Accelerating the Proﬁtability of Hewlett-Packard’s Supply ... · Billington et al.: Accelerating the Proﬁtability of Hewlett-Packard’s Supply Chains 60 Interfaces 34(1), pp

Vol. 34, No. 1, January–February 2004, pp. 59–72issn 0092-2102 �eissn 1526-551X �04 �3401 �0059

informs ®

doi 10.1287/inte.0103.0054© 2004 INFORMS

Accelerating the Profitability of Hewlett-Packard’sSupply Chains

Corey BillingtonGlobal Procurement Services, Hewlett-Packard Company, 3000 Hanover Street, Palo Alto, California 94304-1185,

[email protected]

Gianpaolo CallioniStrategic Planning and Modeling, Hewlett-Packard Company, 3000 Hanover Street, Palo Alto, California 94304-1185,

[email protected]

Barrett CraneDigital Imaging, Hewlett-Packard Company, 3404 East Harmony Road, Fort Collins, Colorado 80528, [email protected]

John D. RuarkOptiant, Inc., 4 Van de Graaff Drive, Burlington, Massachusetts 01803, [email protected]

Julie Unruh Rapp, Trace WhiteInkjet Supplies, Hewlett-Packard Company, 1000 NE Circle Boulevard, Corvallis, Oregon 97330

{[email protected], [email protected]}

Sean P. WillemsSchool of Management, Boston University, 595 Commonwealth Avenue, Boston, Massachusetts 02215, [email protected]

Hewlett-Packard (HP) developed a standard and common process for analysis coupled with advancement ininventory optimization techniques to invent a new and robust way to design supply-chain networks. This newmethodology piloted by HP’s Digital Imaging division has received sponsorship from HP’s Executive Supply-Chain Council and is now being deployed across the entire company. As of May 2003, a dozen product lineshave been exposed to this methodology, with four product lines already integrating this process into both theconfiguration of their new-product supply chains and the improvement of existing-product supply chains. Theteam will highlight the application of these new capabilities within HP’s Digital Camera and Inkjet Suppliesbusinesses. The realized savings from these first two projects exceeds $130 million.

Key words : industries: computer, electronic; manufacturing: supply-chain management.

In 1939, the Hewlett-Packard Company (HP) intro-duced its first product, an audio oscillator, which

it sold to Walt Disney Studios for creating someof the special sound effects for the movie Fantasia.In May 2002, HP merged with Compaq, with thegoal of becoming the world’s leading provider ofinformation technology (IT) solutions. By the endof 2002, HP’s revenues were $72.3 billion. Today,HP operates in over 170 countries and is a leadingprovider of fault-tolerant servers, Windows servers,Linux servers, storage systems, management soft-ware, personal computers, pocket PCs, inkjet printers,multifunction printers, laserjet printers, flatbed scan-ners, printing services, and IT services.

To maintain its position as an industry leader, HPmust keep pace with constant advances in technol-ogy in a highly competitive marketplace. To do so, itintroduces thousands of new products every year. Theolder units lose value so quickly and competition is

so fierce that it must properly design, configure, andoptimize the supply chain wherever possible.

Cost factors, such as material devaluation, scrapcosts, write-offs, and fire-sale discounts, have becomethe single biggest detriment to profitability, some-times eclipsing the already-slender profit margins. Tocalculate these inventory-driven costs, HP must con-sider service levels, demand and supplier uncertainty,and end-to-end process times across the entire supplychain. The stochastic nature of many of these factorsmakes this problem very challenging.

Throughout the 1990s, HP tackled the problemof first determining and then minimizing inventory-driven costs through a combination of homegrownspreadsheets coupled with the modeling expertiseof HP’s internal Strategic Planning and Modelinggroup (SPaM). Managers and planners from differentHP organizations commonly conducted supply-chainanalyses on their own by building models in Excel.They would then share their models with each other

59

Billington et al.: Accelerating the Profitability of Hewlett-Packard’s Supply Chains60 Interfaces 34(1), pp. 59–72, © 2004 INFORMS

and spend considerable time trying to get their peersacross the supply chain to buy into their analyses.

With 141,000 employees in 178 countries who coor-dinate their operations with a global network of sup-pliers, HP cannot rely on a few experts to make goodsupply-chain analysis decisions. For some time, SPaMhad been searching for a robust methodology thatwould enhance business units’ ability to conduct end-to-end modeling of the supply chain. Over the yearsit had developed robust methodologies and analy-sis techniques to solve complex supply-chain prob-lems. However, the sophistication of the modelingoften made SPaM a bottleneck in the process of mak-ing supply-chain decisions. Although the process andmethodology were clearly transferable, SPaM couldnot find a modeling tool that it could easily trans-fer to the business units. More than 60 percent of thegroup’s resources were allocated to network analy-ses, but demand was still not satisfied. In addition,because so many of its resources went to supply-chainanalysis, SPaM could not focus on complex prob-lems outside the supply-chain domain. HP needed anapplication that wouldn’t require business users tobecome expert modelers. It had to have the supply-chain theory packaged so that users could analyzeand model the key supply-chain parameters througha controlled process. It needed an application it coulddistribute to users in all business units and regionsso that they would all have access to the same set oftools. SPaM was monitoring supply-chain advancesfrom academia to find such a solution.

In 2000, HP began working with Optiant, a Boston-based developer of supply-chain design software, tobring HP’s long-standing tradition of supply-chaininnovation to a new and broader level. Optiant’sproduct, PowerChain Inventory, combined a robustmultiechelon inventory-optimization engine with aneasy-to-use Web-based architecture and interface.PowerChain grew out of work originally conductedat the Massachusetts Institute of Technology (MIT)(Graves and Willems 2000), continued at MIT andBoston University (Graves and Willems 2002), and atOptiant (Humair and Willems 2003). A notable fea-ture of the software was its drag-and-drop interface,which allowed users to quickly create and optimizeany network topology.

We describe both the process HP developed fordetermining total supply-chain cost, and the supply-chain theory that we developed to solve these real-world supply-chain problems. We highlight the firsttwo applications of this methodology inside HP: theinitial pilot project completed in 2001 by the DigitalCamera business and the initial application conductedby Inkjet Supplies. For just these two product linesalone, HP reduced total supply-chain costs by over$130 million while maintaining already-high servicelevels.

