “bum rate” from the large-scale production facility and its attendant
multi-shift crews meant that top management was always concemed. Any suggestion of scaling back
until yields could be improved was quickly dismissed. This MCM facility was known throughout the IC
and computer business worlds, and scaling back the effort would have proven embarrassing to the
parent company. After eight years of heavy losses, the MCM facility was shut down in 1992.
Part 2 - The Model
Y ears afterwards, the project’ s facility manager was introduced to system dynamics, and the Product
Development (PD) model resulted from the initial modeling effort. The PD model, shown below in
Figure 2, was created to explore, in retrospect, how the project might have been better understood,
funded, and managed.
THE PRODUCT DEVELOPMENT MODEL
= ee
4, DELIVERY
DELAY
6. CUSTOMER
Figure 2. The Product Development (PD) Model
In keeping with the traditional strengths of system dynamics modeling, the actual physical processes and
behaviors of the MCM manufacturing effort are modeled rather than the abstract financials. The model
is comprised of six sections: (1) manufacturing, (2) productivity, (3) labor, (4) delivery delay, (5) sales,
and (6) customer, each of which are discussed below.
Section 1, Manufacturing
This section contains the key relationship in the model - that between the manufacturing yield
(Mfg_Y ield) and customer order work (Cust_Order_ Work) stocks. The relationship between the two is
simple: the more customer order work, the slower the increase of manufacturing yield. This relationship
requires some explanation. To begin, recognize that the MCM manufacturing process is complex,
which means there are many process parameters that can be varied, and the yields change from batch to
batch even when the parameters are ostensibly kept the same. Leaming takes place slowly in such an
environment as knowledge and experience build gradually. Experience accumulates more quickly if
those responsible for manufacturing are given the chance to ‘play’ with the process - that is, if they are
given the chance to vary the parameters and experiment with unusual configurations in a low-pressure,
consequence free environment. Conversely, experience accumulates more slowly if pressure is placed
upon the manufacturing team per the theory of bounded rationality’. In times of stress people tend to
rely on few and certain data. Here, if the pressure is on to deliver an order, people naturally tend to
concentrate on what worked before rather than risk an even lower yield. So the more customer orders,
the more stress, the slower the learning process, and the lower the yield. Upon recognizing the
importance of manufacturing yield, the strategy should have been not to, “book big orders”, but rather
* For more on this topic, see John D.W. Morecroft’s 1983 article, “System Dynamics: portraying bounded
rationality,” in the International J ournal of Management Science 11(2), pp. 131— 42.
to keep the project small and minimize market contact until production capability (i.e., manufacturing
yield, product design, and other skills) had been leaned and demonstrated.
Section 2. Productivity
If the previous section concems production “know how,” this section concerns its application to actual
production. Thus manufacturing yield is used to generate both usable and flawed product. Usable or
finished product is shipped, which when multiplied by price generates revenue. Costs are then
subtracted from revenue to determine profit.
Section 3. Labor
This section covers the normal hiring and training of workers in accordance with the production
demands as represented by order backlog. As demand increases, it also drivess new investment in
production capacity.
Section 4. Delivery Delay
A basic axiom of business is that low price leads to more customers and high price to fewer customers.
Likewise, timely deliveries have a positive influence on customers and late deliveries a negative
influence. This section tracks delivery delay and relates it to the sales process as a function of backlog -
the larger the backlog, the longer the delivery delay.
Sections 5 and 6. Sales and Customer
These sections concentrate on the process of acquiring customers and generating business from them.
Competition was never a serious factor for MCM - the main problem was delivering product in an
efficient manner. These sections include several influences including price and delivery delay on
customers. Product price was a major issue in the MCM project, which is captured by the key price
factors of shipment size, production capacity, and manufacturing yield.
