General Motors’s Detroit headquarters lies a comfortable 6,400 miles from Japan. Whereas Intel had 600 people working in Japan and depended heavily on Japan’s extensive network of semiconductor suppliers, GM’s ties to the area were much less intensive. A scant 2 percent of its vehicle parts came from Japan, and only 25 of GM’s 18,500 Tier 1 suppliers were in Japan. Yet GM’s experience with the 2011 quake shows just how deeply interconnected companies have become, and how unknown risks can outweigh known ones.
The quake and tsunami struck around 1 o’clock in the morning Detroit time on Friday, March 11. When GM’s executives learned of the disaster in the morning, they were somewhat worried about those 25 Japanese Tier 1 suppliers but did not think that they faced a corporate-wide crisis. GM’s purchasing department worked through the weekend, trying to get information from Japanese suppliers. However, like many other companies, GM had a difficult time reaching these suppliers. Power and telecommunications were down in the affected area. Japanese roads and railroads were closed pending inspection for damage, so workers could not get to suppliers’ factories. By Monday, GM received some initial reports of the severity of the damage and which suppliers were impacted.
By Tuesday, March 15, GM estimated that 30 suppliers and 390 parts were affected by the quake and tsunami. Although 390 parts out of a total of about 30,000 parts for an average car seems minor, a single missing part can prevent completion and shipment of a car. Initial estimates based on available inventories showed that outages of these parts would halt production at many GM assembly plants in only eight days. More ominously, the initial estimate was that by the end of March, all of GM’s factories worldwide would be down. Worse, initial estimates suggested that production might be disrupted for at least seven months. That’s when alarm bells rang throughout GM.
Past crises can help a company develop new skills, and GM was no exception. In 2005, Delphi Corporation, which was a spin-off from GM, declared bankruptcy. To work through the staggering implications of losing its biggest supplier, GM created “Project D.” That effort provided a backbone of people and methods for what became “Project J”—the response to the Japanese quake, tsunami, and nuclear disaster six years later. Bankruptcies and earthquakes may be vastly different causes, but they have similar effects in terms of disrupting critical supplies.
Step one of GM’s method was to create a crisis room. The room was actually three rooms: one for the central coordination of Project J, one for working on supply chain solutions for specific elements of the crisis, and one for working on engineering solutions for affected parts. All three rooms sat in the GM’s futuristic VEC (Vehicle Engineering Center) in Warren, Michigan, 15 miles from GM headquarters at Detroit’s Renaissance Center. The VEC houses GM’s engineering and design functions as well as its supply chain management organization. Other smaller crisis rooms in other GM’s global locations also joined the effort.
With 390 affected parts and all 16 assembly plants at risk of a shutdown within a couple of weeks, GM needed a team of people who were adept at solving tough problems quickly under pressure. That’s where the experience of Project D provided its first benefits. “I knew exactly—I mean literally day one—I knew the people we wanted, because I pulled in people who had worked on Project D,” said Rob Thom, manager, Global Vehicle Engineering Operations.1
Because the crisis was a supply chain disruption, supply chain people took the lead for managing the entire effort. Bill Hurles, executive director of Global Supply Chain, led the effort with a strong collaboration between the supply chain organization and the engineering organization.
The team quickly established a daily routine heavily focused on communication and coordinated action on the highest priority issues. Candid communications helped focus the efforts on the toughest problems. “This is where it’s hard because you don’t want to convey bad information but sometimes you have to. I shared everything, I mean everything,” said Ron Mills, director, GM Components Holding, who served as the key spokesman for the engineering side of GM.2
Each day began with a 6 am call to senior leaders, including Vice Chairman Steve Girsky. “I had every region on the phone; all North America was impacted by this crisis and we’d go through ‘here’s where we are, here’s what we learned, here’s our downtime, here’s our challenge for the day,’” Mills said.
Next, the team defined the day’s activities to reach that day’s goals. Sub-teams had their own meetings, such as a 7:30 am supply chain meeting. By 8 am, the crisis team leaders were rolling out information to all the teams working on the crisis. The crisis team also held a 10:30 am update to the sales, service, and marketing teams.
“Then, at 4:00 pm, we did a follow-up on what we learned for the day and what progress we had made,” said Mills. This call closed the loop on the team’s daily recovery activities. Regular and consistent communications of status, goals, actions, and results kept the group steadily working toward their goal. Given the global nature of the event and the magnitude of the problems, many team members worked around the clock. “It was seven days a week, 24 hours a day that we ran,” Mills said.
Minimizing the impact of the crisis on GM’s production and subsequent sales called for a multifaceted plan of attack. The crisis room teams worked on five basic tasks: identify disrupted parts, assess the risk involved, delay the shortage of parts, reduce any shortages, and optimize production during the disruption. These tasks mainly involved the supply chain, purchasing, engineering, and marketing sides of the organization. And all the tasks had to take place concurrently.
GM’s first task, starting on the day of the quake, was to identify all the affected parts and their impact on GM operations. Although the team immediately knew of some two dozen affected suppliers and 390 affected parts, those were only the direct impacts based on very preliminary estimates.