Optimizing the DigitalCamera BusinessDigital imaging is a key growth area for HP. HP’sDigital Imaging organization produces an array ofimaging products. Consumer offerings include digitalcameras, flatbed scanners, and photo inkjet printers.HP’s commercial products include Phogenix photo-processing equipment and Indigo digital presses. TheDigital Imaging group develops complete imagingsolutions for its customers, everything from imagecapture to image output, and all of the software andinfrastructure connecting the two.

HP entered the digital camera market in 1997 witha single camera. Today, it sells over half a dozen cam-eras, ranging from aggressively priced autofocus cam-eras to high-optical-zoom, high-megapixel products.The digital camera business is extremely competitive;HP competes with Canon, Kodak, Nikon, Olympus,and Sony. Product life cycles commonly run less thana year, with prices rapidly declining over this period.

In the fall of 2000, the manufacturing manager forthe digital camera organization asked his team toimprove the performance of the digital-camera sup-ply chain. Led by supply-chain engineers, this effortwas supported by resources from finance, manufac-turing engineering, research and development, andlogistics. After considering various potential supply-chain alternatives, the team identified five scenariosto analyze in depth (Figure 1).

Scenario 1 was business as usual. In the existingsupply chain, digital cameras went from a dedicatedfactory in Asia to a single worldwide hub, alsoin Asia. This hub localized the cameras (packagedindividual items for regional consumption) and sent

WorldwideManufacturing

Facility

WorldwideLocalization

Hub

HP RegionalDistribution

Centers

Scenario 1 (as is)

Scenario 2

Scenario 3

Scenario 4

Scenario 5

Manufacturing/production

Localization

Distribution/order fulfillment

Figure 1: The five scenarios for digital camera supply-chain configura-tions considered various levels of consolidation and localization betweenworldwide manufacturing facilities, a worldwide localization hub, andseveral regional distribution centers.

Billington et al.: Accelerating the Profitability of Hewlett-Packard’s Supply ChainsInterfaces 34(1), pp. 59–72, © 2004 INFORMS 61

build-to-stock shipments to regional distribution cen-ters (DCs). The regional DCs shipped the finalproducts to retailers.

Scenario 2 eliminated the Asia hub facility, insteadopting to postpone localization until the camerasreached the regional DCs. Scenario 3 also eliminatedthe hub but moved localization back to the factory.Scenario 4 bypassed the regional DCs and delivereddirectly to the retailers. In this case, the Asia hublocalized the products and shipped them directly tothe retailers.

Scenario 5 represented the highest level of con-solidation, with a worldwide factory performing allmanufacturing and localization and then shipping theproduct directly to retailers in the regions.

Determining Supply-Chain CostTo decide which configuration to adopt, the teamneeded to calculate the total supply-chain cost foreach scenario. HP’s costing method divides the totalsupply-chain cost into variable, fixed, and inventory-driven costs. Variable costs consist of the value addedat each stage in the manufacturing process, includ-ing per-unit transit costs, direct manufacturing costs,and labor costs. Fixed costs are those that are not vol-ume sensitive and depend only on the supply-chainconfiguration. They include such items as the portionof a DC’s costs that are allocated or charged againstthe particular product line. Inventory-driven costs aredictated by the cost of the product, the holding-costrate, the service level, the supply variability, and thedemand uncertainty.

Aggregate fixed and variable costs are fairlystraightforward to calculate. For fixed costs, oneneeds to know the cost of the fixed assets. For vari-able costs, one has to know the volume going throughthe facility. The uncertainty surrounding inventory-driven costs creates the dual problem that these costsare both difficult to calculate and difficult to fore-cast. In particular, one year ahead in the budgetingprocess it is impossible to know for certain whichproducts will be in high demand and which will be inlow demand; which components will be delayed; andwhich suppliers will have quality problems. However,these uncertainties are some of the reasons for holdinginventory and major causes of inventory-driven cost.

The most difficult part of minimizing the inventory-driven costs in the supply chain is identifying thelocation and size of buffer inventory at each point inthe supply chain. Too much inventory of the wrongtype or at the wrong place in the chain can increaseobsolescence costs; on the other hand, too little of themodels in demand can create allocation conflicts andlost sales.

Within the worldwide supply chain for digital cam-eras, the team modeled 44 potential inventory loca-tions. Calculating a globally optimal solution for a

problem with so many variables requires a depth ofknowledge and modeling expertise beyond those ofbusiness users armed with spreadsheet-based tools.

The team turned to the PowerChain applica-tion specifically to solve this inventory-optimizationproblem and to quantify inventory-driven costs.The optimization engine coupled with an intuitiveinterface allowed us to introduce user-friendly, robustsupply-chain modeling to business users. For each ofthe five scenarios, the team developed a supply-chainmodel similar to the example shown in Figure 2.

FindingsFor each supply-chain configuration, the cost wasdescribed by a point along an efficient frontier. Sce-nario 4 (ship directly from the Asia hub) has the lowestoverall cost (Figure 3). The output of the model pro-vided the quantitative basis for the recommendation,enriched by consideration of other costs not affect-ing the inventory policy of the supply chain. Adopt-ing Scenario 4 provides two benefits. First, we cancreate a step-function improvement in supply-chainperformance, which translates directly into cost re-ductions, by restructuring the supply chain. Evenwith no other improvements in the business process,HP benefits from the efficient structure provided byScenario 4. Second, knowing the optimal inventorypolicy for HP’s global supply chain, we can implementit through changes to business processes.

We compared the optimized states of the five sce-narios. In practice, few current supply chains arealready optimized for inventory because the informa-tion required was not readily available until now. Thismeans that, in practice, HP’s savings from movingto a new supply chain are likely to be even greaterthan projected in this comparison because even keep-ing the as-is structure would result in some savings ifthe inventory were optimized.

Implementation ResultsBased on the results of the modeling exercise, animplementation team larger than the analysis teamtransformed the existing supply chain to the Sce-nario 4 configuration (Figure 4).

