Hewlett-Packard also faces these challenges. Over the years, our team of operations research professionals has developed an approach that we believe is widely applicable. In this article, we describe our framework for managing one aspect of complexity - namely, product line complexity - and share some examples of how we have applied the approach at HP.

Product Line Complexity


The term "complexity" means different things to different people. To a procurement professional, complexity may manifest as a large and growing number of suppliers and supplier relationships. In the factory, complexity may refer to the number of both components and final products in inventory. For a product planner, complexity may mean the scheduling and allocation of finished goods across multiple factories and sales channels. A change at one point - a supplier, a component, a product - can have cascading effects across organizations and even customers.

This article focuses on complexity in product line offerings - in terms of the number of products and the degree of part commonality among the products - and its impact across the supply chain. We share our approaches for measuring the costs and benefits of product line complexity, and for making tradeoffs between the two to right-size one's product line.

Complexity Management at Hewlett-Packard


Last year, HP generated $80 billion in revenue and $3.5 billion in profit, and offered more than 90 different product lines for sale in 160 countries. To a company of this size, the impact of successfully managing product line complexity, or the cost of its mismanagement, can easily reach into the hundreds of millions of dollars.

As an example, consider HP's product line of consumer desktop PCs. In 1998, HP and Compaq combined offered a total of 88 unique desktop PC systems to North American consumers. In 2002, after the companies merged, this total had reached 110 systems. By 2004, the number had grown to 170 unique systems, with a complete set of new models introduced every three months. The proliferating number of products also triggered corresponding increases in unique and custom parts. While a broader product line allows HP to offer a larger selection - ranging from no-frills low-cost PCs to "gaming" PCs offering enhanced video and audio - the managerial, marketing and supply-chain costs of adding this variety can amount to tens of millions of dollars per year.

A Framework for Managing Product Line Complexity


Our baseline rule for managing complexity is to add complexity only when the benefits outweigh the costs. This may sound simple, but identifying and isolating these costs is itself a highly complex task. For example, adding new models to a product line may result in incremental margin from increased sales, but estimates of these cash flows must be adjusted to account for cannibalization of existing products. In HP's business model, the hidden costs of adding products may be spread across several areas on the income statement. Opportunity costs, or indirect (secondary) benefits, such as increased retail shelf space, are even more difficult to quantify.

We have developed a five-step process that measures the costs and benefits of complexity, and then uses these measurements to guide product line planning. A schematic overview of this process is illustrated in Figure 1.

Figure 1: Five steps to managing product line complexity.

We discuss each of these steps in turn, and review how we have applied the approach at HP.

Step 1: Identify cost areas impacted by product line complexity. The first step is to identify which cost areas are impacted by product line complexity. To avoid overlooking hidden costs, conduct a thorough review of material, information and financial flows along the value chain.

At HP's consumer PC division, we conducted interviews with operations, finance and marketing staff to gain a broad perspective on how complexity impacted their organizations. We subsequently identified the five cost categories shown in Figure 2.

Figure 2: Complexity-driven costs in HP's consumer PC division.

The breadth of costs thus uncovered had far-reaching impacts on the organization as a whole. Increasing the number of desktop PC offerings greatly influences HP's PC-assembly processes and increases the likelihood of error. Figure 3 shows photos from HP's assembly, packaging and testing operations, all of which are impacted by greater product line complexity.

Figure 3: Assembly, testing and packaging of desktop PCs at Hewlett-Packard. More product line complexity means more work, and more chance for error, at each step.

Step 2: Estimate complexity costs per unit in each area. The next step is to estimate the effects of complexity in each cost area on a per-unit basis. This aids in quantifying the cost impacts of changes in product line size or configuration. The estimation approaches used will vary by the type of business and by cost type. Generally, this requires a combination of theoretical principles (such as statistical techniques for calculating inventory-pooling benefits) and empirical measurement.

To perform this estimation at HP, we developed an Excel-based model showing the impacts of product line changes on each of the cost categories we had identified. To build the model, we gathered detailed information on components, SKUs (stock keeping units) and retailers, and combined this with our understanding of business operational policies (i.e. planning, forecasting, batch size, shipment frequency, etc). Figure 4 shows a simplified list of analysis inputs. The expected VCM (Variable Contribution Margin or "profit") for each SKU is the most important, as that number is closely linked to key business performance metrics and goals.