Thom described the commencement of the recovery effort. “When we had the first meeting, it was like, ‘OK, who are the players and who knows what?’ Well, we didn’t have any information yet. So, they handed us the 390 part numbers, and then an hour later said, ‘Oh, by the way, we’ve got a hundred more.’”3 Disruptions deeper in the supply chain would surface over time, but the sooner GM could detect those hidden problems, the sooner it could solve them.
Modern-day cars are electronic miracles with all manner of sensors, computers, and actuators for managing the engine to produce efficient and reliable power with the lowest possible tailpipe emissions. The cars’ dashboards use computerized displays and touchscreens to provide more control and convenience for drivers and passengers. Although a dashboard assembly or antilock brake module might be made in America by an American Tier 1 supplier, some of the components on the circuit board may have come from Japan. Given the Japanese prowess for electronics manufacturing, it’s no surprise that a big part of the disruption lay in the electronics. All of GM’s cars had computer chips, sensors, displays, radios, and navigation systems made with parts from Japan.
Electronics weren’t the only concern, however. GM soon discovered that almost every type of part on many different vehicles required something from Japan. Xirallic, a sparkly additive in the paint used on the Corvette, came from Japan. Special plastics for the body trim came from Japan. Rubber seals and gaskets came from Japan. High-tech chrome plating on turbochargers came from Japan. Cooling fans, radiator caps, air conditioner compressors, starter motors, and many more parts had some tie to Japanese suppliers.
Like most companies, GM did not have direct business relationships with those deeper-tier suppliers. Each Tier 1 supplier was typically responsible for its own engineering and sourcing decisions from deeper tiers. Furthermore, suppliers consider these relationships to be proprietary and confidential—part of the supplier’s intellectual property—thereby limiting GM’s awareness of how the quake might affect various suppliers’ production and recovery alternatives. Fortunately, GM’s team benefited from some ongoing efforts to map the deeper tiers of the company’s supply chain. Even before the quake, the company knew it was exposed to risks in semiconductors. “We already had a pretty good roadmap, and it helped us navigate our way through,” said Bill Hurles.4 Collaboration with the engineering side of GM helped, too. In many cases, GM’s engineers knew some of the Tier 2 and Tier 3 suppliers because they knew which sensor or chip was used inside a particular subsystem.
The deeper the team dug, however, the more problems they found. Some of GM’s non-Japanese suppliers had Japanese suppliers. And some of GM’s non-Japanese suppliers had other non-Japanese suppliers who had Japanese suppliers. And so on. “The list kept growing. And every day, it went up. It was a moving target for us,” Thom said.5
From the known 390 affected parts on March 14, the number grew to 1,551 parts on March 24, 1,889 on March 29, and to a staggering 5,329 on April 13. During the month after the quake, GM discovered an average of 160 disrupted parts each day. Nor was the problem helped by the ongoing crisis with the Fukushima nuclear plant and the persistent power shortages in Japan.
As the number of affected parts grew into the thousands, tracking each individual part became difficult. The volume of inscrutable part numbers impeded communications and obfuscated the effort. To cope with this, the team switched from tracking individual parts to tracking what GM called “commodities,” which were common categories of parts or subassemblies used on most vehicles, such as a seat, door pad, or radio. The long list of 5,329 affected parts became a more manageable list of 116 affected commodities. And because GM’s purchasing, engineering, and supply base functions are largely organized around commodities, this view made sense in terms of finding the right internal and external people to address each affected commodity.
In the end, it took more than two months to even know how many parts were affected. The final figure of 5,830 affected parts was nearly 15 times higher than the initial estimate of 390 parts associated with the Tier 1 suppliers in Japan. And each missing part raised the specter of halting production somewhere in GM’s system.
From the beginning and as the number of affected parts grew, GM needed a way to track the impact of the developing crisis. It needed a visual “dashboard” to show at a glance which vehicle platforms were affected, which parts were affected, when critical parts would run out, when fresh parts or alternatives might appear, and when each one of its 16 assembly plants might be shut down for lack of parts. Again, the company’s experience with Project D provided the solution.
“We call these white-space charts,” said Mills.6 The solution was a very long, room-spanning whiteboard chart showing the timelines on each of GM’s 16 global assembly plants along a horizontal axis. “This is something that we’ve come up with to show impact. It helps us communicate to our leadership where we are with the problem,” Mills said.
On the left edge of the time axis was the current day and near-term weeks. These near-term weeks were shaded, marked, and annotated to show when shortages of any parts might affect that assembly plant. A circle on the timeline showed when a part would run out for some vehicle options but that GM could “build through” by continuing to produce the other variants of that vehicle. A triangle indicated a potential problem affecting production. And an “X” marked a definite disruption to production. As GM gathered data from suppliers, marks began to pepper the various timelines for each plant on the giant chart.
The timeline for each plant covered further months—out to almost a year on the right-hand-side of the chart. This side of the chart showed when GM expected to restart production through one of three methods: the recovery of the original supplier, bringing alternative suppliers on line, or finding an engineering work-around. It also showed previously scheduled halts in production, such as the traditional midsummer shutdown—the annual suspension of production while the company did maintenance and retooling, switched over to the new model year, and gave workers a vacation.
