Professional Documents
Culture Documents
Simchi Levi2015
Simchi Levi2015
Interfaces
Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org
This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@informs.org.
The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.
INFORMS is the largest professional society in the world for professionals in the fields of operations research, management
science, and analytics.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
Vol. 45, No. 5, September–October 2015, pp. 375–390
ISSN 0092-2102 (print) ISSN 1526-551X (online) http://dx.doi.org/10.1287/inte.2015.0804
© 2015 INFORMS
William Schmidt
Samuel Curtis Johnson Graduate School of Management, Cornell University, Ithaca, New York 14853,
wschmidt@cornell.edu
Yehua Wei
The Fuqua School of Business, Duke University, Durham, North Carolina 27708, yehua.wei@duke.edu
Keith Combs, Yao Ge, Oleg Gusikhin, Michael Sanders, Don Zhang
Ford Motor Company, Dearborn, Michigan 48126
{kcombs@ford.com, yge@ford.com, ogsukhi@ford.com, msande22@ford.com, xzhang35@ford.com}
Firms are exposed to a variety of low-probability, high-impact risks that can disrupt their operations and supply
chains. These risks are difficult to predict and quantify; therefore, they are difficult to manage. As a result,
managers may suboptimally deploy countermeasures, leaving their firms exposed to some risks, while wasting
resources to mitigate other risks that would not cause significant damage. In a three-year research engagement
with Ford Motor Company, we addressed this practical need by developing a novel risk-exposure model that
assesses the impact of a disruption originating anywhere in a firm’s supply chain. Our approach defers the
need for a company to estimate the probability associated with any specific disruption risk until after it has
learned the effect such a disruption will have on its operations. As a result, the company can make more
informed decisions about where to focus its limited risk-management resources. We demonstrate how Ford
applied this model to identify previously unrecognized risk exposures, evaluate predisruption risk-mitigation
actions, and develop optimal postdisruption contingency plans, including circumstances in which the duration
of the disruption is unknown.
Keywords: risk management; automotive; manufacturing industries; disruption; risk-exposure index.
History: This paper was refereed.
Normal accident theory holds that because major been given to evaluating the impact of a disruption
disruptions are an inherent property of such com- based on the optimal response of an existing network
plex and tightly coupled systems, they should be once that disruption has occurred. A recent exception
considered unavoidable or normal (Perrow 2011). It is Schmitt (2011), which evaluates response strategies
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
falls to operations and supply chain managers to that minimize the service-level impact when disrup-
navigate this new normal. Traditional operational- tion occurs on a multiechelon network for a random
disruption risk-assessment methods oblige firms to duration. Another is MacKenzie et al. (2014), which
identify the probability and magnitude of disrup- evaluates the interaction between the supplier and
tion risks early in the analysis process (Sampson and buyer response strategies under a random-duration
Smith 1982, Knemeyer et al. 2009); however, managers disruption.
face a number of challenges in implementing such We make three important contributions to this liter-
a solution. First, it is difficult and often impossible ature. First, we develop our model for practical appli-
for managers to accurately estimate the likelihood of cations using large-scale supply chain data from Ford.
low-probability, high-impact disruptive events (Banks Second, we evaluate the optimal contingency plans
2005, Taleb 2007). Second, managers tend to misal- for settings in which the disruption duration is either
locate resources when facing low-probability events known exactly or described by an uncertainty set.
(Kahneman and Tversky 1979, Johnson et al. 1993), Finally, our model quantifies the disruption exposure
ignore risks regardless of their potential significance across all the nodes in the company’s supply chain
(March and Shapira 1987), and distrust or disregard based on company-level performance impacts.
precise probability estimates (Kunreuther 1976, March The second stream of research seeks to classify
and Shapira 1987). This can lead to inaction; Mitroff operational disruptions and quantify their impact.
and Alpaslan (2003) found that most firms do little Scholars and practitioners generally agree that oper-
to proactively prepare for such low-probability, high- ational disruptions materially and negatively impact
impact disruptive events. company performance on average (Sheffi 2005,
In this paper, we apply a new model, proposed Hendricks and Singhal 2005, World Economic Forum
by Simchi-Levi in March 2012 (Gilmore 2012) and 2013). There is less agreement, however, on how
described in Simchi-Levi et al. (2014), for analyz- we should classify and forecast such disruptions
ing operational-disruption risk and detail the devel- (Kleindorfer and Saad 2005, Tang 2006, Wagner and
opment and implementation of this model at Ford. Bode 2006, Sodhi et al. 2012). Researchers are only
Throughout the paper, we share the primary results beginning to understand which disruptions have the
of our analysis using masked versions of Ford’s oper- greatest impact on firm performance. Answering this
ational and supply chain data. research question is important because it informs
firms on which disruptions warrant mitigation invest-
ments. Craighead et al. (2007) propose that supply
Literature Review chain density, complexity, and node criticality con-
We leverage two streams of research in our work. tribute to the severity of disruptions. Tang (2006)
The first area of scholarship pertains to supply chain theorizes that a firm’s vulnerability to disruption
network modeling and optimization, which broadly depends on its supply chain strategies, including
consider the optimal network structure under steady postponement strategies and inventory placement.