The implementation had three important results.First, HP realized the optimal inventory policy inpractice. Implementing the results and seeing thenumbers realized is critical to gaining the usercommunity’s confidence in the tool. Second, the movefrom Scenario 1 to Scenario 4 has reduced not onlyinventory-driven cost but also the aggregate amountof inventory in the supply chain, freeing up cash forrunning the business. Third, this reduction reducesthe downside risk in the supply chain. Given all thevariabilities in a supply chain, being wrong aboutsomething is a given. By lowering the aggregate


Figure 2: The PowerChain depiction of the Scenario 1 digital-camera supply chain consists of 44 stages repre-senting the raw materials, manufacturing, logistics, and customer demands corresponding to the specifics of thescenario’s supply-chain configuration. Each box on the supply-chain map represents a value-added step, suchas product transformation or transport. A triangle within a box denotes safety stock held at that stage. The teamoptimized inventory levels and locations across the multiechelon network. This supply-chain structure was oneof the first models HP built, and it is among the simplest.Note. The text associated with each stage is intentionally blurred to protect confidential information.

inventory level, HP minimized the cost of beingwrong and consequently the overall inventory risk.

The digital-camera supply chain reduced its inven-tory levels by over 30 percent and reduced totalsupply-chain costs by over five percent. Furthermore,this new supply chain maintained the existing highlevels of service while reducing new products’ timeto market by two to three weeks. We did not calculatethe revenue and cost benefit from reducing the timeto market, but getting product to market faster is asignificant benefit.

The project is 100 percent implemented. In fact,HP has produced three generations of digital camerasusing Scenario 4’s supply-chain configuration. Thatis, after its initial success in converting the existingdigital-camera product lines from Scenario 1 toScenario 4, HP has built all subsequent generationsusing Scenario 4’s structure and inventory targetsthat depend on the new product line’s demand char-acterization. The five-year net present value (NPV)of savings is well over $50 million calculated usinga one-time reduction in inventory plus annual sav-

Tot

alsu

pply-ch

ain

cost

Service level

1

2

3

4

5

Desired service level

Figure 3: The total supply-chain costs for the five digital camera scenariosare expressed as efficient frontiers.


Optimal Expected Inventory

Level for Scenario #1

Begin

Planning for

Scenario #4

Scenario #4 Goes Live

Time

Invento

ry (

units)


Level for Scenario #4


Level for Scenario 1

Begin

Planning for

Scenario 4

Scenario 4 Goes Live

Time

Invento

ry (

units)


Level for Scenario 4

Figure 4: The postimplementation inventory levels drop below the optimized levels that would have occurredhad the group stayed with Scenario 1, and the expected and actual inventory levels for Scenario 4 are closelymatched. The preimplementation actual levels for Scenario 1 differ from the expected levels because theexpected level for Scenario 1 was only a baseline and never implemented.

ings in variable and inventory-driven costs achievedthrough the adoption of Scenario 4. By the sum-mer of 2003, HP realized more than $35 million insavings.

Disseminating Proven TechnologyThe success of these optimization techniques withinthe Digital Camera business prompted a wider dis-semination across HP. SPaM sponsored and deployedthe processes and the accompanying technology toother business units through a central server sup-ported by HP IT resources. SPaM provides con-sulting support to help business units build theirmodels. By utilizing SPaM to disseminate these capa-bilities across HP, HP obtained three clear benefits.First, it established a shared horizontal process thatspans various business units, and that allows them toleverage each other’s work in supply-chain analysisusing a common language. This is important becausethe supply chain of any product crosses multipleorganizations inside and outside of HP. Second, thenew offering still leverages SPaM knowledge andexpertise, and the group is positioned to act as acentral broker for knowledge the business units cantap into, helping them to share best practices. Third,SPaM was able to shift resources to other domains ofexpertise outside of the supply chain.

HP’s supply-chain council sponsored and sup-ported these activities. The council consists of the sup-ply chain vice presidents across the company, andit sets and manages HP’s supply-chain strategy andpolicies.

Inkjet Supplies: Modeling a VerticallyIntegrated Supply ChainHP pioneered inkjet printing technology in 1984 andcontinues to lead the world market in this area. Today,HP’s Inkjet Supplies organization manufactures anddistributes ink cartridges, known internally as pens,and distributes them through many commercial andconsumer channels. These cartridges are used in HPprinters, plotters, copiers, and fax machines.

The Inkjet Supplies business includes approxi-mately 15 product families and over 250 manufac-turing stock keeping units (SKUs). Unlike digitalcameras, inkjet supplies have product life cycles in thetens of years, because they are manufactured to sup-port a large installed printer base (over 150 millionprinters sold around the world since 1987).

To protect confidential information, we have alteredthe supply-chain maps in this section. We have notchanged the qualitative and quantitative results fromthose results achieved in practice.

The Inkjet Supplies ChallengeThe Inkjet Supplies network is very complex. HP-owned factories ship inkjet cartridges in bulk toregional completion centers for localization and pack-aging. These centers, called pen completion centers(PCCs), ship finished product to channel partnersand to numerous HP distribution centers. The PCCsalso ship cartridges to HP-owned printer postpone-ment centers, which bundle printers and cartridgesfor sale. The supply chain supports two demand


FactoryRegional

PCCs

Regional

DCs

Regional

PPCs

Regional

DCs

Host

Demand

Trade

Demand

Circuits raw Bulk peninventory

Pen FGI

Pen FGI

Printer FGI

HP Back-end

Operations

HP Front-end Operations Customer

Cartridge sales

Printer salesPen FGI

material inventory

Figure 5: In the vertically integrated Inkjet Supplies business, HP’s back-end operations ship bulk pens (inkjetcartridges) to regional pen completion centers (PCCs). After the PCCs localize and package the pens, they shipfinished products in two streams. Replacement supplies (trade demand) for previously purchased printers goto regional distribution centers (DCs) and channel partners. Cartridges to be bundled with new printers (hostdemand) go to regional printer postponement centers (PPCs), which ship the bundled printer-cartridge productsto regional DCs.

streams: demand from customers who already haveprinters, served by HP’s channel partners and theHP distribution centers, and demand from printersales, because all printers come bundled with a starterkit that typically includes an initial set of cartridges(Figure 5).

HP holds safety stock at various points in the sup-ply chain to ensure high service levels. Bulk peninventory at the PCCs helps decouple customer-facingoperations from the long lead times of the facto-ries and to buffer the system from the demand vari-ability associated with the dozens of pen-packagingconfigurations HP offers. Bulk pen inventory is thelargest safety-stock buffer in the system; by poolingunpackaged, bulk inventory at the PCCs, HP post-pones packaging decisions so that it can respond toshifts in regional demand. After packaging at theregional PCCs, the pens are stored as finished goodsinventory (FGI) and ready to distribute for sale.