In the middle of the chart was the namesake white-space—the ominous time gap during which GM would run out of supplies and before it expected to have a solution to the disruption. Very early in the crisis, every one of the 16 rows of the chart had sickening “X’s,” sometimes many, somewhere on the timeline. Those marks showed the date when GM would be forced to shut down production at that plant as a result of parts shortages.
The crisis team shaded in and color-coded the two sides of the chart to reflect progress on managing the gap. Red meant they did not yet have a plan; yellow meant they had a plan but had not implemented it yet; and green meant they were executing the plan. Because different groups were working on the two sides of the chart by extending existing suppliers on the left side and resuming supplies on the right side, the two sides might have different colors. The primary goal of GM’s response to the crisis was to eliminate all white-space. Moreover, as the recovery teams worked, they also aimed to make both sides of the white-space chart turn green to demonstrate confidence about parts supplies and recovery plans.
GM staffers also used this same color notation to talk about the status of different commodities: “We’re red on paint. We’re yellow on heated seat modules.” By April 13, of the 116 commodities known to be affected, GM was executing recovery plans for 44 (status: green), had plans for another 61 (status: yellow) and only had 11 “red” commodities. By May 27, the scope of the problems had grown from 116 to 118 commodities, but the engineers and Japan’s suppliers had made significant progress in finding alternatives or in restoring production. The number of commodity groups marked as “red”—a serious problem—dropped from 11 groups to only 2. And the number of commodities deemed to be “green” grew from 44 to 82.
“The first couple of weeks were kind of white knuckle time,” said General Motors Chairman & CEO Daniel Akerson.7 With each bit of news of another part disruption, the team quickly re-estimated when GM would run out and which vehicles would be affected by that new shortage. Those estimates added more “X” marks on the left side of the white-space chart and defined that boundary of the white-space.
The supply chain team worked hard to find any extra inventory of the affected parts that might be available—in an undamaged warehouse in Japan, in transit to GM, in a contract subassembly maker’s site, and so forth. The team searched up and down the supply chain. They even tapped into the replacement parts inventories in the field and at dealers’ locations. Each time they found more parts, they were able to push critical “X” marks a little further into the future and shrink the white-space on that side.
These delaying tactics worked, and the left edge of the white-space moved further and further into the future. The first assessment of the crisis on March 14 estimated that all plants would shut down by the end of March—only two weeks away. By March 24, the team had found enough supplies to keep all the plants running until April 11. By the end of March, the shutdown had been pushed to May 16—providing more than six weeks for finding other solutions to the toughest problems.
At the same time that the supply chain professionals of GM worked to extend supplies to delay the shutdown, the engineering side of GM looked for ways to recover parts’ supplies and production as quickly as possible. To reduce the time-to-recovery, both procurement and engineering helped with the recovery of the affected suppliers and, at the same time, looked for substitute parts. Part substitutions involved a double search both for new suppliers and for ways to adapt well-stocked parts to cover the shortages in parts affected by the Japanese quake.
For example, Renesas Electronics, maker of 40 percent of the world’s automotive microcontrollers, was severely hit by the disaster. Fortunately, General Motors CEO Dan Akerson had been on the board of directors of Freescale, another supplier of these kinds of chips. “So I picked up the phone, I called the CEO of Freescale and I said, ‘I know you make chips of this type.’ We came up with a solution,” Akerson said. At the same time that GM was seeking second sources, the company also had supplier quality engineers in Japan helping suppliers recover and restart production. “So we have a two-track solution if this becomes a problem for us,” Akerson concluded.8
Yet finding an alternative part meant more than just slapping a different part into the vehicle. Under normal circumstances, engineers take six to twelve months to qualify and validate a new part or new supplier. In building cars designed to last for years and run hundreds of thousands of miles, every part must be carefully checked to ensure it will perform its job effectively, safely, and reliably. Validating a new part requires checking that part’s ability to withstand the rigors of years of expected life on the road in terms of heat, cold, mechanical stress, and exposure to gasoline, motor oil, antifreeze, and so forth. The size, weight, and materials of the proposed substitute must be compatible with the other parts and materials already on the vehicle. Finally, engineers need to make sure each part is compatible with manufacturing (e.g., if assembly workers must bend a substitute hose to maneuver it into place, the chosen hose won’t be damaged).
Engineers also had to make sure that any change they made didn’t adversely affect the rest of the vehicle. For instance, the weight of the latch in the hood affects the required strength of the spring needed to hold the hood open. Substitute a heavier latch, and the hood needs a heavier spring. But then the weight of the hood assembly affects the front suspension, and so on. An important part of the qualification process was the effort to minimize the threat of cascading changes. This effort was complicated by the fact that alternatives for several parts were considered simultaneously.
Recovery also meant that suppliers had to work to create capacity for new or changed parts. Vehicles depend on a great many custom-made parts that fit a very limited set of vehicles. Suppliers often use specialized tooling to make complexly-shaped parts such as seals, gaskets, floor mats, rubber boots, and plastic trim, as well as stamped, forged, or cast metal parts. And any visible parts on the exterior or interior need to match in color. For these reasons, qualifying a part from a new supplier, or even just a new plant of an existing supplier, takes time.