state operations (Fisher et al. 1997, Graves and Braunscheidel and Suresh (2009) identify that a firm’s
Willems 2003) or under the possibility of a disrup- organizational integration practices are associated
tion (Snyder et al. 2006, Peng et al. 2011, Mak and with the firm’s ability to mitigate the consequences
Shen 2012). Closely related is research that evaluates of disruptions. Kleindorfer and Saad (2005) provide
coordination strategies between buyers and suppli- evidence that changes to risk-assessment and risk-
ers in the presence of disruption risk (Tomlin 2006, mitigation practices reduce the impact of disruptions
Chopra et al. 2007, Tomlin 2009). Less attention has in the chemical industry.
Simchi-Levi et al.: Managing Risks and Disruptions in Automotive Supply Chain
Interfaces 45(5), pp. 375–390, © 2015 INFORMS 377
We contribute to this body of research by identi- Given these limitations, legacy risk-management
fying the specific nodes in a firm’s operations and processes often focus on tracking the status of only
supply chain that would, if disrupted, result in the a handful of suppliers and part numbers. These tend
greatest damage to firm performance. We believe that to be suppliers that provide major assembly compo-
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
this result is particularly beneficial in an applied set- nents and represent a large portion of the total compo-
ting because it allows firms to understand their expo- nent costs. Many large manufacturers recognize that
sures at specific operational locations and put in place material exposures are likely to be hidden among the
countermeasures that address the greatest sources of suppliers who are not included in this regular review
exposure. process. Because of the difficulties in predicting dis-
ruptions, the data limitations, and the size of their
Our research generally aligns with concepts applied
supply chains, companies often cannot identify where
in other disciplines, including estimating maximum
these exposures are, much less quantify their impact.
foreseeable loss (i.e., the maximum loss if all safe-
For example, managers at Ford estimate that conduct-
guards in a system break) in the insurance indus-
ing a traditional risk analysis for all of Ford’s more-
try and conducting failure analysis (i.e., assessing the
than-4,000 Tier 1 supplier sites would likely take two
structural resilience when a critical member of a sys- or three years, at which time the analysis would be
tem is removed) in structural design. Until now, how- obsolete.
ever, the field of operational risk management has not
given these principles much attention.
Our Approach: Risk-Exposure Index
Recognizing that managers have limited ability to
predict low-probability, high-impact risks or collect
Limitations of the detailed data on lower levels of their supply chain,
Legacy Risk-Analysis Approach our approach, initially described in Simchi-Levi et al.
For many companies, even those that have world-class (2014), advocates integrating a vulnerability-based
operations and supply chain management systems, analysis into supply chain risk assessments. In such
proactively managing high-impact, low-probability an analysis, the focus is on understanding the impact
disruption risks is challenging. One obstacle to con- of a disruption, regardless of its source. This defers
ducting a more insightful analysis of disruption risks the need to estimate the probability associated with
is that operational disruptions are both difficult to pre- any specific risk and collect detailed information from
dict and have a highly uncertain impact on perfor- subtier suppliers until after Ford has determined the
mance. In Ford’s case, the scale and dynamic nature impact a disruption will have on its operations. At
of its supply chain further complicate this problem. that point, Ford can make a more informed deci-
sion about where to focus its limited risk-assessment
These factors increase both the number of disrup-
resources. Our approach suits the goal of analyzing
tion scenarios to consider and the frequency at which
supply chain disruptions because the impact of a dis-
we should evaluate those scenarios. A second obsta-
ruption often does not depend on the cause of the
cle is data availability, particularly on suppliers at
disruption but rather on its duration. In addition,
lower tiers within the supply chain. Supply chain
the potential mitigation actions that a company can
transparency is a challenge for the entire automotive practically employ in response to a supply chain dis-
industry. Suppliers to the industry have historically ruption are often the same regardless of the specific
been reluctant to provide the automobile manufac- causes of the disruption. Finally, our approach implic-
turers with detailed information about their suppliers itly recognizes that supply chains are in a continu-
and their suppliers’ suppliers. As a result, although ous state of flux. In the face of such constant change,
manufacturers typically have good information on maintaining up-to-date predictions of the likelihood
Tier 1 suppliers (i.e., companies that supply directly to of specific risks is nontrivial; however, given that a
the manufacturer), they have considerably less infor- disruption does occur, estimating a firm’s vulnerabil-
mation on lower-tier suppliers in the supply chain. ity is more tractable.