ANALYSIS INPUT
For each component		. Order lead time . Cost . SKU allocation . Inventory holding policy (weeks of stock) . Salvage value
For each SKU		. Manufacturing factory allocation . Expected net revenue ($/unit) . Expected variable contribution margin (VCM $/unit) . Forecasted lifetime volume . Eligibility for price protection
For each retailer		. SKU allocation . Return rates (historical) . Marketing fund liability (historical)
Other inputs		. Lifetime volume forecast variability (historical) . Retailer order lead time . Supplier holding cost (% per year) . Factory capacities . Depreciation rates

Figure 4: Analysis input for measuring complexity-driven costs.

In addition to the component-level and product-level costs, complexity can also affect costs on a product line or product portfolio level. These costs are not always apparent to the person making localized decisions about configuration, price or sales forecast. To address this issue, our model includes portfolio-level effects when estimating the total cost of complexity.

Step 3: Define "cutoff" margin threshold per SKU. Measuring the per-unit costs of complexity has some challenges. Most troublesome is the fact that these costs are generally nonlinear, and vary depending on the characteristics of the portfolio being offered. In order to enable rapid decision-making, we found it necessary to devise a simpler set of "complexity guidelines" that could be used to evaluate individual products without having to model an entire portfolio.

To achieve this at HP, we used the complexity model we developed in Step 2 to simulate many different scenarios, and then derived a simplified set of complexity guidelines based on the simulation trial results. The guidelines consisted of threshold margin contribution requirements for evaluating individual SKUs under various circumstances. In the PC division, product offering plans are amended and altered over many weeks in response to retailer requests and new market information, so that the final product line is frozen only at the last minute, leaving no time to train and run a detailed model. The complexity cutoffs we developed, while not absolute, offered good estimates of hidden costs. The cutoffs are helpful for guiding day-to-day product planning decisions. The structure of these guidelines is shown in Figure 5.

COMPLEXITY COST THRESHOLDS

HP Consumer Desktop PC Products in North America (specific values are confidential) Each SKU in the portfolio must meet business objectives AND contribute acceptable contribution margin (VCM) to offset these complexity: Organizational and manufacturing impacts: Portfolio Size Fixed Cost Adder Per-Unit Variable Cost Adder 35-40 SKUs $A - 40-45 SKUs $A $B 45-50 SKUs $A $C > 50 SKUs $A $D Inventory / shortage costs: If the SKU contains unique parts: add E% of the unique part cost per unit Price protection (amount HP pays retailers when unsold retail inventory is affected by an HP price drop), and return costs: Increase in returns costs from SKU addition: add F% of total retailer revenue If the SKU is a price protected derivative: add G% of total retailer revenue

Figure 5: Complexity cost thresholds for HP's North American consumer desktop line.

Step 4: Evaluate incremental margin of each proposed SKU. Increased product line complexity can yield benefits as well as costs. Benefits include greater consumer choice, increased shelf space at retailers, increased consumer mindshare and reduced pricing transparency. Complexity also benefits retailers who can offer custom-built SKUs to differentiate their product offerings. Thus, offering a broader product line not only brings in additional revenue but also strengthens HP's relationships with retail channel partners.

While all of these benefits of product line complexity are genuine, it is not necessary to quantify all of these benefits in cash-value terms. We found that the incremental margin contribution generated by potential additions to the product line, after adjustment for cannibalization and attached sales, captured the essence of the benefits for our needs. The other benefits qualify as "strategic considerations" that could be applied prior to a final decision about inclusion or exclusion of a SKU. At HP, sales forecasts were the primary input for calculating incremental margin contribution projections.

To understand why we could exclude some of the secondary benefits of complexity from our analysis, it helps to look at an example. In general, we have found that the targets of complexity reduction are naturally those products whose incremental margin contributions are the smallest. In the theorized example shown in Figure 6, the "products" shown inside the circle constitute merely 6 percent of the total projected margin contribution, yet represent 38 percent of the total product count. In other words, one must increase the product count by 63 percent in order to achieve the benefit of a 6 percent increase in projected margin. The strategic value of "marginal" products such as these is generally far less than the cost of retaining them in the portfolio.