Planners for the front-end supply chain knew thatdemand for bulk pens varied far less than demand forindividual SKUs. While HP’s existing inventory pol-icy relied heavily on bulk pen inventory to respondto swings in demand for individual SKUs, the front-end organization wanted to know if HP would benefitby moving the bulk pen buffer, or some portion ofit, further upstream to the factories. Because eachfactory supplied multiple PCCs, they reasoned thatthe supply chain might benefit from pooling demanduncertainty at the factory stage, before shipment tothe regional PCCs.

Unfortunately, the inventory-modeling tools avail-able to the front-end planning community could not

perform this evaluation. Using high-level assump-tions concerning cycle time and variability, the toolscould establish buffer strategy between any two loca-tions (local optimization) but could not identify net-workwide opportunities (global optimization).

To consider the issue of bulk-inventory placement,the InkJet Supplies organization initiated the suppliesinventory modeling (SIM) project with a twofold mis-sion: (1) to investigate the benefits of holding bulksafety stock at the factories and (2) to identify otherareas of opportunity as time allowed.

Team CompositionPrevious modeling and analysis projects had failedto result in major change within the Inkjet Suppliesbusiness because the organization had relied onconsultants to perform and present their analyses. Atthe end of the engagement, the consultants wouldleave, taking with them the modeling and implemen-tation skills that were necessary to transition frommodeling to pilot stages and beyond. The businessstakeholders also could dismiss findings they disagreewith when the models had no internal champion todefend them.

To address this common problem, the Inkjet Sup-plies organization assigned collective responsibilityfor the SIM project to roughly a dozen supply-chainplanners and financial analysts from the factories andregions across three continents. These people wouldbuild a common set of analysis skills, develop thesupply-chain models, complete the analysis, and com-municate the conclusions themselves. The day-to-dayexperts who run the Inkjet Supplies business werethe people analyzing the problem, building their own


modeling proficiency, and answering detailed ques-tions posed by management.

PowerChain was instrumental in enabling thisteam’s work, providing it with an intuitive means ofaccessing some advanced mathematical programmingtechniques it could not have used with traditionalmodeling tools. Furthermore, once they were trained,the geographically dispersed team members couldcollaboratively develop and manage models over theWeb. Thus, their outputs were truly team deliverablesand represented a common understanding among allthe team members.

Marrying Process and TechnologyThe initial training session consisted of two days offormal instruction (Figure 6). Over half of this instruc-tion was hands-on learning to build a model, with theremainder of the time spent on interactive lectures.For the first time in their careers, business analystsglobally optimized supply chains they drew on com-puter screens.

On the third day, the team began building thesupply-chain map for inkjet cartridges. The ability todraw supply chains rapidly, combined with a favor-able learning curve, meant that after two days offormal training and a third day of project-focusedmodel construction, the team had built much of thesupply-chain structure. Two factors contributed tothis achievement. First, the optimization algorithmswere flexible enough to “solve what you can draw,”making rapid modeling possible. Second and equallyimportant, the SIM team used a divide-and-conquerapproach to global supply-chain modeling. Individu-als built those portions of the supply chain for whichthey were responsible day to day. This approach dif-fered from a traditional modeling effort in which one

Training (2 days)

Mapping (1 day)

Data Collection (6 weeks)

Validation of Model Structure (2 days)

Optimization (<1 day)

Constraining Stages (1 week)

Optimization (<1 day)

Validation of Model Outputs (2 days)

Analysis (1 week)

Communication (<1 day)

Implementation (2-6 months)

Weeks

Figure 6: In the timeline for the process of designing a supply chain, data collection and implementation consumemost of the time.

or two expert modelers construct the entire modelbased on interviews with individuals. The visualnature of the tool helped the team to quickly addresslogic issues and flow and data questions between por-tions of the supply chain.

Next, the team had to populate the models withdata. Of the overall two months of the project, three-quarters was devoted to gathering data. If that datahad been readily available, model development couldhave been completed in less than two weeks. Theteam spent much of its effort taking existing data fromHP systems and extracting enough detail to meet thegranularity requirements of the process steps mappedin the supply chain.

To validate the model, the team verified the flow oflogic and input data, such as stage costs, lead times,and demand. The team wanted to be sure that itcorrectly captured the process details, included therelevant cost drivers, and used the correct data. Modelvalidation took a day or two.

Once the team finished its initial validation of themodel, it immediately ran the model to optimizeinventory levels and locations. This laid the founda-tion for the next step of comparing the initial model’ssolution against the business constraints.

Determining and introducing constraints at the var-ious stages was the reality step in the process. Theinitial optimization run was only the starting pointfor the team’s analysis. The SIM team was experi-enced enough to know that a model’s optimal solu-tion need not be the best to implement in practice. Inreality, many near-optimal solutions that have slightlyhigher costs may either be far easier to implement orreflect operating constraints not factored into the orig-inal model. In this step, the team revisited the model’s


recommendations for the inventory policy and thesupply chain to make sure they matched real-worldconditions. For example, contract terms may or maynot require a supplier to carry inventory or a certainwork center may not actually be an ideal candidatefor postponement. During this phase, the team alsoadded existing supply-chain inventory policies to themodel to provide a baseline comparison. The teamdid not have to go outside the model to capture thesereal-world constraints. It could just enter the relevantconstraints for each stage and let the model figure outthe best solution given these new constraints. Modify-ing the supply-chain model to accurately reflect real-ity took roughly one week.

With the new capability for optimizing across anentire supply chain, the SIM project went far beyondlocalized inventory calculations. For the first time,analysts could evaluate the best holding locationsand the optimal quantities for those locations tomeet customers’ predefined requirements for ser-vice. Instead of performing a localized analysis thatwould likely suboptimize total system performance,the team could evaluate an entire system and, withinthat global context, determine the optimal inventorystrategy.

The team performed a second validation step intwo phases. First, it revisited the assumptions in themodel and compared the overall supply-chain costswith aggregate financial data from HP systems. Sec-ond, HP’s Finance group verified the current-statemodel against actual costs by comparing the model’soutputs with actual booked values. This crucial val-idation step showed that the model’s supply-chaincosts were very close to projected costs, not only as anaggregate number but at the per-unit SKU level. Themodeled and projected per-unit costs were within apenny of one another. The similarity in cost outputswas another major success factor in the project; theteam believed the one-penny differential was belowthe noise or measurement limit of their efforts.