Simchi-Levi et al.: Managing Risks and Disruptions in Automotive Supply Chain
378 Interfaces 45(5), pp. 375–390, © 2015 INFORMS
Time-to-Recover Model may occur when multiple firms try to adjust to sup-
The model considers the supply chain as a graph rep- ply disruptions at the same time. For example, if a
resenting the movements of supplier parts from each supplier fails to deliver to one firm, it may have gone
supplier facility to each of a firm’s facilities and prod- down for multiple firms. Such an event makes other
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
uct lines. A node, also referred to as a stage, in the potentially compensating nodes (e.g., backup suppli-
graph is equivalent to a part or manufacturing pro- ers) more congested.
cess at a particular supplier or Ford facility. Inputs to
the model include operational and (or) financial mea- Time-to-Survive Model
sures (e.g., unit profitability) and in-transit and on-site In many cases, accurate TTR information may not be
inventory levels for each node. Our model incorpo- available. More importantly, a supplier may be opti-
rates the time-to-recover (TTR) of each node in the mistic when assessing its TTR; that is, a supplier may
supply chain network, which represents the time it underestimate the time required to recover and hence
takes for a node to recover to full functionality after may underestimate Ford’s exposure to a disruption.
a disruption (Miklovic and Witty 2010, Simchi-Levi Therefore, Ford is interested in identifying suppli-
et al. 2014). This value can be unique at each node in ers whose disruption impact is sensitive to the given
the firm’s supply chain. TTR information. For this purpose, we introduce the
The model iterates over each node in the graph, time-to-survive (TTS) concept, which we define as the
disrupting the node for the duration of its TTR and maximum amount of time the system can function
calculating the corresponding impact on the firm’s without performance loss if a particular node is dis-
performance. It determines the performance impact rupted (Simchi-Levi et al. 2015). As we will show,
assuming the firm responds optimally to the disrup- we determine the TTS associated with a specific node
tion scenarios, where the model simulates the optimal by solving an optimization problem that takes into
responses by solving an associated linear optimiza- account the disruption of this node, inventory levels,
tion problem; see Appendix A for details. The model and alternative sources of supply; see Appendix B
can accommodate different performance measures as for the model formulation. The firm can determine
the objective for this optimization, including mini- whether a more accurate measure of TTR is neces-
mizing the lost units of production, lost sales, or lost sary by comparing the TTS value associated with a
profit margin. For each disruption scenario, the model specific node with the TTR estimate of that node. If
the TTS far exceeds the TTR, it implies that a large
searches on how to reallocate existing inventory, redi-
change in TTR will have little impact on the firm’s
rect supply alternatives, and idle downstream plants
risk exposure; however, nodes with short TTS values
such that the disruption has the smallest impact. The
require Ford to engage these suppliers in a detailed
resulting performance impact (PI) is the impact of
discussion about their TTRs.
that disruption scenario on the firm’s chosen perfor-
mance measure during the TTR. To simplify cross-
scenario comparisons, the model can also calculate Implementation at Ford
a risk-exposure index (REI) (Simchi-Levi et al. 2014), We implemented our model as a decision support
which normalizes the PI for each scenario by the max- system during a three-year research engagement
imum PI over all scenarios considered in the analysis. between MIT and Ford. The first phase of the project
The model can accommodate simultaneous disrup- included the assessment of existing risk-management
tions in multiple supply chain nodes. This allows approaches. In the second phase, we worked with the
management to analyze complex disruption scenar- Ford optimization and information technology (IT)
ios, including disruptions that affect all the parts from teams to focus on model design and implementa-
one supplier plant or disruptions that affect all the tion, and the integration of the optimization model
same part regardless of the supplier. We can extend and Ford’s IT system. The modeler and optimization
the model to account for alternative sources of supply specialists communicated weekly, and received help
and supplier capacity commitments. This facilitates from Ford’s procurement team to validate the model’s
an explicit examination of interactive effects, which output.
Simchi-Levi et al.: Managing Risks and Disruptions in Automotive Supply Chain
Interfaces 45(5), pp. 375–390, © 2015 INFORMS 379
250
200
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
Number of suppliers
150
100
50
0
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2.0
10.0
20.0
30.0
40.0
50.0
More
TTS (weeks)
Figure 1: A significant portion of the suppliers have very low TTS values, thus requiring more accurate TTR
evaluation and closer monitoring for risk-exposure assessment. In addition, some suppliers have very high TTS
values, possibly because of redundant inventory buffers.
Ford’s procurement staff used the decision sup- Evaluation of Node Criticality with the TTS Model
port system in three ways: (1) strategically, to identify As discussed in the Time-to-Survive Model section, TTR
exposure to risk associated with parts and suppliers, information is not known accurately in many practi-
effectively prioritize and allocate resources, segment cal situations because of information uncertainty and
suppliers, and develop mitigation strategies; (2) tac- optimistic supplier assessments. Therefore, the first
tically, to track daily changes in risk exposure to step of our risk-analysis process is to identify the dis-
alert procurement executives to changes in their risk ruption scenarios that would lead to immediate per-
position; and (3) operationally, to identify effective formance deterioration, namely, to find nodes with
ways to allocate resources after a disruption. Using small TTS values. Nodes that represent higher expo-
the model to conduct a comprehensive analysis of sure levels will have a TTS value that is lower than
its risk exposures (i.e., the strategic level), Ford iden- a threshold value, for example, TTR plus a safety
tified several supply chain nodes that would have allowance.
a large impact on its operations if disrupted. These Figure 1 shows that the suppliers included in the
large exposures lie in unlikely places, such as non- analysis have a range of TTS values. Many suppli-
strategic suppliers or parts that the company spends ers have TTS values of less than one week. Ford’s
relatively little money to procure. Armed with this management can use this information to concentrate
information, Ford can make more informed decisions on the PI of low-TTS suppliers and acquire corre-
on how to deploy its risk-assessment resources and sponding TTR information. In addition, by identify-
mitigate the effects of a disruption to these nodes. ing the nodes with high TTS values, this analysis can
In this section, we describe the insights our model identify potential waste, caused by excessive protec-
provides at the strategic, tactical, and operational lev- tion (strategic inventory), within the system. For such
els for Ford’s risk-analysis, procurement, and man- nodes, a firm may reduce (strategic) inventory, thus
agement teams. providing significant cost savings.