“We still have issues, and the issues we have now are getting tougher to solve,” said Robert E. Socia, GM’s vice president for global purchasing and supply chain in mid-May.9 Despite the heroic efforts to find inventories, alternate suppliers, and alternative engineering solutions, GM could not totally close the gap. “Now the long tent pole appears to be semiconductors,” said Akerson.10 The specialized nature of chips, the extent of the damage, and the long recovery times meant that some white-space remained. Because an unsolvable white-space gap had been a looming threat from the beginning, GM had thought carefully about what to do if it could not eliminate all of the white-space.
In many cases, parts such as engine controllers, mass airflow sensors, and brake control modules were shared across multiple vehicles. As Bill Hurles noted, “a lot of the stuff we were working with, you could allocate. We could make a decision to build and use that same engine controller in multiple products in Europe, multiple products in China, and multiple products in North America.”11
“Our goal was to keep them all,” Hurles continued, “but we set up a protocol if we got into a situation and needed to make a prioritization.” First, and from the beginning, GM prioritized all the assembly plants on the white-space chart using a proprietary measure of each vehicle’s financial contribution to the company. The rows of the chart listed the plants in financial priority order. This ensured that the crisis team focused its efforts on the vehicles that could help GM sustain itself if the disruption proved to be worse than expected.
Second, GM also factored in the operating stock levels in the field. How many cars, in terms of days of sales, did it have on dealers’ lots, and was that number lower or higher than planned? Stopping production on a vehicle that had a low number of days in inventory raised the risk of lost sales. “So the stock level would kind of determine the relative importance of vehicles,” explained Bob Glubzinski, manager of North American scheduling and order fulfillment at GM.12
On March 21, concern about a very near-term shortage of airflow sensors led GM to idle production of Chevrolet Colorado small trucks in Shreveport, Louisiana, for one week. Although this step made headlines, it did not influence sales because GM and its dealers had sufficient stocks of finished vehicles. GM took this step because managers knew that the sooner they idled a low-priority plant, the longer they’d be able to guarantee the supply of airflow sensors to plants making full-size trucks that had both a larger profit margin and lower field inventory of finished vehicles. As it turned out, the closing was not even needed; GM subsequently found it had enough sensors to supply all its truck plants, and Shreveport restarted the following week.
In the end, GM weathered the crisis well. From the perspective of the average car buyer, almost nothing happened. Dealer lots still offered plenty of cars, although a few colors of some models weren’t available for a time. Inside GM, it was a major event and hundreds of people put in very long hours to insulate dealers and customers from the disaster half a world away. As one GM staffer joked, “My neighbor thought I was divorced because I wasn’t home for nine weeks.”
The success of GM’s efforts illustrates an important lesson in coping with large supply chain disruptions. When dealing with complex manufacturing operations, team members should stick to their roles and expertise in order to avoid impeding progress and minimize extra needless work. It also highlighted the role of senior management.
“Because this was such a visible crisis, everyone was trying to be a hero in their own function,” said GM’s Dr. Marc Robinson, assistant director & economist, enterprise risk management.13 One of the toughest challenges for GM was in reining in the good intentions of everyone’s attempts to find solutions to the many different part shortages. GM employs a broad staff of veterans with decades of experience and extreme loyalty to the company. They knew their cars and factories inside and out. That meant that everyone had lots of ideas on how to tackle the various shortages.
Yet people’s well-intentioned interventions could easily create other problems. In the same way that changing a part has cascading engineering implications on the engineering of other parts, changing a manufacturing plan has cascading implications on other plans and on the supply chain. As an example, consider heated seats, which faced a disruption in their electronic control modules. “We received a lot of pressure from engineering to stop ordering these seats for vehicles yet-to-be-manufactured,” Glubzinski said.14 But GM wanted to avoid three types of impacts that this intuitive and well-meaning change to the company’s mix of products could create.
The first cascading impact of changing the mix was a potential capacity issue. Hurles explained the chain of logic: “I can build vehicles without heated seats. But now that starts shifting the mix because almost all heated seats go with leather. If I go out of heated seats, I now go into cloth. Now I can create an issue with cloth.”15 Moreover, the cloth vs. leather mix affects other vehicle features linked to selling basic vs. sport vs. luxury variants of a given model. So one small change could have significant impacts on the volumes of other parts and create shortages in other parts that were not disrupted. “We could move ourselves right into a hurricane,” Hurles concluded.
The second cascading impact was caused by the sheer complexity of GM’s supply chain and manufacturing systems. Building cars involves a carefully choreographed, weeks-long pipeline of coordinated production activities around the globe. Canceling heated seats means that all subassemblies and components that were destined to go into vehicles with heated leather seats become stranded somewhere in the supply chain. “And the supply chain, which was already in a very delicate state, could not survive more change,” Glubzinski said.
Finally, the third impact came on the sales side. Dealers and consumers may not want more cloth seat vehicles. And the lack of heated seats might put GM at a competitive disadvantage in the marketplace relative to other car makers who could offer heated seats. “We want to provide the dealers the product that’s going to turn. They know what they want. They know what they can sell. We want to supply it to the best ability that we can,” Glubzinski said. “All of our efforts were to keep building every product every day to the plan, trying to minimize changes to the mix,” Hurles added.