The team then set out to interpret the results of itsglobal inventory modeling effort and to do this withina week. In the past, it would have treated each stagein isolation, assuming that each stage held a decou-pling safety stock. Because of the SIM project, theteam could model the entire end-to-end supply chain(Figure 7).

The results were very surprising. Contrary to theteam’s expectations, the optimization did not show abenefit from holding bulk pen inventory at the fac-tory. Instead, it suggested that the system would bebetter served if HP held safety stock in forms otherthan bulk pens altogether. Specifically, it suggestedincreasing the amount of circuits raw-material inven-tory in front of the factory and increasing the amount

of packaged finished goods inventory near the cus-tomers. The team might never have made their find-ing had it been forced to model only a small numberof supply-chain stages explicitly.

With the original bulk-inventory question an-swered, the team used its remaining time to evalu-ate other possibilities. In fact, the team had to restrainitself from exploring too many possible avenues,because it was so easy to duplicate, edit, and comparethe initial supply chain against what-if improvements.

Within a few days, the very people who wereresponsible for day-to-day operations of the supplychain made a discovery. To construct a global model,they had to detail the costs and times of indi-vidual freight lanes within the model. If the teamwere to move transoceanic freight lanes from air tosea, they calculated that HP could realize an NPVgreater than $80 million. But this would require anincrease in supply-chain inventory, a challenging sellin an environment focused on inventory reduction.The modeling team needed to demonstrate that theproduct cost savings far outweighed the additionalinventory-driven costs incurred by the longer transittimes.

Moving shipments from air to sea seemed com-pletely counterintuitive. Traditional rules of thumbhad focused on responsiveness through speed. HPsaw air shipment as a way to help minimize bothcycle times and inventory levels. Also, the planninggranularity of the current inventory-management sys-tem and a highly variable demand signal madeincreasing transportation time seem risky. With-out such countermeasures as holding inventory, HPrisked harming end-customer availability. The modelshowed that by sizing the inventory at the DCs appro-priately, HP could avoid expensive air-freight chargeswithout changing its service to customers.

The project team would have missed many of theseinsights if it had focused on a localized modelingapproach centered around bulk inventory of penswith outcomes constrained to only two choices (holdbulk at PCCs or hold bulk at factories). In this exer-cise, the team answered the bulk-inventory questionand, in addition, identified other opportunities basedupon insights gathered from using systemwide opti-mization and the team’s detailed knowledge of theday-to-day supply-chain operations.

The purpose of communicating results is toinfluence change. From a change-management per-spective, the analysis stage is the easy part. Even arecommendation that sounds simple, switching fromair to sea, can take tremendous effort to implement,particularly when the business is large. When thearguments in favor of the change are largely basedon arcane-sounding modeling exercises performed by


Factory PCC

Echelon Echelon

Figure 7: The Inkjet Supplies organization initially asked the team to determine whether it was optimal to shiftinventory of bulk pens from the PCCs back to the factories. With the end-to-end formulation of the supply chain,the team quickly determined this was not a cost-effective change. Instead, it was optimal to increase the amountof circuits raw-material inventory and the finished goods inventory near the customer. Next, the team identifiedsignificant savings from shipping by sea versus air between the factories and the PCCs.Note. The text associated with each stage is intentionally blurred to protect confidential information.

outsiders using black-box tools, selling the idea canbe a Herculean task.

A graphical modeling tool that the change agentsthemselves could use and understand helped themto accept and implement the changes. In this case,the team was trying to persuade the organizationto change processes that had been in place foryears and were ingrained conventional wisdom. Theirwork also challenged the notion that the InkjetSupplies business was too large for the applicationof management science. The graphical nature of thetool was a key communication aid during presenta-

tions and working sessions with various managementtiers. Managers could see the structures of their sup-ply chains clearly in a way that hadn’t been possi-ble before. With a picture to look at, stakeholders feltthey were in control, because they could understandwhat they were seeing. The black box of traditionalsupply-chain modeling had been replaced by an openwindow into the cause and effect of supply-chain per-formance.

In an environment where senior management hadrequested a reduction in overall inventory, the InkjetSupplies team instead successfully influenced stake-


holders to increase inventory in the supply chain toreduce total supply-chain costs!

For implementation to succeed, all stakeholdersmust understand the analysis, and they did becauseof the portability and standardization of the analy-sis techniques. Furthermore, because the people whowould do the implementation did the modeling, theydid not need to be convinced of the results. Thestakeholders could all agree on what the supply-chaincharacteristics were without resorting to higher math.For example, a question about postponement at a par-ticular point could be quickly answered by looking atthe supply-chain maps.

The team finalized and implemented the plan twomonths after the SIM project’s completion. HP tar-geted several products to move to sea shipment. Forothers, switching to sea transport did not make sensefor such reasons as product economics or a short shelflife. For those that can be shipped by sea (the pop-ulation of interest), the team took several considera-tions into account. Engineering had to confirm thatthe products’ quality would not be harmed by anyaspect of sea transportation. Manufacturing neededenough capacity to produce the extra units required tofill the new pipeline and still keep regional buffer lev-els at target. To address manufacturing’s concern, HPmodified factory schedules with a nine-month transi-tion plan.

Within six months of the implementation’s startdate, more than 90 percent of the units targeted forocean freight were indeed on the ocean, giving thebusiness an annuity stream in excess of $20 millionper year, which goes directly to the bottom line. Withour systems approach, we did not implement local-ized solutions that had unreported negative conse-quences for HP’s partners who might eventually pushback or be forced to end the business relationship.The model considered the overall impact on the end-to-end supply chain, including partners, thus reduc-ing the possibility of consequences that would laterundermine any reported gains.

For the Inkjet Supplies exercise, the team quicklydiscovered and implemented opportunities beyondthe one it was asked to evaluate. After deciding whereto hold inventory, it quickly discovered the air-versus-sea transportation opportunity and began to imple-ment it almost immediately.

This project highlighted the benefits of couplinga global inventory optimization algorithm with anintuitive interface that allowed the real-life businessusers to conduct modeling exercises rapidly. The in-house team identified opportunities in their field ofexpertise that an external consultant might never haveuncovered.