Simchi-Levi et al.: Managing Risks and Disruptions in Automotive Supply Chain
380 Interfaces 45(5), pp. 375–390, © 2015 INFORMS
2,773
1,801
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
1,601
1,401
1,201
Number of sites
1,001
801
601
401 805
201 408
252
142 154
1
No impact Very low Low Medium High Very high
Performance impact (lost sales in vehicle volume)
Figure 2: (Color online) Among 4,534 sites examined, 2,773 sites have zero impact at the time of analysis and
408 have very high impact.
Figure 3: (Color online) Impact of a disruption at a supplier site (node) on Ford’s lost profits is not correlated
with the amount Ford spends at the supplier. Each circle represents a unique supplier site.
suppliers, which accounts for about 80 percent of total shifts to a few manufacturers that dominate the mar-
spend. Typically, each of these components has a sin- ket because of their lower costs and superior qual-
gle strategic supplier. An appropriate supply strategy ity; as a result, engaging in a dual-sourcing strategy
for these items is to focus on long-term partnerships is difficult. In our experience, one possible mitigation
with suppliers and implement effective supply con- strategy involves a new product design in which com-
tracts where Ford can share risks with suppliers and ponents are standardized, allowing the firm to shift
track performance. Importantly, because of the high more volume and more spend to the supplier; hence,
total spend with these suppliers, Ford may be able to the firm would be in a good position to require dual
compel some of these suppliers to have backup sup- sites.
ply sites in different regions.
The most challenging suppliers are those whose Application to Tactical Decisions
total spend is low and PI is high (i.e., suppliers at Recall that for some components, risk exposure is
the bottom right side of the chart). To ensure sup- directly proportional to the level of inventory of that
ply, a firm may invest in inventory, require the sup- component in Ford’s supply chain. To identify risk
plier to have dual sites in different regions, or apply exposure, pipeline inventory information is uploaded
a dual-sourcing strategy. Unfortunately, each of these to the system on a regular basis, and the system deter-
mitigation strategies may cause a problem. Invest- mines the performance impact by component any-
ing in inventory may not be consistent with the where in the supply chain. When performance impact
lean strategy the company is applying. Low total is above a specific level, procurement specialists initi-
spend implies that the firm is not in a good posi- ate a process to understand the reason and take cor-
tion to require the supplier to have multiple sites. rective action. In that respect, our system serves as a
In addition, some of these suppliers are associated control tower that allows the firm to monitor suppli-
with high-volume, low-cost, and low-margin compo- ers’ performance and inventory trends to take action
nents. For these components, competition typically before problems occur. Because the company takes
Simchi-Levi et al.: Managing Risks and Disruptions in Automotive Supply Chain
382 Interfaces 45(5), pp. 375–390, © 2015 INFORMS
s 0ARTNERSHIP
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
s ,ONG TERM CONTRACTS 0ERFORMANCE IMPACT LOST PROFIT s )NVENTORY
s 4RACK INVENTORY s $UAL SOURCING$UAL SITE
s .EW PRODUCT DESIGN
Figure 4: (Color online) This graph suggests supplier segmentation and different risk-mitigation strategies for
different groups of suppliers.
actions in anticipation of a potential adverse event, and capacities under such uncertainty in TTR values
it can minimize the financial impact if such events is important.
happen. Figure 5 provides a stylized example that compares
the impact of different resource-allocation strategies
Application to Operational Decisions when the length of the disruption varies. In this fig-
Operationally, Ford supply-risk specialists use the ure, each curve represents the financial impact of one
model to respond to a disruption event. For example, resource-allocation strategy. For example, the solid
a few months ago, political problems in one region curve corresponds to the optimal resource-allocation
motivated the procurement department to identify the strategy for TTR = 1; we evaluate the performance
high-exposure suppliers in that region and find alter- of this resource-allocation strategy for all TTR val-
native sources of supply for these components. ues between 0 and 2. Similarly, the dotted curve is
In such situations, our TTR model optimizes inven- associated with the optimal resource-allocation strat-
tory and capacity allocation decisions when a dis- egy when TTR = 07. Figure 5 suggests that neither of
ruption occurs (Appendix A), assuming that accurate the two strategies dominates; that is, neither strategy
TTR information is available immediately after a dis- outperforms the other on all TTR values between 0.7
ruption occurs. Unfortunately, TTR may be different and 1. This is not always the case. Another stylized
for different modes of disruptions (e.g., process dis- example (Figure 6) shows that the strategy associated
ruption versus tooling damage), and the firm may not with the solid line outperforms the strategy associated
know the exact TTR value when a disruption occurs. with the dotted line. The former strategy outperforms
Therefore, identifying robust allocations of inventory all other strategies for TTR values between 0 and 2
Simchi-Levi et al.: Managing Risks and Disruptions in Automotive Supply Chain
Interfaces 45(5), pp. 375–390, © 2015 INFORMS 383
3.50
3.00 information. Figure 7 describes the system architec-
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
2.50
ture in which various data sets, including bill of
2.00
1.50 material, part routing, inventory levels, and plant
1.00 vehicle volumes are used to map Ford’s supply
0.50
chain. Gusikhin and Klampfl (2012) describe the basic
0
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 methodology of mapping the Ford supply chain.