Hurles continued, “I still remember the night I called John Calabrese [vice president, Global Vehicle Engineering] and Ron Mills and just said, ‘guys, I need your help. I’m seeing people starting to make decisions to shift mix. I can’t have them do that. We don’t need to do that. It’s a good intention, but if they start removing part A, I’m going to have a problem with part B. So I need you to make it very clear, they’re not to make a change without our authorization.’”
This battle to limit unintended consequences from well-intentioned interventions led to a mantra: “Stay in your swim lane.” Although the crisis team wanted creative solutions from everyone, they didn’t want everyone implementing independent decisions that affected other parts of the organization and created more problems elsewhere. GM had to balance between people having the flexibility to solve the problem any way they could and the discipline not to disrupt the functioning of the rest of the company.
From the beginning, the team had strong support from GM’s executives. And, possibly, the biggest contribution of senior management was not to get involved in the details. Steve Girsky, vice chairman of General Motors told the war room team, “You are the brain trust. Just tell me what you need; otherwise I’m staying out of the way.” And the executives meant it. One team member was very pleasantly surprised to notice that the “execs were going to Jimmy Johns [a local sandwich shop] to pick up food so that we [the crisis team] could keep working.”16
That support from top management grew stronger during the crisis as the team steadily found solutions to almost every element of the disruption. “Here was this life-threatening crisis that the team showed they could manage, and it gave the leadership confidence. The leadership didn’t need to tour the crisis room again in later crises,” Hurles said.
GM found that this crisis, like others, helped build bench strength by revealing people who had the required combination of the right psychological profile and the right skills. In the wake the 2011 Japan earthquake, GM’s Mills said, “I know who to call in the organization, who works well in crisis, who has the right skill set, who has the stamina to survive in the environment, and who also has know-how to work within their function and across the functions.” And he added, “You can’t just have an engineer who knows a steering system. You’ve got to have an engineer who knows the steering system and knows the supply base and knows the competitive alternatives to your solutions. You’ve got to have some breadth and depth. I know who those guys are in the organization, which helps.”
Many companies used similar strategies and had similar experiences during the Japan quake and during other disruptions. Delphi noted that it faced a five-day information blackout during the Japanese quake and tsunami that made assessing the situation difficult. Intel also had to contend with the time required for engineering validation of alternatives involving complex products with intricate interactions. Yet other companies used somewhat different response tactics to handle the multiple tasks demanded in a disaster response.
In October 2011, a massive wave of flooding reached Bangkok, Thailand, inundating the country’s industrial parks. Over 1,000 factories were flooded, including many suppliers to the computer and electronics industries (see chapter 7). Among the affected companies was Flextronics, the large global contract manufacturer headquartered in Singapore. Flextronics knew that it bought some 2,000 different electronic components from dozens of suppliers that might be in the affected area. But with little information coming from the flooded area, Flextronics couldn’t know exactly which parts were actually affected or how best to respond.
To focus its efforts in the midst of unresolvable high uncertainty, Flextronics defined four mutually exclusive, collectively exhaustive categories of parts. The “green” category included well-stocked parts with more than three months of inventory. These were parts that were not likely to cause production and customer shortages problems because they were not likely to run out. This large supply of inventory likely exceeded the time-to-recovery, or, in GM’s vernacular, these parts had no white-space. The “yellow” category comprised dual-sourced parts, which were less likely to be disrupted because Flextronics had access to supplies from its (already qualified) second supplier. Contacting the second source would presumably ensure a fast recovery.
The last two categories covered single-sourced and low-inventory parts—ones that had a high likelihood of causing supply disruptions. These parts were prioritized by revenue impact. The “orange” category parts were those feeding into lower-revenue business activities. And the “red” category parts were the high-revenue-impact parts.
Overall, this categorization let Flextronics winnow down the daunting list of 2,000 potentially disrupted parts to a short list of about 100 high-priority parts. The team then identified recovery strategies, such as placing risk buys, qualifying alternate sources, or adjusting build schedules and allocating available inventory to the higher priority products. Using this information, the team was in a position to have targeted conversations with suppliers and customers about the 100 parts that were on the “high risk” list. Categorization by sourcing, inventory, and revenue impact was critical to ensure that the parts with the highest risk were addressed first.
GM’s approach to assessing the white-space provides a general framework for companies to roughly estimate the impact of future disruptions. The white-space duration of a potential disruption in days or weeks can be converted into a financial number—the value-at-risk (VaR)—by multiplying the white-space duration of disrupted production (or sales) in days times the daily impact (e.g., revenues or profits) of a day’s worth of production (or sales). Although the actual VaR may vary as a result of the many nuances of disruptions and the company’s response, this VaR estimate can provide a starting point for quantifying impact. Estimating VaR requires the terms described in the four sections below.
The VaR calculated below is an effects-focused view on risk in that it is an estimate of the financial effects of a disruption—regardless of the cause—on the company. It is a conditional estimate of loss, rather than an expected value of loss: if disruption occurs, then VaR may be the impact. Calculating the VaR for multiple types of disruptions helps companies to prioritize proactive risk mitigation efforts or reactive recovery efforts during crisis response.