LessonsWe discussed two business cases to illustrate theapplication of a methodology in two very differentbusiness environments and to demonstrate the porta-bility of these techniques within diverse organiza-tions. However, HP now uses these techniques inareas beyond these two business units. It uses thismodeling method for such product lines as flatbedscanners, inkjet printers, servers, and laser printers.The user community includes analysts in HP regionsall over the world (Figure 8).

The techniques are currently disseminated on anas-requested basis. Through 2002, we had conductedmore than 10 training sessions around the globe forover 120 people. Clearly, companies need standard-ized, portable inventory-optimization techniques, andHP now has a portable tool that business users canuse in concert with HP’s internally developed processfor decision making in the supply chain.

This deployment to a broad distributed user basegoes beyond reaching the people we have actuallytrained. These very people influence a broader groupof people who are not interested in the deeper analy-sis but are interested in the broad implications of theanalysis. With a common language and framework,the efficiency of supply-chain analysis for inventoryoptimization has dramatically increased within HP.

Between January 2001 and January 2003, HP usedPowerChain to model over 1,500 supply chains. Inearly 2001, the pilot study in the Digital Camera busi-ness and the initial application in Inkjet Supplies wereeach creating multiple supply chains to represent var-ious supply-chain configurations. Because these earlyteams were so successful, the adoption process wasessentially viral. During 2001 the use grew on anannual basis of 200 unique chains modeled per year(this is the number of original maps created, notincluding revised versions of those maps). While eventhis growth rate was unprecedented within HP, thecatalyst for exponential growth, now 1,400 chainsper year, was the implementation of a formal train-ing class sponsored by HP Corporate Education. This

Figure 8: The PowerChain user community includes HP facilities in Boise,Idaho; Corvalis, Oregon; Fort Collins, Colorado; Houston, Texas; PaloAlto, California; Roseville, California; San Diego, California; Vancouver,Washington; Guadalajara, Mexico; San Juan, Puerto Rico; Grenoble,France; Boeblingen, Germany; and Singapore.


training class exposed new users to the business pro-cess and tools and equipped teams with the processknowledge to improve their supply chains.

Supply-chain engineers have the task of re-engineering business flows without having any realauthority to do so. The people they depend onfor implementation do not report directly to them.Achieving buy-in across the supply chain from a posi-tion of nonauthority is very difficult.

To make changes and obtain the cooperationof diverse groups and individuals, all the playersmust understand the details and trade-offs. With-out the ability to determine and validate the totalsupply-chain cost, one cannot possibly achieve amajority buy-in. While sophisticated business peoplerecognize that optimizing safety-stock levels canreduce total supply-chain costs, they also know thatdetermining an optimal inventory position is very dif-ficult. Sometimes, the business users spend quite a bitof time developing their own models. Others rely ongut feel. Either way, they commonly fail to persuadeothers to share their views.

Having an intuitive and proven analysis toolremoves this major barrier between analysis and im-plementation. The entire user community candetermine global inventory levels within a struc-tured framework. This translates into the followingbenefits:

—With a structured method for determiningsupply-chain costs, business users learn which param-eters are important and how the parameters interact.Best of all, the modeling team can demonstrate thebehavior of the supply chain to nonmodelers.

—Stakeholders from across the organization canuse the same framework in discussing the supplychain, which is critical for holistic analysis. Userscan see the trade-offs between local optimization andglobal optimization and persuade others within theirlocal divisions to buy into their recommendations. Byusing an algorithm that is flexible enough to solvewhat they can draw, users get the benefit of a picturethat is worth a thousand spreadsheets in communi-cating their findings to a broader audience.

—The ability for each user to add his or her partof the supply chain greatly aids the decision-makingprocess. It is difficult for one analyst from a particularbusiness operation to capture the entire network; it isequally difficult for analysts from two business oper-ations to combine their respective custom-built mod-els. By creating a modeling framework that allowsfor decentralized management and control, the teamremoved a barrier to creating, maintaining, and inter-preting an end-to-end model. This framework greatlyfacilitates the implementation process.

—One of the unforeseen benefits of using this toolis that each business user can input his or her own

0

500

1000

1500

2000

Cumulative number of unique chains modeled

Jan 2001 Jan 2002 Jan 2003

Figure 9: For 2001, supply chains were being modeled at a creation rateof 200 supply-chain maps per year, which was unprecedented adoption bybusiness users at HP. At the start of 2002, HP Corporate Education begansponsoring training on the business processes outlined in this paper, cre-ating an order-of-magnitude utilization increase to a new rate of 1,400supply chains per year.

portion of the data. In other situations, modelers tendto model supply chains using data that is readilyavailable to them. Many times, they miss or cannotget pertinent data that might be much more availableto users in the field.

ConclusionHP has benefited from a standardized process to opti-mize total supply-chain cost. HP uses a single pro-cess and optimization technology coupled together toobtain benefits in radically different business situa-tions and to identify different opportunities.

Large numbers of business analysts can nowmodel a supply chain together, make global deci-sions together, conduct sensitivity analyses together,and make unified recommendations to a cross-organizational management team based on theirefforts. Seeing these unified recommendations, man-agers throughout HP tend to quickly authorize imple-mentation. Stakeholders from across the supply chainunderstand the global impact of their inventory deci-sions, and HP’s implementation teams can respondto on-the-fly changes in the supply chain and modelthem rapidly.

For the first time within HP and, we believe, in theworld, we have brought inventory-optimization tech-niques to a broad audience. They no longer resideonly in the hands of highly trained OR individu-als. HP is staying true to its heritage of innova-tion. Not only have we introduced a better technicalanalysis method, but we have also introduced aprocess and a framework that helps people to com-municate, collaborate, model, and understand theirsupply chains. This new combination of technicalinnovation enabling broad-based organizational deci-sion making is noteworthy, and we believe that it isexactly what Bill Hewlett and David Packard wouldexpect from the employees of HP.


Appendix: PowerChain Intuition andAlgorithmThe authors of several excellent supply-chain papershave competed for the Edelman award in previousyears. In particular, Arntzen et al. (1995) and Cammet al. (1997) considered network design problems. Linet al. (2000) considered multiechelon inventory opti-mization. In our work, we continued this stream ofwork in focusing on developing a pragmatic solu-tion to optimizing inventory levels and locations in asupply chain. The success of PowerChain rests on itsability to marry supply-chain optimization algorithmsand an intuitive application environment to supportthe decisions planners make every day.