TTR Our optimization engine uses the results to generate
the various performance impacts. These performance
Figure 5: (Color online) Each curve represents the financial impact of one measures are then presented to the users by Tableau
resource-allocation decision. The solid curve is optimal for TTR = 1, but
suboptimal for TTR = 07; the reverse is true for the dotted line. data visualization, which includes a geographic map-
ping capability. Thus, users can view results both in
tabular form and in various graphical formats. Fig-
1.60
ure 8 provides a screenshot of our interface; the size
1.40
1.20 of the circles identifies the performance impact of a
Financial loss
Vehicle volume
planning system
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
Supply chain
mapping
(java graph ETL)
Materials
planning and
logistics
Data
Central visualization
Purchasing repository (tableau)
system (SQL server)
Model interface
Risk exposure
Vehicle profit model
margins ( java-CPLEX)
Figure 7: (Color online) Ford’s risk-analysis framework integrates databases, our risk-exposure model, and an
output visualization tool.
procurement costs. Second, the cost of failure can be exposures as high-priority issues that require a for-
huge because supply chain disruptions can have a mal remediation analysis. Identifying these suppliers
significant impact on Ford’s ability to match supply is particularly compelling because they represent 1,500
with demand. Indeed, Ford estimates that the lost rev- additional supplier sites that will now receive a larger
enue associated with a disruption can be significant. share of Ford’s risk-management resources. Identify-
To illustrate this point, recall that in 2011 Toyota lost ing and addressing such risk exposures directly sup-
800,000 units of production volume as a result of the ports Ford’s corporate strategy.
Japan earthquake and more than 240,000 units of pro- A second benefit is identifying low-exposure sup-
duction volume as a result of the flood in Thailand. plier sites that are currently receiving an excessive
Honda faced similar challenges (Schmidt and Simchi- allocation of Ford’s risk-management resources. The
Levi 2013). model has identified over 400 supplier sites that
The risk-exposure model produces important and Ford includes in its risk-monitoring program, but
tangible benefits for Ford to help it effectively iden- which pose insignificant exposure to the company
tify and manage its risks. First, Ford has identified if disrupted. This information has allowed Ford to
supplier sites that have a material impact if dis- more efficiently allocate its supplier risk-management
rupted, but that it did not recognize as high-exposure resources.
sites. Based on the model results, 2,600 Tier 1 sup- By reallocating these resources, Ford is better able
plier sites have nonzero vehicle-volume impact that, to protect itself from the highest-impact exposures
if disrupted, would adversely impact Ford’s rev- in its operations and supply chain. For example, the
enue by up to $2.5 billion. Ford now classifies these lost revenue associated with a two-week disruption to
Simchi-Levi et al.: Managing Risks and Disruptions in Automotive Supply Chain
Interfaces 45(5), pp. 375–390, © 2015 INFORMS 385
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
Supplier Vehicle impacted Total part cost Financial impact Volume impact Supplier Part names
x11 cc1 $$$ $$$ vvv x11 y11
x12 cc2 $$$ $$$ vvv x11 y12
x13 cc3 $$$ $$$ vvv x11 y13
x14 cc4 $$$ $$$ vvv x12 y21
x15 cc5 $$$ $$$ vvv x12 y22
x16 cc6 $$$ $$$ vvv x13 y31
x17 cc7 $$$ $$$ vvv x13 y32
x18 cc8 $$$ $$$ vvv x13 y33
x19 cc9 $$$ $$$ vvv x13 y34
x20 cc10 $$$ $$$ vvv x14 y41
Figure 8: (Color online) Critical suppliers are mapped to geographical location. The size of a circle indicates
the magnitude of the impact on Ford’s performance if a supplier is disrupted. The table view gives detailed
information regarding the financial and vehicle-volume impact associated with these suppliers.
the newly classified high-impact supplier sites ranges Finally, our model detects hidden risks in Ford’s
from several hundred thousand dollars to $2.5 billion; supply chain. For example, it identified a low-cost
in contrast, the lost-revenue impact associated with a sensor that has high vehicle exposure; however,
two-week disruption to each of the formerly classified because of the low total spend, Ford’s procurement
high-impact supplier sites is minimal. In the words of group was not paying much attention to this com-
Ford manager Michael Sanders, “This has been one ponent. Following the risk analysis, the commodity
of the key game changers for us. This enables us to team acknowledged the sourcing concentration and
associated risk and developed a mitigation strategy.