The first factor for estimating value-at-risk is defined by the left edge of the white-space, which is the lag between the disruptive event at its point of occurrence (e.g., the earthquake) and the disruption of production, sales, or deliveries17 for the company. Inventories—often measured as days of supply (DOS)—between the source of the disrupted part and the customer enable the company to maintain operations for some period of time after the disruptive event occurs. This number represents the inventory anywhere in the supply chain where inventories might lurk, such as suppliers’ warehouses, safety stocks of inbound parts, work-in-process, and finished goods. These inventories delay the time when a product’s disrupted part becomes unavailable and consequently that product cannot be made and shipped to customers. Each added day of inventory means one fewer day of lost production, sales (and profits). The duration of normal operations that these inventories offer before customers are affected is the time-to-impact (TTI).
The second factor is the anticipated time-to-recovery (TTR), which is the lag between when the disruptive event occurs and when the company can restart normal production. TTR is the point marked by the right-hand edge of the white-space in GM’s chart. As mentioned in the “Task 4: Reduce Time-to-Recovery” section, the TTR is the earliest of several durations. These include the duration of the recovery efforts to restart production and resume deliveries at the disrupted supplier; the duration of procurement and engineering processes to find, qualify, buy, and use parts from a second source; and the duration of product and manufacturing reengineering processes to use other types of available parts or capacities. TTR also includes any transportation lead-time, which may be expedited.
The concept of TTR raises an issue: what counts as recovery? Often, production from a restored supplier or new second source will take time to ramp from zero volume to full production. For example, automotive chip maker Renesas announced a series of expected recovery levels with different recovery times as it rebuilt after the 2011 quake: a 10 percent capacity resumption at 12 weeks, 35 percent at 16 weeks, 55 percent at 20 weeks, 75 percent at 24 weeks, and 100 percent at 28 weeks.18 Later, the company announced an accelerated recovery schedule that reduced the 100 percent recovery point from 28 weeks to 24 weeks.19 Using new or existing second-source suppliers has similar issues with the lead-time for ramping to volume production or bringing more capacity online to deliver replacement supplies for the disrupted part.
Different companies might use different TTR thresholds to manage risks. For example, Cisco uses the 100 percent TTR definition, which is conservative and may over-estimate the impact because it’s highly likely that a disrupted supplier would resume partial production before it recovers 100 percent. In contrast, Medtronic assesses the 50 percent, 90 percent, and 100 percent TTR points to model the ramp of the recovery. Moreover, to the extent that the affected company can prioritize the use of partial supplies, it can further reduce the impact by allocating the limited supplies to its most profitable or important product lines and customers—a 50 percent recovery in supplies is then likely to produce more than a 50 percent recovery in sales or profits (see the section titled “Mitigating the VaR”).
The gap between the estimated time-to-impact (when inventories run out) and the estimated time-to-recovery (when production can resume) is the estimated customer impact time (CIT). In other words, CIT = TTR – TTI. For GM, this was the white-space or production down time, when GM would not be able to fulfill dealers’ orders. For other companies, CIT is the time during which they will not be able to fulfill customers’ orders.
The third factor for estimating the value-at-risk for a given disrupted part or facility is the financial impact of the disrupted element to the company. To estimate the daily financial impact, one uses Bill of Materials (BOM) data to identify all the products that use the disrupted part and then uses Enterprise Resource Planning (ERP) order data to estimate the daily financial impact on different channels, products or customers.
The financial contribution per day of each disrupted product can be calculated from the financial contribution per unit of the product sold and the number of units sold per day. The daily financial contribution is then the sum of the financial contributions per day of all the products that use the disrupted part (accounting also for the number of such parts used in each product). It can be based on revenue (e.g., sales/day), gross profit, net profit, or some other measure of the financial outcome per day. Naturally, the contribution per unit of product may be different across customers as well.
As mentioned above with respect to the VaR, the financial impact is just a rough initial estimate that assumes demand does not change, that the disruption has no other side effects on customer relations, and that the company serves customers fairly.20 A company’s response actions (see the section titled “Mitigating the VaR”), customer demand patterns, customer importance, parts’ interactions, competitors’ capacities and behaviors, and many other factors mean that during a disruption the actual impact may be different.
In many disruptions, parts production does not cease entirely. Second sources, undamaged capacity at the disrupted supplier, and alternative parts can be used to provide partial parts supply for the duration of the TTR and allow some partial rate of production during the CIT. In some cases, predisruption inventory plus partial supply may suffice to satisfy demand during the entire time-to-recovery.21 However, once existing inventories run out (assuming that both finished product inventories and part inventories will be used to satisfy the full demand as long as possible), then the partial supply of parts implies that the company can only satisfy a portion of the demand until the TTR is reached.22
The impact of the lost sales during the CIT depends on the unsatisfied sales volume, which in turn depends on the level of missing parts (calculated as one minus the partial supply). Thus, if the partial supply can cover 25 percent of the normal volume, the fraction of missing parts would be 75 percent and, as a rough approximation, the company would be losing 75 percent of its usual revenue or profits during the CIT. To a first approximation, the VaR of a disrupted part or ingredient under conditions of partial supply is the total daily impact multiplied by the CIT and by the fraction of missing parts.