Profit and Loss
As noted previously, revenue is generated in the productivity section when product is shipped and
billed. A separate section, not shown in Figure 2, calculates profit and loss essentially by counting the
costs of production including materials, labor, marketing, technical staff, overhead, and intellectual
property. Profit is calculated by subtracting these costs from revenue.
The resulting model, while not a perfect representation of reality, provides a reasonable measure
of correspondence to the MCM project as it was actually run. In the next section, the model is used to
explore and improve the operational strategy that eventually led to the demise of the MCM project.
Additionally, several alternative policies are proposed, implemented, and evaluated in an effort to craft a
more workable strategy that might have led to a more positive outcome.
Part 3 - Crafting Alternative Strategy
The key variable for this discussion is marketing effect, a graphical function that relates the effort and
expense necessary to identify and acquire new customers with the technical state of the production
process as measured by manufacturing yield.
1.000
—~=mM On-r=Z
0.000
Mig_Yield
Figure 3. Actual MCM Policy - Acquire Customers Early
Figure 3 captures the product development strategy as it was actually practiced. The horizontal
or X-axis measures the manufacturing yield in terms of percentage of usable product. The scale’s
maximum range is 60%, with postulated breakeven being about 50% and the best achieved yield being
about 25 to 30%. The vertical or Y -axis shows the output of marketing effect, the effort expended to
acquire customers using a dimensionless measure ranging from 0 to 1 with 0 being the minimum effort
and 1 the maximum. Note that the graph depicts fairly consistent and high effort to acquire customers
regardless of the value of manufacturing yield, starting off at a value of 0.6 and increasing to a value of
0.9. In this fashion, the graph captures management's policy of booking orders early and learning on
the fly.
1: Mfg Yield 2: profit loss from operations 3: Cum Prof Loss
Bhs
VS nae
50000.00
0.00
+ FE
z “BODO DOS eeeeeeeeeeeeeeeee
3: -3000000.00
1: i
2: -150000.00 i i i i
3: -6000000 .00: T T T 1
1.00 16.00 31.00 46.00 61.00
Months
Figure 4. Actual MCM Performance
Figure 3 demonstrates the consequences of the pursued strategy, consistent losses on the order
of $100,000 per month. Driving this consistent loss of money is the presence of customer orders that
effectively prohibit experimentation and learning about the production process - such leaming would
have allowed the manufacturing yield to improve. In this scenario, while manufacturing yield increases,
it never reaches the 50% mark deemed necessary for profitability.
fi 000
M
k
t
9
E
f
f
Jo 000
I
fo 000 fo 600
Mig_Yield
Figure 5. Acquire customers after production understood
Figure 5 graphically describes the strategy of expending serious energy acquiring customers only
after production yields approach a close to profitable level, in this case about 35%. Below that, only
token customer acquisition efforts are undertaken.
1: Mfg Yield 2: profit loss from operations 3: Cum Prof Loss
1: [te eaereerrer peers
2: §0000.00
3 0.00
1: 0.30
2 SOODD DDfesesseeeeeeeseeyeee
3: -3000000.00
ig 0.00
2: -150000.00
3: -6000000.00:
1.00 16.00 31.00 46.00 61.00
Months
Figure 6. Performance of delayed customer acquisition
Figure 6 shows the results obtained by implementing the graphical policy presented in Figure 5.
Note specifically an improved rise of manufacturing yield, which leads to improved financial
performance in the form of lower losses. So while profitability remains elusive in this scenario, the
MGM project's financials are at least headed in the right direction.
[roo
M
k
t
9
E
Jooor
I
p.000 [eo
Mfo_Yield
Figure 7. Acquire customers after production well understood
If waiting a while before acquiring customers is good, how about waiting even longer? Figure 7
graphically describes the strategy of expending serious energy to acquire customers only after
production yields approach an even more profitable level, in this case about 45%. Once again, only
token customer acquisition efforts are undertaken below that.