focus on the supplier sites which would have a high
or very high impact on performance if disrupted, and
lets us put all our resources and all our knowledge Discussion
into making sure we have robust plans to protect us Firms operate in a constantly changing environment
in the event that something happens with any one of in which operational risks are increasing. In the auto-
those sites” (Simchi-Levi and Sanders 2013). motive industry, four factors contribute to increasing
Simchi-Levi et al.: Managing Risks and Disruptions in Automotive Supply Chain
386 Interfaces 45(5), pp. 375–390, © 2015 INFORMS
levels of operational vulnerability. The first factor is chain, and rapidly and consistently assess its sup-
the proliferation of global programs and the related ply chain risk-mitigation initiatives. Ford also takes
need to maximize the operational scale of these pro- advantage of the model’s capability of running at var-
grams. This results in less redundancy and more ious levels of detail. For example, in some applica-
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
dependence on fewer suppliers, increasing the supply tions, the company runs the model by aggregating
chain’s exposure if one of these suppliers is disrupted. nodes within a geographic region, and then drills
The second is the ongoing consolidation in the sup- down into more detail by running it using more gran-
ply base and the fiscal incentives to maximize the use ular representations for nodes.
of supplier resources. This also results in greater sup- Our risk-exposure model augments rather than
plier concentration and less slack capacity for the most replaces traditional risk-analysis methods. Ford incor-
critical subtier manufacturing components, including porates the results of the model with other indicators
electrical components, raw materials, and chemical that measure each supplier’s financial risk, including
precursors. Third, manufacturers’ efforts to push their metrics for financial health, and steady state opera-
Tier 1 suppliers toward lower costs ultimately drive tional risk, including metrics for service level perfor-
those suppliers to pursue subtier sourcing in emerg- mance and quality control compliance. Suppliers that
ing markets. This further extends the manufacturers’ trigger one or more risk areas (i.e., disruption, finan-
supply chains, adding more dependencies and poten- cial, or operational) are identified for follow-up with
tial points of failure. Finally, unlike the situation in the Ford’s supplier risk-management team. By including
PC industry, in the automotive industry, no common the model in its broader supplier risk-analysis pro-
standards are applied across other equipment man- cess, Ford can more confidently and accurately iden-
ufacturers (OEMs) for electronic components; hence, tify the areas in its supply chain and operations to
very few suppliers are available for these components. allocate its limited risk-management and mitigation
Any supplier disruption can shut down Ford’s ability resources.
to match supply with demand.
The automotive industry is not alone in facing Acknowledgments
We acknowledge the contributions of many people in Ford
increased disruption risk. Trends toward more ex- Purchasing and Information Technology (IT) for their sup-
tended supply chains and reduced operational buffers port of this project. We especially acknowledge Steve Faraci
are gripping many industries. As a result, supply from Ford Purchasing for his invaluable contributions. We
chain executives have a dual mission—to systemati- thank John Knowles and Dong Ruan from Ford IT for their
cally address extreme risks, such as hurricanes, epi- consultations and programming in data collections. We also
gratefully acknowledge John Ginder from Ford Research
demics, earthquakes, or port closings, and to manage
and Advanced Engineering for his leadership and strong
operational risks, such as forecast errors, sourcing support of the project. This work was funded by the Ford-
problems, and transportation breakdowns. Succeed- MIT Alliance.
ing in this dual mission is difficult because com-
pany operations and supply chains are increasingly Appendix A. Time-to-Recover Model
dynamic, and the occurrence and impact of disrup- We first present a single-tier (ST) supply chain model to
tions are difficult to predict. illustrate some of the main concepts, and then extend it to
In this paper, we provide a new approach for a multiple-tier (MT) model that encompasses more compo-
nents. The basic premise of both models is that, given a
supply chain risk management, which reduces the
supply chain structure (a graph) and a disruption scenario
need to estimate the likelihood of low-probability, (interrupted nodes and edges), we determine how to allo-
high-impact events. Our method focuses on eval- cate the firm’s remaining resources to optimize its ability to
uating a firm’s vulnerability, given that a disrup- satisfy exogenous demand. A node (or stage) in the graph is
tion could occur anywhere across its supply chain. a component or manufacturing process at a particular sup-
plier or assembly site; an edge is a directed flow of materials
This approach helps Ford streamline and better tar-
from an upstream stage to a downstream stage. We formu-
get its operational-disruption risk-assessment process, late both models as linear optimization programs. We sum-
deepen its understanding of its disruption risks across marize our notation for the single-tier model in Table A.1
both its internal operations and extended supply and for the multiple-tier model in Table A.2.