Figure 6: Illustration of typical complexity reduction targets.

We conducted a similar analysis at HP, looking at the incremental projected margin contribution for each SKU in HP's proposed product line. The picture did not look much different from the theorized example shown above.

Step 5: Eliminate proposed SKUs that do not exceed threshold. The last step involves eliminating products under consideration for which the projected margin contributions do not exceed the thresholds assigned. At its most basic level, this is simply a SKU-by-SKU comparison of projected margin impacts against the measured cost thresholds.

Figure 7 shows data from HP's consumer desktop product line illustrating this approach. We adjusted the projected margins of each SKU by adding the complexity costs to obtain complexity-adjusted margin projections. Within a limited range of product portfolios, these adjusted projections could then be compared easily against one another by sorting them and then plotting as shown. All SKUs with adjusted VCM projections below zero were deemed "red zone SKUs," which became candidates for elimination from the product line.

Figure 7: Comparison of predicted VCM (variable contribution margin) percentages for SKUs in a proposed consumer desktop product line at HP, adjusted for complexity costs. SKUs with VCM percent below zero were candidates for elimination from the product line.

While the approach shown above appears to identify with certainty which SKUs should be eliminated, analyses of this type inevitably carry some degree of uncertainty. To understand the robustness of our results, and to communicate this to our clients, we performed sensitivity analysis to identify areas of greater or lesser confidence.

Consider cannibalization as an example. Adding a new product might cause some customers, who would have bought another similar model from the same manufacturer, to choose the new product instead. Similarly, eliminating a product may not result in the loss of 100 percent of the forecasted revenue for that product, as customers may purchase another product in the same manufacturer's lineup if the eliminated SKU is not available. To test the impact of cannibalization on our margin forecasts, we modeled the extreme cases of 100 percent cannibalized and 0 percent cannibalized.

We went even farther to adjust for "attached" product sales (such as when a consumer purchases a monitor with their PC), and for some of the other intangible complexity benefits. We extended our range to cover everything from "100 percent cannibalized" to "-50 percent cannibalized." The "-50 percent" covers the cases where adding the product not only generated 100 percent new demand, but also generated other benefits beyond its own margin. Even in the most generous case, there were still many cases where the cost of introducing a new product still outweighed the incremental margin.

We applied a similar approach to our cost estimates as well, assessing the impacts on each cost category across of range of logical values. By doing so, we were able to construct an overall confidence interval for the estimated impact of a proposed set of changes in HP's desktop PC product line. Figure 8 shows this analysis for the proposed elimination of "red zone" SKUs in Figure 7.

Figure 8: Model output showing the projected value of complexity reduction, along with confidence intervals for each assessment.
(Click here to view a larger version.)

The left-most bar shows total expected contribution margin from offering all proposed products (the "full complexity" offering). Removing products from the portfolio causes a drop in total margin. The magnitude of the decrease depends on the level of cannibalization. This effect is minimized when all demand for the canceled products shifts over to non-canceled alternatives. However, if no cannibalization exists, then all of the margin from the canceled products - and possibly also margin from other connected products such as monitors and printers - is lost. The model conservatively assumes that each SKU provides incremental volume and that demand for cancelled configurations is not transferred to other products. The third bar shows the new VCM total for the simplified portfolio, which is the delta between the first two bars.

If we knew with certainty that all costs were already fully reflected in our existing pricing models, the first three bars would tell the whole story. Unfortunately, this is not the case, as many costs are spread across multiple business functions and are difficult to track. The model shows that eliminating SKUs offers multiple financial benefits. Light gray bars show the expected cost savings in each of the five cost categories: manufacturing ramp, component inventories, marketing liability, organizational performance and returns/warranty. When these savings are added into the analysis, they outweigh the lost margin and suggest that cutting SKUs will improve overall profitability.