1: Mfg Yield 2: profit loss from operations 3: Cum Prof Loss
$0000.00
0.00
0.30
-§0000.00-
-3000000.00
ees
-150000.00
-6000000 .00:
+ T 1
1.00 16.00 31.00 46.00 61.00
Months
Figure 8. Performance of customer acquisition delayed still further
Figure 8 shows the results obtained by implementing the graphical policy presented in Figure 7.
Note an even sharper rise of manufacturing yield, which once again leads to improved financial
performance in the form of profitability around month 50. Had such performance actually been
achieved, the MGM product development project would have been deemed a success. Note also the
oscillations of the profit and loss curve. These stem from costs on the operational side, most notably
the salaries of the technical staff. Staff levels oscillate due to the time delays associated with acquiring
and training staff in response to product backlogs.
1.000
mam Ors
0.000
Mfg_Yield
Figure 9. Gradual acquisition of customers as production is understood
One aspect of the policies presented in Figures 5 and 7 is their troubling binary nature. That is,
there is a tendency to work at a reduced level until manufacturing yield reaches a previously defined
target. Often however, such targets are ill specified and difficult to define. Figure 9 shows a policy
curve with a different shape denoting the gradual increase of marketing effort as manufacturing yield
improves. This seems a more plausible and sustainable policy.
1: Mfg Yield 2: profit loss from operations 3: Cum Prof Loss
Pe etn re gms
a | he
0.604
50000.00
0.00
-50000.00'
-3000000.00
-160000.00
-6000000.00:
y T T 1
1.00 16.00 31.00 46.00 61.00
Months
Figure 10. Performance of gradual acquisition of customers
Figure 10 shows the results obtained by implementing the graphical policy presented in Figure 9.
Even though manufacturing yield, climbs at about the same rate as Figure 8, profitability is much higher
and the response oscillates similarly. The key point is that even though the Figure 9 policy curve is
shaped differently than Figures 5 and 7, it pushes off marketing effort until the production process is
better understood, which is the key to profitability in the MCM domain.
Conclusion
In reviewing the results of the product development model, the question of whether or not success was
“dialed in” is addressed. That is, the conclusion of doing development first and then marketing as
opposed to development and marketing simultaneously is, in retrospect, almost self-evident. Moreover,
the algorithm used to generate increases in manufacturing yield are oversimplified, primarily because
capturing the complexity of the MCM production process is impossible and would not have been worth
the effort even if it were possible.
The benefits of PD model are more subtle. First, the model shows in an operational fashion the
types of losses that can be incurred should the manufacturing problems go unsolved for a long period,
as actually occurred. Similar models could have been run using a spreadsheet model, though they
would have lacked the operational flavor of this model as well as its complex feedback relationships.
Moreover, the model shows the complex, causal connections that link marketing effort with the
production effort. Without such explicit connections, one might assume that the two efforts are
independent - that decisions made in marketing are unrelated to production or vice versa. One might
also assume, as did the senior executives of the MCM project's parent company, that marketing effort
would spur the development effort. This proved not to be the case as the drive to deliver product to the
customer effectively stifled leaming about the manufacturing process. This in tum kept manufacturing
yields at low and unprofitable levels.
The product development model shows that a strategy focusing on customer orders at the
expense of research and development lead to high burn rates, dissatisfied customers, and a drawn out
learning curve. In light of this, it seems that a strategy predicated on leaming first and selling second
would have lowered the bum rate, retained the goodwill of customers, and accelerated learning. In
making this distinction, recall that the project continued for eight years, spent millions of dollars, and
consumed the time of many bright people. While this lesson might seem obvious or unique, such stories
are all too common in business. The problem was not that the people were insufficiently dedicated,
informed, or intelligent, but that there was no framework to integrate and make sense of the disparate
and diffuse data. The system dynamics model developed and referenced herein provides a framework
that delivers value not because it uncovers new information but because it orders and interprets
available information in such a way as to provide new insights.
CC BY-NC-SA 4.0