Simchi-Levi et al.: Managing Risks and Disruptions in Automotive Supply Chain
Interfaces 45(5), pp. 375–390, © 2015 INFORMS 387
F4n5 Set of production edges under disruption scenario n. A Set of all suppliers sites (plants).
t 4n5 TTR for disruption scenario n. A Set of all nodes produced at supplier and (or) plant .
ci Total production capacity of node i per unit time. V Set of all final nodes (vehicles).
si Finished goods inventory of node i. Vj Set of all final nodes (vehicles) that are of the same type (j).
fj Profit margin of product j. N− 4i5 Set of parts required to produce node i.
dj Demand for j (per time unit). N+ 4i5 Set of nodes that require node i.
lj Lost production volume of vehicle type j. Pjk Set of all nodes that are in the upstream of node j and of part
yij Amount of product j produced at plant i. type k.
t 4n5 TTR for disruption scenario n.
Table A.1: This table lists the parameters and variables of the single-tier ui Total production quantity of nodes i during time t 4n5 .
model and their explanations. lj Lost production volume of vehicle type j.
yij Allocation of upstream node i to downstream node j during
time t 4n5 .
In the ST model, the firm has a set of plants (A), which si Finished goods inventory of node i.
produce a set of products (V). The firm’s objective for each rkj Number of type k parts required to make one unit of node j.
disruption scenario is to minimize the impact of the dis- fj Performance impact (e.g., profit margin) of one unit of
ruption on its chosen performance metric. We capture this product j.
through the following linear program. dj Demand for j per time unit.
X 4n5 ci Production capacity of node i per unit time.
Minimize fj lj
j∈V Table A.2: This table lists the parameters and variables of the multiple-
X 4n5 4n5 4n5 tier model and their explanations.
s.t. yij + lj ≥ dj t − sj 1 ∀j ∈ V
i2 4i1 j5∈F4n5
X 4n5
yij ≤ ci t 4n5 1 ∀i ∈ A to identify the most crucial node of the system, the disrup-
j2 4i1 j5∈F4n5
tion scenarios are constructed as all events that relate to
the removal of a single node from the graph. This is the
4n5 4n5
yij 1 lj ≥ 01 ∀ i ∈ A1 j ∈ V paradigm adopted for the analysis at Ford.
4n5 Although the ST model explicitly captures only the last
In this model, decision variable yij is the cumulative pro-
4n5 tier of the production system, it can be used to analyze
duction of j at plant i in disruption scenario n. Variable lj
a disruption at a supplier in an upstream tier. To do so,
is the amount of lost demand for product j in disruption
4n5 we disrupt the nodes in the final tier that depend on the
scenario n. Parameter fj refers to the impact of one unit
upstream supplier, and solve ST. This is reasonable if the
of loss in sales for product j, for example, the profit margin;
firm has little control over the nodes prior to the last tier
t 4n5 is the TTR for this disruption scenario. dj and sj are the
and if the firm knows which final-tier nodes will be affected
demand and inventory for product j, respectively. Flexibil-
by the disruption. These assumptions may be too conserva-
ity design F4n5 is the set of edges that are still alive during
tive, for example, in situations in which the firm has control
disruption scenario n.
over upstream resource allocation and routing. We present
The objective function is the minimization of the total
weighted loss as a result of the disruption. The first con- a multiple-tier model that addresses this more general case.
straint is a lower-bound constraint for the number of units The MT model is similar to the ST model. We include the
lost for product j, given the production and inventory con- concept of parts, which refers to the set of nodes that are
ditions. The second constraint is a total capacity constraint functionally equivalent in the manufacturing process, but
on the assembly plant i. We can replace the linear objec- potentially processed at a different plant or supplier site.
tive function with a convex one in a more general case, for X
example, accounting for lost market share if the loss exceeds Minimize f j lj
j∈V
a specific threshold.
∀ k ∈ N− 4j51 ∀ j ∈ D
X
Solving one instance of this linear program measures the s.t. uj − yij /rkj ≤ 01
impact of one disruption scenario. A crucial step of using i∈Pjk
uk ≥ dj t 4n5 1
X
lj + ∀j ∈ V as finding a lexicographically optimal solution in multi-
k∈Vj objective optimization (Ehrgott 2005), where the n objectives
correspond to the performance impact under these n TTR
uk ≤ c t 4n5 1
X
∀ ∈ A
values. Using x = 4l/t1 u/t1 y/t5 to represent the resource-
k∈A
allocation strategy, and f 4x1 t5 and 8x Ax ≥ b4t59 as the
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
References https://mitsloan.mit.edu/LearningEdge/CaseDocs/13-149%20
Nissan.Simchi-Levi.pdf.
Banks E (2005) Catastrophic Risk: Analysis and Management (John Schmitt AJ (2011) Strategies for customer service level protection
Wiley & Sons, Chichester, UK). under multi-echelon supply chain disruption risk. Transporta-
Braunscheidel MJ, Suresh NC (2009) The organizational ante- tion Res. Part B 45(8):1266–1283.
cedents of a firm’s supply chain agility for risk mitigation and
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
Professor Simchi-Levi has consulted and collaborated experience in supplier quality, notably directing Automo-
extensively with private and public organizations. He was tive Components Holdings’ Supplier Quality Department
the founder of LogicTools that provided software solutions for five years. Prior to his present assignment, Combs spent
and professional services for supply chain optimization. four years directing material cost and risk management
LogicTools is now part of IBM. activities in North America.