Results in HP's Desktop PC Division


HP's desktop PC division was able to eliminate several unattractive SKUs from its product portfolio and change its processes for evaluating proposed products in the future. Armed with this information, the team had clearer guidelines on when to create new SKUs, and is now enforcing stricter minimum quantities with retail accounts. We estimated that the financial impact from this application to North America desktop PCs was on the order of $4 million per year.

Other Applications at HP


Our teams have applied a similar approach across many businesses. Examples include commercial as well as consumer desktop PCs, notebook computers, monitors, servers, storage product lines and spare parts.

The technique has proven helpful even to businesses that are quite different from our desktop PC example, such as HP's spare and trade parts business. HP Global Supply Operations is responsible for selling hundreds of millions of dollars worth of spare parts every year. Consider that LaserJet printer products are frequently sold with service contracts covering repair and replacement of parts for three years after the date of purchase. HP builds and sells these parts in some cases for up to seven years after a product line is discontinued. In total, HP offers more than 14,000 different spare printer parts for sale. A variety of this magnitude creates significant challenges for planning and inventory management, as support contracts from HP's suppliers often require end-of-life part buys prior to the expiration of HP's support period. These parts are managed at warehouse facilities such as the one shown in Figure 9.

Figure 9: HP's spare parts warehouse in Roseville, Calif. This is one of eight HP global spare parts central warehousing facilities, each of which manages from several hundred.

Our team worked with the spare-parts business to rebalance the parts product line. This involved extending support lives on some of the most commonly failing LaserJet parts while discontinuing parts with zero or virtually zero demand that were offered years beyond the expiration of HP's service contract obligations. In all, the Global Supply Operations team removed more than 1,000 SKUs from HP's LaserJet spare-parts offering, eliminating the need for manufacturing capabilities, inventory and management attention to these parts. At the same time, HP extended the support life on a handful of high-failure parts, generating $500,000 in incremental annual parts sales and filling a previously unmet customer need.

We sorted all products by the incremental margin they were projected to bring in, as illustrated in Figure 10. Forty-seven percent of the parts more than six years old contributed zero revenue, and could be eliminated immediately. Beyond these, there was a tradeoff between projected margin contributions and the cost of supporting each additional part. Again, we focused on the cost side to identify a threshold of incremental margin contribution below which the benefits of adding a support part would not outweigh the costs.

Figure 10: Analysis of HP's LaserJet spare parts product portfolio. HP was able to eliminate over 1,000 parts with low projected demand, thereby reducing scrap and other complexity-driven costs. Further, HP extended support life on a small number of parts, resulting in $500,000 in incremental annual revenue from the portfolio.to tens of thousands of SKUs to support repair services and part sales for HP's products.

Conclusions


Managing complexity is of critical and growing importance to today's businesses. We have investigated one aspect of complexity - product line complexity - and have developed a five-step process for managing product line tradeoffs quantitatively. At HP alone, we project that these approaches can yield hundreds of millions of dollars in impact through cost avoidance, performance improvement and margin generated from servicing unmet customer needs.

Our experiences have shown us that reducing product line complexity without appropriate analysis can be detrimental to business results. Complexity has an intrinsic value, in many cases improving profitability. The difficulty in managing complexity is in separating "good complexity" from "bad complexity" in a systematic way. Customers are happy to pay for "good" complexity. "Bad" complexity needlessly increases costs and jeopardizes profitability. We encourage business leaders to follow steps 1-3 of our approach before a complexity crisis occurs so that as new products are proposed, the organization understands the costs that new SKUs will bring and can control their growth appropriately.


Brian Cargille manages Hewlett-Packard's Product Design for Supply Chain program. Chris Fry is owner and president of Strategic Management Solutions Group, a management consultancy and HP business partner. Aaron Raphel is a former HP Design for Supply Chain intern from the MIT Leaders for Manufacturing Program (http://lfm.mit.edu/). Please direct questions or comments to brian.cargille@hp.com.

The authors thank HP operations research scientist Dr. Thomas Olavson for his co-development of these techniques for spare parts applications, and HP managers Jorge Arreygue, Paul Coggeshall, John Fisher, Rob McDowell and Sam Szteinbaum for their tireless sponsorship and support.