A simplistic estimate of the value-at-risk can overestimate the potential impact of a disruption because companies have many tactics for minimizing the financial impact of a disruption, such as: preferential allocation, auctions, dilution, and substitution. Each tactic can mitigate the VaR by increasing the amount of demand satisfied using the available disrupted supplies or by increasing the financial returns from those disrupted supplies.
Both GM during the 2011 Japan earthquake and Flextronics during the 2011 Thai floods allocated scarce parts to those products or customer segments with the highest financial performance. GM used a confidential profit margin metric for its allocations during Project J as well as finished product inventory levels in the field. Flextronics based its allocation decisions on value-at-risk calculations but adjusted the decisions because the different product lines represented different contract manufacturing clients. Issues such as profitability, the overall value of the client to Flextronics, and the impact on the client were taken into account.
Customer size matters. During the Thai floods and the disruption of the hard disk industry, large PC makers such as HP, Dell, and Apple were “highest on the priority list to get the products,” said Bob O’Donnell, program vice president at IDC, the market intelligence company.23 The top makers were followed by PC original design manufacturers (ODMs), while channel retailers were last.24Yet it is not always just the large customers who get parts in short supply. Verifone’s senior vice president of global supply chain, Patrick McGivern, commented that tiny customers might get good allocations because their volumes are very small and suppliers may be sympathetic to the plight of the smallest customers who may have fewer options and may face an existential threat from a significant disruption.
Of course, suppliers are likely to use allocation, too. Flextronics’s head of procurement and chief supply chain officer, Tom Linton, said that during the first two days after the 2011 Japan quake, everyone was frantic to find out what happened; then it was a mad rush to lock in suppliers. Some suppliers allocated “fairly” (that is, they did not commit their limited supplies to those who called them first or to the best customers only) and others didn’t. When floods in Thailand devastated some hard disk drive makers, Stephen J. Luczo, CEO of Seagate, said, “It’s going to be very interesting to see who gets drives and who doesn’t.”25 Financial and marketing considerations, as well as company culture and regulations, can affect how companies decide which customers to serve and how.
Preferentially allocating scarce supplies can mitigate the financial impact of a disruption by making only those products that get the highest financial return from the fewest number of parts. That allocation depends on both the financial return on each unit of product and the number (or amount) of the scarce part used to make that product. For example, a product with a financial contribution of $1 that requires only one part of the disrupted type to build, should be preferred to a product with a higher financial contribution—say $2, but which requires five units of parts to build. Thus, for the simple case of a single disrupted part, products could be ranked by their financial contribution per part and orders for higher-ranked products could be fulfilled ahead of orders of lower-ranked products.
The feasibility of preferential allocation depends on three primary prerequisites. The first is the availability of some parts via inventories or partial supply to support some rate of production and sales. The second is that the products’ financial contribution per unit of the scarce part must vary across products or customers. If all products and customers have the same financial returns per unit of scarce part, then preferential allocation offers no benefits. The third prerequisite is that the company must have the regulatory freedom, contractual flexibility, and cultural willingness to sacrifice low-performing products (and customers).
If the company can use preferential allocation, then the allocation-mitigated VaR can, in theory, be estimated via a formal optimization process that allocates units of each partially available part to the various products so as to minimize the VaR. The result is a mitigated VaR, which is smaller than the VaR under fair allocation, in which each product or customer is given the same partial fraction of demand.26
The optimization needed to calculate the mitigated VaR can grow quite complex if the disruption affects multiple parts and multiple products. During an actual supply disruption, such an optimization is not practical because of the time constraints, changing dynamics, and ongoing uncertainty. As demonstrated by GM’s Project J and Flextronics’s actions during the Thailand floods, companies rely in such cases on their knowledge of engineering, supply chain, and customers, using heuristics and inventiveness to mitigate the VaR.
Even in a planning mode, when the objective is to prioritize risky parts (and suppliers), formal optimization for risk prioritization may not be useful given the large number of parts and products at even medium-sized manufacturers. More important, formal optimization may not be useful given the uncertainties and unpredictable dynamics in the levels of demand, pricing, inventory, time-to-recovery, and partial supply levels. For purposes of risk prioritization, the mitigated VaR can be estimated numerically by simulating a wide range of scenarios and modeling the mitigated VaR as a function of key parameters.
The most important parameter is the variability of the financial contribution per unit part across all the products using the disrupted part. A set of one-time simulations (with an embedded optimization to calculate the mitigated VaR for each realization) can give a company the data to estimate an approximate mitigated VaR resulting from a disrupted part or a combination of parts. The mitigated VaR estimate for various potential disruptions can then be expressed as a function of the variability of financial contributions, as well as the number of disrupted parts per each product, expected sales volumes, inventory levels, partial supplies, and CIT.27
Preferential allocation will have no effect if all the products using a disrupted part have the same financial contribution, but will have a large effect if the variation is high. The greater the differences in financial outcomes between high-value and low-value products, the more the company can mitigate VaR by preferentially making only high-value products or serving only high-value customers. The impact of preferential allocation will be especially high (per available part) in the case of low partial supplies because if the company has only a scant supply and allocates it to the highest of the high-value products, then that will have a higher mitigation effect than if the company has greater partial supply and is producing more of the average-valued products.