Problem ComplexityIn even the simplest multiechelon supply chain, theproblem of choosing inventory-stocking locations andthe target levels at those locations has exponentialcomplexity. Consider a supply chain with five poten-tial stocking locations in series, and suppose we areto decide only where to buffer safety stock. Assumebuffering at a location means the location fully satis-fies demand from inventory, whereas not buffering ata location means the location never satisfies demandfrom inventory. In a five-stage serial supply chain,there are 32 possible buffering policies. While com-plete enumeration is feasible for this stylized exam-ple, real-world supply chains have hundreds to thou-sands of stages; furthermore, once a stocking locationis known, the actual inventory level to maintain atthat location must still be determined.

Traditionally and with pre-existing tools, analystsleverage experience and intuition to reduce a pri-ori the possible sets of locations to make theiranalyses manageable. In our five-stage supply chain(Figure 10), a typical intuitive reduction might resultin three scenarios for consideration: stock everywhere,stock at the final stage only, and stock at the finalstage and at some upstream stage.

In the stock-everywhere scenario, each stage fullybuffers. This is a common scenario and is typically

Stage 1 Stage 2 Stage 3 Stage 4 Stage 5



Figure 10: For a five-stage serial supply chain, there are 32 possibledecoupling inventory policies. This figure shows three of them.

the output of single-stage solutions, because buffer-ing effectively decouples the echelons. The commondownside of buffering everywhere is that variabilityof lead times cannot be pooled across stages, and wehave found in practice that maintaining many stock-ing locations increases the risk of mismanaged (in par-ticular too much) inventory throughout the supplychain.

In the stock-at-final-stage-only scenario, the finalstage buffers the entire supply chain, while upstreamstages have no buffering. The problem is that, at thefinal stage, the product held in inventory is fully dif-ferentiated, so inventory is its most expensive at thispoint, making the cost of maintaining high service pro-hibitively expensive.

In the stock-at-final-stage-and-some-upstream-stagescenario, the final stage holds safety stock to satisfycustomer demand, but this inventory is not bufferingfor the entire supply chain, only part of it. One of theupstream stages also buffers, providing an intermedi-ate decoupling point. This scenario might be an appro-priate solution, and would certainly be considered byan analyst, if one of the stages has a very high process-ing cost. Buffering before this high-cost stage and notafter it might perform better than either of the othertwo scenarios because the expensive, processed inven-tory would be minimized.

The challenges of relying on experience and intu-ition are manifold. It is unlikely that the necessarilypartial analysis yields a true optimal solution. With-out the ability to justify a solution in financial terms,one would have difficulty obtaining support fromkey stakeholders, especially when they must makepersonal sacrifices for a global improvement. Morethorough analyses require substantially more time.Therefore, what we need is a method of determin-ing globally optimal stocking policies in a reason-able amount of time, while providing visibility tostakeholders.

Algorithmic FormulationIn this section, we outline the algorithmic fea-tures of the model. Graves and Willems (2000) andHumair and Willems (2003) provide more detaileddescriptions.

We model a supply chain as a network where nodesare stages in the supply chain and arcs denote that anupstream stage supplies a downstream stage. A stagerepresents a major processing function in the supplychain, such as procurement of a raw material, produc-tion of a component, manufacture of a subassembly,assembly and test of a finished good, or transporta-tion of a finished product from a central DC to aregional warehouse. Each stage is a potential loca-tion for holding a safety-stock inventory of the itemprocessed at that stage. Let the directed, acyclic, con-nected network G = �V �A� represent the stages and


arcs of the supply chain, where V is the set of stages,indexed by i, and A is the set of all arcs �i� j�. The setVd is the set of demand stages, such that stage k is inVd if there are no arcs �k� j� in A.

Each stage has a stage time, Ti, which is the timefrom when all of the inputs are available until pro-duction is completed and available to serve demand,including waiting time, processing time, and trans-portation time to put the item into inventory.

The decision variables are, for each stage, Si andSIi. Si is the service time quoted by stage i to eachof its downstream stages, while SIi is the maximumservice time being quoted to stage i by its upstreamstages. For each of the demand stages in Vd, there isa maximum service time bound that can be quoted tocustomers, represented by bi.

Cost is captured arbitrarily by a real functionci�Si� SIi� for each stage i, representing the holdingcost of inventory at each stage, given the outgoingand maximum incoming service times.

The objective function is therefore

minN∑

i=1

ci�Si� SIi��

The feasible region is constrained as follows. First,all service times are nonnegative and integral:

Sj� SIj ≥ 0 and integer for j = 1�2� � � � �N �

The maximum incoming service time to a stage jis greater than or equal to all of the outgoing servicetimes of that stage’s upstream stages:

SIj − Si ≥ 0 for all �i� j� ∈A�

The service times may be bounded arbitrarily bythe user at every stage (Lj and Uj are the lower andupper bounds the user specifies):

0≤ Lj ≤ Sj ≤Uj for j = 1�2� � � � �N �

The expression SIj +Tj −Sj , named the net replenish-ment lead time, is the time of exposure over which thebase stock must cover demand. It is the time requiredto replenish an item at a stage (the maximum incom-ing service time plus the stage’s processing time), net-ted by the service time being quoted by the stage. (Ifthe stage quotes a service time equal to its replenish-ment time, it needs no safety stock because all orderswill be satisfied after one full replenishment cycle;this corresponds to a net replenishment lead time ofzero.) The net replenishment lead time can be arbi-trarily bounded by the user at every stage to reflectsituations in which stages are either not allowed tohold inventory or, by contractual terms, are requiredto hold a certain amount of inventory:

NLj ≤ SIj + Tj − Sj ≤NUj for j = 1�2� � � � �N �

The cost functions ci�Si� SIi� are obtained from theinventory dynamics. Graves and Willems (2000) showthe cost function to be

ci�Si� SIi�= hi

{Di�SIi + Ti − Si�− �SIi + Ti − Si��i

}�

where hj is the holding cost of inventory at eachstage, Dj�x� is the maximum possible demand seenby the stage over the time interval of length x,and �j is the average demand in one time period.Extensions to the formulation to account for stage-time variability and production-capacity constraintscan be captured by more complex treatments ofthe ci�Si� SIi� functions. In even the simplest forms,these are general nonlinear functions, so that thisproblem is nonlinear with integral decision vari-ables. We have developed dynamic programmingtechniques, described by Graves and Willems (2000)and Humair and Willems (2003) for solving thisproblem.