Downloaded from informs.org by [129.93.16.3] on 24 October 2015, at 08:40 . For personal use only, all rights reserved.
William Schmidt is an assistant professor in the Oper- Yao Ge has BS and MS degrees in mechanical engineer-
ations, Technology and Information Management group at ing. He currently leads analytics research at Ford IT. Ge
the Samuel Curtis Johnson Graduate School of Management specializes in distributed data processing, machine learning,
at Cornell University. He received his doctorate from Har- and natural language processing. Previously, Ge worked
vard Business School in 2013, Master of Business Adminis- on large-scale search-based applications, rule-based AI sys-
tration from the University of Florida, and BS in aerospace tems, as well as engine simulation codes. Before joining
engineering from the University of Florida. Before his doc- Ford, Ge worked as product design engineer in automotive
toral studies, Schmidt worked for 12 years as a consultant supplier industry.
and product manager, primarily for Electronic Data Sys- Oleg Gusikhin is a technical leader at Ford Research
tems, FreeMarkets, and Ariba. and Advanced Engineering. He received his PhD from
Yehua Wei is an assistant professor in the decision sci- the St. Petersburg Institute of Informatics and Automa-
ences area at the Duke Fuqua School of Business since 2013. tion of the Russian Academy of Sciences and an MBA
Prior to that, he received his PhD in operations research from the Ross Business School at the University of Michi-
from Massachusetts Institute of Technology. His primary gan. He has worked at Ford Motor Company for over
research interests lie in the area of operations research. More 20 years in different functional areas including informa-
specifically, he is interested in the benefit and design of tion technology, advanced electronics manufacturing, and
limited resource pooling, supply chain risk mitigation, and research and advanced engineering. During his tenure at
decision making under uncertainty. Ford, Dr. Gusikhin has been involved in the design and
Peter Yun Zhang is a PhD candidate in the Institute implementation of advanced information technology and
for Data, Systems, and Society (IDSS), Massachusetts Insti- intelligent controls for manufacturing and vehicle systems.
tute of Technology. Prior to joining IDSS, Peter received Dr. Gusikhin is a recipient of two Henry Ford Technology
his Bachelor of Applied Science in engineering science and Awards, 2009 Institute of Industrial Engineers Transactions
Master of Applied Science in mechanical and industrial Best Application Paper Prize in Scheduling and Logistics,
engineering from the University of Toronto. He is interested and 2014 INFORMS Daniel H. Wagner Prize. He is an indus-
in supply chain risk management, optimization, and more try vice-chair of IFAC Technical Committee “Manufacturing
generally, decision making under uncertainty. Modeling for Management and Control,” and a lecturer in
Keith Combs is Purchasing Director, North American the Industrial and Operations Engineering department at
Engine, Transmissions, and Driveline Components at Ford the University of Michigan.
Motor Company where he is responsible for supply into Michael Sanders brings a wealth of knowledge to Ford’s
all of Ford’s Automatic Transmission, Engine, and Driveline evolving Supply Risk Management team with more than
Plants in North America. His office is located at Rotunda 17 years of purchasing experience. Sanders is responsible
Center in Dearborn, Michigan and he has held this posi- for monitoring, analysis, and communication of potential
tion since June 2014. Keith attended GMI Engineering and Supply Disruption Events for North America, as well as
Management Institute (now Kettering University) and grad- lead for the Global Production Supply Risk Strategy. In 1995
uated with a Bachelor of Science in management in 1987. Sanders earned a degree in engineering arts with a business
He later attended Indiana University and graduated with a cognate from the College of Engineering at Michigan State
Master of Business Administration. University.
Keith was employed by General Motors Corporation’s Don Zhang is a technical expert in Global Data, Insights
Central Foundry Division from 1982 to 1991. He held a and Analytics Team at Ford Motor Company. He has more
number of plant positions within GM’s Defiance, Ohio and than 10 years of experience at Ford in supply chain analytics,
Bedford, Indiana facilities. He started as a cooperative edu- cost optimization, and forecasting. He is currently part of the
cation student and was a manufacturing supervisor when Manufacturing and Purchasing Strategy group to develop
he chose to leave GM to complete his education and ulti- advanced analytic models for businesses. Prior to Ford, he
mately pursue other interests. Combs has been employed worked for U.S. Steel Corporation (1994–2004) where he held
by Ford Motor Company since 1991 and has held a number the positions of senior research engineer, associated research
of positions within the Purchasing organization. His pur- consultant, and research manager in steelmaking and casting
chase responsibilities have included various component technologies. He holds a BE and MS in physical chemistry
purchases (rubber and plastic parts, cast and powder metal of metallurgy from University of Science and Technology
aluminum/steel/iron parts, stampings, resistors, and vari- Beijing. He holds a PhD degree in materials science and
ous other commodities) as well as assemblies such as air bag engineering (1993) and an MBA in operation research and
modules, pumps, and torque converters. Combs also has finance (2003) from Carnegie Mellon University.