In the wake of the Thai floods and the decimation of Western Digital’s disk drive production capacity, Seagate temporarily took the #1 disk drive maker’s crown from its disrupted rival. Seagate also took a rather unusual approach to allocation—auctioning some disk drives to the highest bidder.28 “In addition to making drives available for those qualified customers not covered by an LTA (long-term agreement) or where LTA volumes are not sufficient to cover their needs, this will allow us to fully understand and gauge marginal pricing,” said William David Mosley, Seagate’s chief operating officer.29 Seagate’s move also prompted more customers to sign LTAs to avoid potential price spikes that might result from an auction.30
But the auction approach may have affected customer relations. “Seagate, where they are playing the auction approach—whoever is going to pay the most money is going to get the product—has left a real bad taste in many OEM’s mouths,” explained a source within the industry.31 “OEMs are doing business with (Seagate) because they have to right now. Once everything is back to normal, I feel that they are going to lose a lot of market share for that.”32 Tellingly, when Western Digital recovered its capacity, it also retook the lead.33
Auctioning a scarce commodity after a disruption might seem extortionate, but a well-designed auction actually improves economic efficiency by allocating the resource to those who can create the greatest value with that resource. Moreover, a high post-disruption price encourages those companies who have ways of foregoing the scarce commodity to deploy their flexibility or substitute options, thereby leaving more of the scarce commodity for those who have no such options. In fact, government auctions are justified in many cases on this characteristic of auction mechanisms. For example, in the opening letter of the information package for the US broadband personal communications service auctions, Reed Hundt, chairman of the US Federal Communications Commission wrote: “I am confident that the auction method we have chosen will put the spectrum in the hands of those who most highly value it and who have the best ideas for its use.”34
The response of some of Seagate’s corporate customers to the auction demonstrates that this approach should be used with care and probably never in consumer markets, where social networks and “equality activists” can hurt a company’s reputation. Using an auction during a disruption—despite its theoretical appeal—reeks of profiteering.
Other companies insist on fair allocations for commercial, cultural, or legal reasons. The most common fair allocation gives every customer the same fraction of their orders. Joe McBeth, vice president of global supply chain management at Jabil, noted that many Japanese companies employed strict fair allocation schemes after the 2011 earthquake. Intel said that as a large supplier in the PC industry, it generally uses a fair allocation approach to avoid the appearance of favoritism.
Yet being fair isn’t easy, especially if customers try to game the system by asking for large orders knowing that they may get only a fraction of what they ordered. Customers could also over-order in an attempt to hoard supplies if they anticipate future shortages. Naturally, over-ordering makes it difficult for a company to serve all its customers. To avoid these difficulties, many suppliers used an allocation formula based on historical sales level.
Continental Teves Inc., a supplier of automotive, industrial, and agricultural products, had to make tough decisions when 9/11 shut down all US airfreight traffic and disrupted cross-border freight flows between the United States and both Canada and Mexico. On the afternoon of 9/11, the company assembled a list of all outstanding orders from customers and to suppliers. Most important, it collected data on its North American customers’ inventory levels. Knowing these customers’ production rates from past order patterns, Continental Teves calculated the number of days of parts supplies each customer had before its operations would run out of parts. This statistic—days of supply—was the one on which Continental Teves based its fair allocation, trying to ensure that all its customers had the same days of supply.
Another approach is dilution—using less of a key raw material when formulating a product in order to extend partial supplies. Dilution reduces VaR by extending both the DOS of existing inventory as well as partial supplies. Chapter 1 described how Intel successfully used this strategy for stretching disrupted supplies of some specialized chemicals after the Japan earthquake. But whereas a properly reengineered and qualified manufacturing process might tolerate dilution, customers might not be as accepting.
In February 2013, premium bourbon distiller Maker’s Mark faced a shortage of bourbon. “Fact is, demand for our bourbon is exceeding our ability to make it, which means we’re running very low on supply,” wrote Rob Samuels, the company COO and grandson of the founder.35 Choosing not to raise prices, the distiller found a way to add a “touch more water” without affecting the taste—diluting the alcohol content of its product from its historic level of 45 percent, or 90 proof, to 42 percent, or 84 proof. “This will enable us to maintain the same taste profile and increase our limited supply so there is enough Maker’s Mark to go around, while we continue to expand the distillery and increase our production capacity,” wrote Rob Samuels in a letter to customers.36 The action, however, did not sit well with customers, as described in chapter 4.
Another mitigating factor is substitution by customers of an alternate product sold by the same firm. When General Mills estimates the impact of ingredient shortages, it knows that a shortage of one type of General Mills breakfast cereal does not imply that the consumer stops buying cereal for breakfast. Instead, consumers may buy a different flavor, which is often another General Mills product. To the extent that consumers substitute a different flavor of General Mills cereal, General Mills does not lose any sales, which reduces the potential impact. The potential amount of mitigation provided by substitution depends on the percentage of customers who will substitute another product offered by the affected firm, the efforts of the company to promote substitute products, the difference in the financial contribution of the original versus the substituted product (after accounting for special promotional costs), and the availability of spare capacity to satisfy the incremental demand for the substitute product. Substitution reduces VaR. The availability of substitutes and an estimate of the extent of substitutability should be taken into account when calculating the impact of other mitigation methods such as auctions, dilution, and preferential allocation.