ArchitecturePowerChain is a collaborative application for theinteractive visualization, design, and optimization ofsupply chains. Using Web-based technologies in amultiuser environment, the software encompasses theentire modeling life cycle, including model creation,supply-chain structure mapping, data loading, opti-mization sensitivity analysis, model versioning, andworkflow. A typical project will analyze one or moresupply-chain scenarios (such as the five digital cam-era options we presented). Within the software, ascenario is modeled by a construct called the chainobject. A chain object has its own version control, sim-ilar to what is found in document management orsource-code control systems: a chain can be publishedto freeze its state, checked out as a new version tomake changes, and shared among different users whoare stakeholders of the analysis.

A chain object comprises global settings, stages,and links. Global settings include such parameters asthe name of the chain, the modeling horizon, andwhether advanced features, such as batching andcapacity, are enabled. Most of the data requirementsfor PowerChain relate to stages, including demand,target service levels, holding cost rates, yields, capac-ities, cost structures, and time delays. Links relatestages to each other and represent the flow of mate-rials or information but by themselves contain verylittle data, mainly a multiplier that indicates a goes-into factor from the source stage to the destinationstage.

The flexibility of the individual stage coupled withthe optimization and calculation techniques embed-ded within the software are what enable PowerChainto solve what the user draws. However, becauseof the Web-based, multiuser, distributed nature of


ERP

SCM

MRP

Excel/

Access

ClientComputer

Data Layer

PowerChain Web Server

XML Dispatcher

Command Hierarchy

Business

Objects

Inventory

Optimization

Engine

Architect

Optimization

Engine

Reporting

Engine

PowerChain Database

Job ManagerSecurity Layer

Figure 11: A user connects to the PowerChain application using a Webbrowser and draws a supply chain using a Java applet. The Web servergenerates requests for performing supply-chain actions using an XMLschema specifically designed for supply-chain modeling. These XMLrequests are dispatched and turned into commands, such as “Add Stage”or “Optimize.” An asynchronous job manager handles complicated andtime-consuming actions, such as optimization, while business objects,reporting, and security layers handle traditional application-tier activities.All of these operations rely on a relational database.

the application, the path from drawing a supplychain to optimizing inventory policies is hardly direct(Figure 11).

Web technologies enable users to come up to speedquickly while minimizing the total cost of ownership.New users can be added to the system and broughtonline with minimal IT investment as soon as theyare trained on the software.

ReferencesArntzen, B. C., G. G. Brown, T. P. Harrison, T. L. Linda. 1995. Global

supply-chain management at Digital Equipment Corporation.Interfaces 25(1) 69–93.

Camm, J. D., T. E. Chorman, F. A. Dill, J. R. Evans, D. J. Sweeney,G. W. Wegryn. 1997. Blending OR/MS, judgment, and GIS:Restructuring P&G’s supply chain. Interfaces 27(1) 69–93.

Graves, S. C., S. P. Willems. 2000. Optimizing strategic safety stockplacement in supply chains. Manufacturing Service Oper. Man-agement 2(1) 68–83.

Graves, S. C., S. P. Willems. 2002. Strategic inventory placement insupply chains: Nonstationary demand. Working paper, BostonUniversity, Boston, MA.

Humair, S., S. P. Willems. 2003. Optimizing strategic safety stockplacement in supply chains with clusters of commonality.Working paper, Boston University, Boston, MA.

Lin, G., M. Ettl, S. Buckley, S. Bagchi, D. Yao, B. Naccarato, R. Allan,K. Kim, L. Koenig. 2000. Extended enterprise supply-chainmanagement at IBM personal systems group and other divi-sions. Interfaces 30(1) 7–25.

Jeff Clarke, Executive Vice President, Global Oper-ations, Hewlett-Packard, writes: “HP’s supply chainscross multiple organizations and companies. Theimplementation of these supply-chain optimizationtechniques and tools, leveraged across multipleorganizations, delivers on this need for the supply-chain development community. This work high-lights the power of standardization and knowledgeleverage, and shows HP’s continual leadership insupply-chain innovation.”

Vyomesh Joshi, Executive Vice President, Imag-ing and Printing Group, Hewlett-Packard, writes:“IPG’s goals in operational excellence require us tooptimize costs, accelerate time-to-market, and maxi-mize flexibility while delivering industry leading cus-tomer experience. This initiative successfully definedand implemented tools that allow IPG businessesto model and analyze the end-to-end supply chaintackling these goals. Through global collaboration, theDigital Imaging and Inkjet Supplies businesses wereable to implement supply-chain improvements yield-ing over $130 million in benefits to IPG.”

Mary Peery, Senior Vice President, IPG Digi-tal Imaging and Publishing Organization, Hewlett-Packard, writes: “Digital Imaging is a significantgrowth business for HP, with a goal of providingimage rich conversation seamlessly between people.Digital cameras are a core element of this strategyto bring affordable solutions to market. By reduc-ing the camera total supply-chain costs by morethan 5% and improving our time to market by morethan two weeks on every camera, our customershave access to the latest technologies at an afford-able price. This is not only a win for Digital Imag-ing, but our customers as well. These supply-chainanalysis techniques have now been adopted by ourflatbed scanner and photosmart inkjet printer prod-uct lines following the success in our digital camerabusiness.”

Boyd Lyon, Enterprise Supply-Chain Manager,Inkjet Supplies Business, Hewlett-Packard, writes:“Within the Inkjet Supplies Business, it is well rec-ognized that our success continues to depend on ourability to compete as an integrated network. We arechallenged with continued pressures to reduce inven-tories, supply-chain complexity, and product costswhile simultaneously breaking new ground in creat-ing new, disruptive customer advantage. These toolsand methodologies are enhancing our understandingof the broader systems we manage, and providingnew tools for shaping action across a distributed oper-ations network. We are able to ask the right ques-tions, focus on the right opportunities, and speed theprocess of recognizing benefits.”

Documents

Accelerating the Proﬁtability of Hewlett-Packard’s Supply ... · Billington et al.: Accelerating the Proﬁtability of Hewlett-Packard’s Supply Chains 60 Interfaces 34(1), pp